Intel's processor is drawing 307W to slightly edge past the 5950X in Cinebench, which only draws around 214W.
It's hard to cool 307W over extended periods without your computer becoming really loud, even with a water cooling setup.
Now, if they had that kind of cooling available, they could have easily gotten a lot more performance out of the Ryzen, while the i9 is already operating at the limits of what consumer grade cooling can sustain.
There's no way that Intel processor is beating a 5950X that is also drawing 300W. At that point some Ryzen's will be operating at close to 6GHz per core.
The most interesting takeaway is that Intel apparently has no qualms selling a processor with a stock configuration that draws 300W.
Also their Ryzen numbers are beyond suspicious. On a silent configuration (peak power draw ~209W, 175W sustained, some undervolting) my Ryzen 5900X hits a Cinebench R20 score of 8835 vs their 8390. I suspect my rig is worse since I cheaped out on the RAM, but I wouldn't know because they omit information about the Ryzen test rig.
Their Ryzen numbers don't add up with other benchmarks around the internet either: The 5950X should hit around 10400, and the 5900X should hit 8600 in their stock configurations with mid-to-high-grade RAM[1]. Their scores are 400 points and 200 points lower respectively.
To get numbers as bad as theirs, I have to mismatch my FCLK and RAM intentionally. But again, we don't know what their setup is because they didn't tell us.
We however do know that they used DDR5 RAM for the Intel processor, which the current gen Ryzen don't support. Likely some performance was gained there as well, but obviously they didn't test the Intel processor with DDR4 RAM to get an idea of how much comes down to faster RAM. It's still fair to use DDR5, since the Intel processor supports it, but it's hard to say much about the Intel processor performance (development) itself without some additional information. Without this information we don't know whether we can expect the next generation of Ryzen processors with DDR5 support to immediately leave Intel in the dust again, or whether DDR4/DDR5 doesn't really matter.
As others have sort of pointed out, your opening sentence is incorrect. That "307W" draw is "Full System Power Consumption", not just the CPU. The 12900K will draw around 230-240W at max load (when doing work like Blender renders), compared to 140W for a 5900X.
But for Gaming loads where the CPU maxes out, I haven't seen reviewers hit those numbers. The 12900K was using ~125W in those workloads, just a few watts lower than AMD.
Now, if you go down to Intel's midrange (i5, not the i9 you were quoting), the power usage is way down (at 125W), vs the AMD 5600X at 75W with a blender workload. Gaming workload, Intel is a few watts higher than AMD.
Yes, Intel is drawing way more power for super CPU intensive tasks, but for what most people will be doing, it's pretty even.
>Yes, Intel is drawing way more power for super CPU intensive tasks, but for what most people will be doing, it's pretty even.
This. I honestly don't get why people are so focused suddenly on the power consumption for a desktop CPU designed for enthusiasts who want top performance at any cost.
GPUs are consuming way more power than that(Nvidia's Ampere is a notorious power hog), pushing consumers to 700W+ PSUs and nobody bats an eye, but seem to throw their hands in the air when their CPUs get close to that of a GPU.
Like why? It's the top spec model, it's supposed to be ludicrous, not be good value for money or highly efficient, they have other models more competitive for that. It feels like people are just looking for reasons to to create outrage about a product they're not gonna buy anyway.
> I honestly don't get why people are so focused suddenly on the power consumption for a desktop CPU designed for enthusiasts who want top performance at any cost.
Because if you up the Power/Current limits on a Ryzen, they will run faster - especially in tests like Cinebench. You won't even have to 'overclock' them. Simply allow them to draw more power if your cooling, motherboard, and PSU can handle it.
To make that a fair test the tester would have had to crank those settings up as far as they go, or until they match the Intel processor. If the Intel processor is allowed to draw 250W, the Ryzen should get the same treatment.
For example a Ryzen 5900X at ~135W (CPU Package) should score around 8200 in Cinebench r20, and at ~200W it should score around 8900.
Since in the test those 200W are apparently for the whole system, it's pretty obvious the tester left those at the stock values.
> To make that a fair test the tester would have had to crank those settings up as far as they go, or until they match the Intel processor.
The test was perfectly fair because they used the CPU as delivered, as the average user would use it.
Nobody is buying these CPUs to overclock them to a specific power usage. They're either running them stock or running them at whatever comfortable overclock they can get away with.
Test stock versus stock and overclocked versus overclocked. Trying to normalize to specific power consumption doesn't make sense for enthusiast builds.
Overclocking is running a CPU at higher clock speeds than advertised.
If you just change PBO limits, it never goes beyond what AMD knows their silicon can handle, and what is sold to the customer. Individual cores might already be boosting to those clock speeds over short intervals. You're just allowing more cores to do that and longer, within your configured power limits. The expectation here is thermal throttling, not an unstable system.
No one pretends PBO is not overclocking. Not even AMD:
> Precision Boost 2 is a performance-maximizing technology available in all AMD Ryzen™ and Ryzen™ Threadripper™ 2000 Series processors (or newer). This technology can improve your PC’s performance by raising clockspeeds
> Because Precision Boost Overdrive enables operation of the processor outside of specifications and in excess of factory settings, use of the feature invalidates the AMD product warranty and may also void warranties offered by the system manufacturer or retailer.
-
Reply because I'm getting rate limited, probably over that one time I wasn't nice to dang
> Changing these values will not make your CPU reach voltages it wouldn't otherwise have reached, nor will it make your CPU reach clock speeds it wouldn't otherwise hit.
This is patently false! I don't know where you got this idea when I literally just quoted AMD saying it raises clocks past factory specifications!
Also sure where you think the additional power draw comes from...
AMD literally uses the tagline: "Precision Boost Overdrive: A Smarter Way to Overclock"
-
I see no point in continuing a conversation with someone who doesn't understand the basic tenets of the technology they're trying to split hairs over.
> What is important to note is that PBO only affects these three power limits. The effect on CPU clock speed is indirect, and PBO will never boost the CPU past the advertised clocks.
I think it's fair to say adjusting power limits falls under a deliberately broad "overclocking" umbrella, but it's not really overclocking, and doesn't require you to do the effort part of real overclocking.
> This is patently false! I don't know where you got this idea when I literally just quoted AMD saying it raises clocks past factory specifications!
The part you quoted doesn't have the word 'clocks'.
> If you don't know clocks is slang for clockspeed, you're out of your depth. So just stop talking.
I phrased that badly. I was talking about the second quote there.
It's a little bit subtle.
The first quote says it will raise clock speeds.
The second quote says it will operate outside of specifications.
But nowhere does it say the clock speeds will be outside the specification.
It's raising the power limits outside the specification so that it can raise clock speeds inside the specification.
> And you're saying that technical documents from AMD don't know what they're talking about (it's a technical document. It's a technical article.)
If you can quote me something from AMD that specifically says PBO (not AutoOC) raises the maximum clock speed above the number you can boost to within spec, I will admit I was wrong and apologize. But the page you linked doesn't say that.
PBO allows overclocking[1]. But it mainly allows you to configure other stuff, liking raising/lowering certain limits, that are not overclocking. Changing these values will not make your CPU reach voltages it wouldn't otherwise have reached, nor will it make your CPU reach clock speeds it wouldn't otherwise hit. It will however make your CPU draw more amps and use more watts to sustain clocks for longer than normal and on more cores at once.
Allowing here means that in the typical motherboard you'll have to change PBO from the default "Auto" to "Manual" to see these settings.
Where do terms like "undervolting", "overvolting", "setting TDP" even fit in your vocabulary if you insist everything is to be called "overclocking"? These terms should be literally self-explanatory.
[1] But only technically, because for some reason the AutoOC feature is found in the PBO menu on motherboards, even though AutoOC is not part of PBO. They are however typically marketed together.
Intel have a rich and storied history of paying for biased and dishonest reviews. You should be thankful that people are willing to fill in the gaps so you hear the full story.
Because there isn't a proportional performance gain, and the heat generated will degrade the computer, at what cost? A few irrelevant percentage points?
If you're concerned about proportional performance gains, this product isn't for you. So why are you complaining about something you're not gonna buy anyway?
Most tech products that are the tip of the spear are bad value for money and less energy efficient than the ones down the range (Nvidia 3090 vs 3080 for example) as they pass the point of diminishing returns of what the design can do and what's cheap and easy to manufacture.
But they exist because there is a marke of enthusiasts for whom efficiency or price does not matter, they just want the best performance money can buy because maybe their business use case benefits from the 10% shorter render/compile times at the cost of 50% extra power consumption. Who are you to judge?
To feel better about my non Intel purchase by laughing at the performance.
If that is true they should be comparing it to the Epyc 7763 or other CPUs that aren't meant to be reasonably priced. No reason a true enthusiast would pick this CPU over Epyc.
That's a server chip/line, no? And therefore will have significantly worse per-core performance, one of the most important things for non-embarrassingly-parallel compute tasks?
Edit: and after a cursory search, is it not also true that Epyc line is 4-8x higher priced than the 12900k MSRP?
It’s a self deprecating joke, dude. People make these to de-escalate situations safely. I sometimes really wonder at the social skills of people on this website.
You really believe that? Feel free run your computer after you take off the heatsinks of your computer, removing all the fans, and removing all the heatsink paste if you truly believe heat doesn't degrade pc components. Feel free to post the results.
I don’t, because you are wrong about heat. Heat matters to everyone. If heat didn’t matter on regular timescales graphics cards would never break and crypto miners wouldn’t care about their hardware getting hot.
The 480 was probably barely used and the mining GPU are used at low power and controlled temperatures. It proves my point.
Old doesn’t mean heat abused. Gaming laptops or ones with discrete GPUs often have issues and it’s always because of the GPU. If it posts it doesn’t mean that the GPU won’t die soon or is problem free, gaming laptops have the most heat abuse and are constantly dying and having issues. Performance degradation because it’s old is a different claim from saying heat damage doesn’t exist.
feels like m1 broke a lot of people's brains about performance-per-watt. like, maybe that metric matters if you're running a laptop that has a battery and physically heats up your lap. but nobody with a desktop pc cares about a 100w tdp difference so long as the chips can be cooled
I care about a 100W TDP difference in that a cheap cooler will sound like a jet engine, and a powerful cooler will cost more, in addition to a hot CPU making my room warmer.
it affects cooling. gpu dies are way bigger than cpu dies. a cpu that draws almost as much power as a high end gpu is way more difficult to cool. I don't really care about price/performance that much, but I don't want my pc to sound like a jet engine at load.
> I honestly don't get why people are so focused suddenly on the power consumption for a desktop CPU designed for enthusiasts who want top performance at any cost.
These are individuals with an interest in harming Intel and propping up AMD.
Intel is doing a fine job themselves in terms of harming intel and propping up AMD. Intel themselves has focused on power efficiency after ivy bridge, and people care about energy bills and being green.
> There's no way that Intel processor is beating a 5950X that is also drawing 300W. At that point some Ryzen's will be operating at close to 6GHz per core.
Modern desktop CPUs are already being pushed well past the peak perf/watt efficiency in order to extract the maximum amount of performance. A Ryzen 9 5950X would need substantially more than 300W to hit 6 GHz, and you would need exotic sub-ambient cooling to keep it stable at those speeds.
A more reasonable comparison would be against EPYC and Threadripper, both of which use their respective power budgets for more cores rather than more clocks. A 64-core EPYC 7763 has a TDP of about 280W, and it's going to substantially outperform the Core i9-12900K when given 300W of power.
Outperform it assuming your workload can actually saturate 64 cores. In the desktop market though, and heck often in the server market, you're going to struggle to do that. Single thread performance is still hyper important and that isn't going to change anytime soon. Intel seem to have the edge there, in a big way.
You can give a 5950x all the wattage all day but it really doesn't get much faster until you start dropping the temps to sub ambient. I have a 480mm radiator dedicated to my 5950X along with high quality RAM and extensive tuning (sub timings, fabric clocks) and I'm not able to get anywhere near 6 GHz at room temp without a voltage level that would destroy the chip after a few benchmarks.
And while I think my highly optimized 5950x setup could probably eek out the 12900k cinebench multicore score I'm sure if you gave me 10 minutes to also be fair and tweak a few settings on the 12900K I could get that to win again, even without messing with max power.
I do agree the Ryzen numbers in the article do in general seem lower than normal though.
Given this is Intel's "Catch up multiple nodes, new performance design, 300 watt class, ddr5" CPU and it's only able to get this eek ahead vs AMD's CPU from last year which is about to get the same litany of improvements plus their stacked cache improvement I'm not holding my breath Intel is really innovating fast enough again but at least they don't seem to be sitting still anymore.
What is the cold temp helping with? AFAIK transistors no longer get faster at cold, at least since 16nm the perforamce at sub 1V thin-ox transistors is pretty flat with temperature. Cold helps a lot with power though, leakage at hot (100C) could be 20W, so maybe going cold allowed more dynamic power dissapation? I'd be intersted to hear what the overclock experts like yourself think.
Transistors have a maximum operating temperature[1] (Tj_max), and the package has a fixed thermal resistance[2].
If the junction-to-case thermal resistance is 0.5 deg C/W and CPU dissipates 100W, then the junction will operate at 50 deg C above the case temperature, whatever that is.
If the case is kept at room temperature (say 25 deg C), which is the best possible a simple air or water cooled system can do, then the junction will operate at 75 deg C.
For simple discrete transistors and diodes, Tj_max is typically 150-175 deg C. For modern CPUs I believe it's closer to 100-125 deg C.
Thus, lowering the case temperature means the chip can consume more power before reaching the maximum junction temperature.
A greater power budget means you could raise operating voltages to help with overclocking for example. Keep in mind power dissipation scales with voltage squared.
There might be other effects too, I'm no overclocking expert, but this is one aspect.
Speed, is that due to Vgs threshold voltage decreasing as temp increases? On the other hand you have Rds(on) rising as temp increases, which would negate some of that effect I would imagine (lower gate drive current).
But not my field so no idea how that interplay works out.
Thanks. You definitely don't want the junction temp anywhere near 125c, the leakage would be very high. I thought most processors in due temp was 75-85c max, and will throttle severely if it goes above that.
Unless anything has changed the silicon/die itself isn't staying sub ambient at all it's just that there's a limit to hot steep temperature gradient you can build when the ambient/sink temperature is fixed.
If 5950x was designed for 240W max power like 12900k, then it is likely that the optimizations AMD would have needed to implement in the silicon would have also provided some additional frequency headroom. It's all hypothetical though since all we can do is compare the chips that were actually produced.
In the Gamers Nexus review, they were using a high end 360MM AIO cooler and the CPU Temp shot up to 74C almost immediately during a blender session. Definitely a hot, power hungry CPU.
It seems like all the focus has been on how these chips are at maximum load. I wonder how they perform under more typical light/mixed loads.
AFAIK, the theoretical benefit of doing the big-little arrangement is improved power scaling with load. At partial load, powering a wimpy "e-core" at its ideal clock speed should be more efficient (instructions per kwh) & possibly have better latency characteristics than the usual approach of doing a drastic underclock & undervolt on big performance-oriented cores.
Naturally I'm interested in seeing whether that theoretical advantage of the architecture has paid off. The maximum speed performance stats and wattage numbers give an incomplete picture. I want to see wattage and performance metrics across the spectrum of load levels, from <1% all the way up.
If the 12900K delivers roughly 5950 level performance using 25% more power than the 5950, that looks pretty bad. But if the 12900K can deliver 25% of its max performance for only 10% of its max power envelope, and the 5950 needs 30% of its max power envelope to keep up with that load, that's a big deal, and the intel chip will actually be the cooler one more often than not. That dynamism is what reviewers need to focus on explaining & quantifying, IMO.
The LTT review [1] compares full-load-render benchmark tests against more real-world gaming performance, and the temps drop from a sustained 90 degrees to 60 degrees, so you seem to be right, there.
> In the Gamers Nexus review, they were using a high end 360MM AIO cooler and the CPU Temp shot up to 74C almost immediately during a blender session.
FWIW, my water cooled AMD Ryzen also shoots up to very high temperatures immediately when loaded. It can touch 80C like a lightswitch.
Water coolers have a lot of thermal mass in the water and the metal radiator. The CPU can't instantly heat all of this up.
What's actually happening is that the turbo boost is pushing the CPU as far as possible within the thermal limits on a short-term basis, then slowly lowering the boost frequency as the temperature saturates. Power consumption is very non-linear, so each additional MHz of boost is a significant amount of additional power. The boost algorithms are designed to spike high for short bursts at the onset of load because most desktop operations require short bursts of single-threaded performance.
> FWIW, my water cooled AMD Ryzen also shoots up to very high temperatures immediately when loaded. It can touch 80C like a lightswitch.
It might be interesting to look into a block which covers the chiplets more. Mine does max 55 at normal speeds and PBO max/unlimited power it does about 70C max (and about 220W draw for a 5900X). I have no peaks like you describe. That used to be a bit more with my old water block which had copper on the chiplets but no water channels.
> FWIW, my water cooled AMD Ryzen also shoots up to very high temperatures immediately when loaded. It can touch 80C like a lightswitch.
you should probably looking at a nocturna dh15, no ryzen (on the market) will get to 80° when loaded, not even in long blender sessions)
so you probably have a faulty water cooling or smth.
In my experience with a 5600X and water cooling (nothing crazy, just an off the shelf AIO) sustained full CPU load will be in the 70s, but I've seen it hit 80 or so briefly.
What's the die size of the ADL silicon? I know my 5800X will rocket up past 70C even on a custom loop with 30C water temps because the chiplets are so small and power dense that there's not enough surface area to transfer heat any quicker to the IHS.
> Also their Ryzen numbers are beyond suspicious. On a silent configuration (peak power draw ~209W, 175W sustained, some undervolting) my Ryzen 5900X hits a Cinebench R20 score of 8835 vs their 8390. I suspect my rig is worse since I cheaped out on the RAM, but I wouldn't know because they omit information about the Ryzen test rig.
Firmware quality for AM4 boards is ... "variable". My board restricted CPU power to 60% of nominal because they put in the wrong ADC scaling factors in some BIOS revision. Some boards have PBO enabled by default (which nets big % on the kill-a-watt, less % in performance). Many boards have bad defaults when you enable XMP which steal a few Watts from the actual CPU cores.
CB mostly scales with core frequency, memory and FCLK are less important.
> At that point some Ryzen's will be operating at close to 6GHz per core.
There's no way any of them will get close to 6 GHz without LN2.
> There's no way any of them will get close to 6 GHz without LN2.
No need for LN2, you could probably achieve this easily using a closed loop CO2 system. But I'm not aware of any such systems readily available. Kinda seems like an unfilled niche to me.
If I had to guess the TDP of the new Intel CPU when maxed out is actually around 270-280W. Not all of that 307W is the CPU, there's power used by the motherboard itself, a mostly idle GPU, fans, NVME SSD, etc.
Still a very serious heat producing thing in one socket compared to the AMD.
This also means it's going to require some rather expensive motherboards since reliably delivering 280W to one socket is no simple task.
You don't have to guess, Intel themselves say the "peak" TDP of their top i9 is 250W. All to slightly beat 5900X which tops out at 150W. It's insanity, literally have no idea who will buy this.
A few lifetimes ago, I worked at a small company where the VP shared with me why we had some "caddilac" service plans (ie, super expensive, cost/benefit=unreasonable).
His reply: These are the showroom-only models where customer won't buy but will feel their (already expensive) not as high end plan is more reasonable.
Essentially it's a part of product marketing segmentation where you create an unobtanium segment that's only there to sell other segments.
> It's insanity, literally have no idea who will buy this.
It's not insanity. Overclocked enthusiast builds have been pushing over 250W and higher for many years. Quietly cooling a CPU putting out that much heat isn't a mystery any more. Modern coolers are very, very good.
That Intel CPU doesn't come anywhere near 250W during gaming or normal operations anyway. It's not like it's consuming 250W every single time you do something.
Modern GPUs also consume well over 300W during normal gaming, which will be double or triple the consumption of most CPUs. I don't understand this sudden concern about Intel shipping CPUs with high peak TDP.
>>That Intel CPU doesn't come anywhere near 250W during gaming or normal operations anyway. It's not like it's consuming 250W every single time you do something.
It does though, that's the whole reason why I'm complaining. While I'm doing a video transcode on my 5900X putting all cores on full load it won't go above 140W. Do the same on the i9 and it will do it marginally faster sitting at 250W.
The point isn't complaining about 250W usage in general. Of course there are other chips, including other CPUs that use that much or more and it's not the end of the world. The problem here is that you're paying a HUGE price for a very minor advantage over the competition or even their own chips. If you clearly can buy a chip that performs very slightly worse for over 100W reduction in power, why would you buy the space heater one.
Yes overclockers have been doing this literally forever, I'm aware, but this is intel making this decision on stock CPU - I just don't find this acceptable at all.
>> I don't understand this sudden concern about Intel shipping CPUs with high peak TDP.
To summarize the above - because clearly the high peak TDP is much higher than that of competing CPUs for a very minor increase in performance. That's just not ok.
You don't get the bleeding edge tech for practicality or reasonableness. You don't buy a Bugatti Veyron for cargo space and gas mileage. You get silly tech for insane performance at massive cost. Eventually this tech is refined, optimized and made reasonable.
It's what humans do. We don't cross the Pacific on a tiki raft or fly to the moon in a tin bucket because our destination is nicer than where we are currently, we do it because we can and because it's awesome.
I think that's an extremely poor comparison, because this isn't some weird one-off extra special product that is meant to push us forward ahead in computing power. This is just a regular competition for AMD at this point.
To stick with the car comparison, let's use Mercedes C class and BMW 3 series as direct competitors, right? They offer broadly the same features, cost the same amount of money, look similar-ish, you get my point.
So now imagine that one year, BMW releases a new 3 series with revolutionary new engine that makes it go faster by few percent than the C class on average, but at the cost of double fuel consumption.
I'm sure you'd go - what are you doing BMW. You aren't Lamborghini, you aren't pushing the envelope here - you are just pushing your existing framework to the absolute max to beat the competition at a huge cost.
This isn't progress, and this isn't a bugatti veyron equivalent. This is just technological dick waving with no regard to power consumption whatsoever.
The AMG and ///M models don't ever get compared on mileage, and almost noone who buys them cares in the slightest. They want to know about the performance, full stop.
Sure, but this isn't a Core Extreme model(remember those?), This is a c220d using twice as much fuel as a 320d, just to be 0.1s faster to a 100. If that's a good trade off for you, then sure, but I stand by my opinion that this isn't progress and this isn't some incredible humanity pushing achievement - this is just allowing the CPUs to run at crazy power just to beat the competition ever so slightly, because intel still hasn't sorted out their manufacturing nodes.
I think you misunderstand me; I agree with you on this being stupid.
I should have worded it better, as having owned quite fast cars when I was younger the running costs (both to my wallet and the environment) were something I eventually realized were more important than the performance.
I’m curious why you think that? Generally if running costs are compared it’s with a laugh. I’m in a position to mentor young people with large amounts of disposable income who disproportionately spend it on cars (I was one of them once) and I’m about the only person I know who would give them actual running costs to consider.
But even people who want performance must do a calculation in their head whether something is worth it or not. A CPU that uses 100W+ over the next competing CPU for a very minor performance gain is clearly not worth it, like, surely this isn't a mad assumption.
Let me make this even clearer - Intel provides two numbers this time round. One "typical" power usage, and one "peak" power usage for every 12th gen CPU. The peak on the top i9 part is 250W, and looking at the reviews that came out today, it seems correct - the CPU will consume 250W of power when running at full load.
I did, my point was more that the Intel and AMD TDPs aren't really super directly comparable because they both use different formulae to compute the numbers.
Maybe I misunderstood you?
I do agree with the second part of your reply, and I also don't know who would buy the Intel part over the AMD part. I think Intel's move into space heaters is disappointing, and once AMD revs their process, micro-architecture and moves to DDR5, I suspect they will once again smoke Intel. Pretty poor showing in my opinion.
>>I did, my point was more that the Intel and AMD TDPs aren't really super directly comparable because they both use different formulae to compute the numbers.
Why do these formulas matter though, if we know, without a doubt, that a fully loaded 5900X/5950X will just sit at 150W usage, but the new 12th Gen i9 at full load sits at 250W. That's not going by what Intel says or what's on the box, that's actual measured power usage which has been now confirmed by reviews.
So yes, sure, let's not compare the given numbers in the spec - but real life numbers do back it up this time.
> We however do know that they used DDR5 RAM for the Intel processor, which the current gen Ryzen don't support. Likely some performance was gained there as well, but obviously they didn't test the Intel processor with DDR4 RAM to get an idea of how much comes down to faster RAM.
>We tested Alder Lake on the latest Windows 11 below, but our Ryzen results are on Windows 10—this made certain to avoid AMD being penalized by the current Windows 11 regressions in L3 cache and Preferred Core selection, while giving Alder Lake the big.little architecture support it needs from Windows 11 itself.
At roughly 25W per Performance Core running at 5.2GHz on Intel 7nm High Performance Node, with a ST Geekbench score of ~1900. Compared to The Apple M1 Max running at 3.2GHz on a TSMC 5nm Low Power node, at roughly 5W per core, with Geekbench score of ~1750.
The is 5x the power for a 8.5% of performance increase.
But I know there are many who just want to get the best ST performance regardless or power usage. And considering someone Overclock the thing past 8Ghz barrier. It is still pretty damn impressive.
That is the difference between a consumer review and tech review. They may be meaningless to consumers, but they are meaningful for those of us only want to learn about tech.
> They may be meaningless to consumers, but they are meaningful for those of us only want to learn about tech.
It's actually misleading to compare apples and oranges like this.
CPU power efficiency is very non-linear. For example (made up numbers for example), it might require double the power to go from 3GHz to 4GHz and another doubling of power to go from 4GHz to 5GHz.
If you're designing a low-power laptop chip, you take advantage of this fact and operate it at the lower frequencies. You may be able to cut the power consumption by 1/4 and still get 60-70% of the performance.
If you're building a desktop for enthusiasts who are buying big liquid coolers, you go the other direction and push the CPU to the limit. That's what Intel did here.
This factors in at the design stage, too. It's not as simple as taking the same design and scaling it up or down. You need to target your timings around a range of clock speeds and voltages.
Drawing comparisons isn't as simple as dividing Geekbench scores by power consumption because the underlying math is very non-linear.
>It's actually misleading to compare apples and oranges like this.
I never make the claim it scale linearly. And I have been very explicit in the node simply because I expect technical reader should know it doesn't scale linearly ( may be I should have wrote it down just in case ) and as I wrote in my another reply you cant even use the TSMC N5 and scale to 5+Ghz. It is not optimised nor tested for that clock speed.
To make the comparison on a low power node, Alderlake using 5W at 3.2Ghz, and scale linearly with GB5 score, it would get a result of 1140. On the same power, M1 is able to score 50% higher. Or 50% increase in IPC.
But that comparison in my view is unfair if not truly misleading. Because there are a lot of assumption involved, from scaling to design and performance characteristics. Where as the previous comparison was done with two actual product based on facts, and you should draw your own conclusion within the context.
OTOH when Apple takes an M1 from a laptop and puts it in a desktop... the performance and power are exactly the same. So it probably is fair to compare M1 Pro vs. 12900K.
Most reviews are meaningless as a whole in the global context; most people can't afford any of the high performance parts of the electricity to power them anyway, and those that can now have the problem of shortages making it hard to get the parts at all.
The only thing all of us get out of this is: context matters.
> parts of the electricity -> parts OR the electricity
Quite the distinction there. I was referring to reviews in the global context where there are plenty of places on earth where even getting power is hard and you might be constrained to limited wattage or battery backups that drain very fast on high performance parts. All to build up an extra example context where there is no 'absolute' truth to some reviews out there.
Yes except I got asked again and again what is "Intel 7". So I am going to stick with a simple naming scheme that right now happens to be about the same across all three foundry working on leading node, Samsung, Intel and TSMC. And everyone can relate or understand.
If that is the case shouldn't you call it "Intel 10nm" since that is what it was called when Intel was still using the same units as everyone else (even if those units are becoming meaningless)?
Not even Intel itself says that "Intel 7" means "7nm", right?
Even if Intel was using "nm" like everyone else, Intel's naming was closer to reality then the rest of the industry. TSMC and Samsung's "nm" is more of a marketing term.
Intel's 10nm is relatively close to TSMC's 7nm in term of density. That's why Intel renamed it to "Intel 7".
Technically, even TSMC calls it N7, and not 7nm.
But as you said, those units are more or less meaningless. Its just marketing.
Intel have a different process that used to documetned and marketed as 7nm. That process since has been renamed to Intel 4.
Regardless of discussions about actual density and the meaningless, it's important to make the discussions between Intel 7 and 7nm (now Intel 4) to prevent confusion. Otherwise people might mistakenly think Alder Lake is on the process previously known as Intel 7nm.
Fortunately, once we get to Intel 4, it becomes unambitious again.
Apple ditched all the x86-32 code with Big Sur (one reason we can't upgrade at work..). So its a subset. They did the same with 32 bit iOS Apps. Apple is happy to leave old software behind.
More seriously, there's always a few comments on threads like this claiming it's unfair to compare Intel to AMD or Apple because "they have a larger node size and are hamstrung by it." Which is Intel's choice, they were happy having a superior node all to themselves for decades and now it's a problem for them.
I don't remember people saying AMD was just as good as Intel but was just on a worse node either, did it matter to anyone but AMD fanboys?
For those that do not care just about the faster CPUs on benchmark and tribal fights, and like instead to nerd out about architectures, it is still interesting to know whether Apple superiority is due the ARM architecture, the Apple microarchitecture improvements or TSMC process.
It is likely a mix of all three and of course microarchitecture cant be easily disentangled from the process.
Unfortunately I’ve seldom seen discussions framed in that architectural curiosity context.
The drowning out discussion noise seems to be much more of the macho-tribal variety of shallow assumptions and absolutist conclusions.
And that’s really too bad, because I do really enjoy conversations framed in honest wondering and non-biased estimation of the future potential of various architectures.
Yes. That is why I have been very explicit about node numbers and characteristics along with clock speed. Apple cant push to 5GHz without using another node and possibly some redesign. Intel using their own ultra low Power node or even TSMC 5nm low power node will not be able to reach 5Ghz either.
And let people make their own judgement. Again this isn't for consumer review, just for those interested in tech and how it all fits together.
These biased sites will declare Intel the winner despite the higher cost in motherboard, CPU, power usage, cost to overclock, and heat creation. These tech sites are obviously nothing but product shilling sites with a cheap veneer that relies on their old reputation, but after seeing them all saying the opposite of reality with Ryzen, I won't give them clicks and I won't buy Intel again.
He's not wrong per se though he's wrong in this case, because many tech reviewers were overly enthusiastic about Ryzen parts and acted as-if they're much better than their Intel counterparts because they performed better in various benchmarks and productivity applications; this is kind of a meme in the gaming scene (where AMD's best CPUs generally still struggle with Intel's 10th gen offerings).
In tech your product has to be REALLY bad in order to receive critical reviews and manufacturers are basically smart enough to just not send these bad products to reviewers, so you basically get no or few reviews on them. Most reviewers are completely dependent on the manufacturers to give them stuff to review at no cost, and obviously when your reviews are too critical they're going to stop sending you stuff, which threatens the reviewer's business. There are very few reviewers who can do highly critical reviews, GamersNexus comes to mind. Notice how they're often not getting stuff from the manufacturer but buy it off the shelf instead.
Lots of tech reviews will also not say anything bad, for instance anyone critial of Apple won't get access to their events, products, and will be blacklisted.
You can't trust these publications at all anymore.
Dude, it's a company that makes consumer goods. It's not the government refusing to credential a reporter.
If you are actually an objective news company trying to evaluate a product, then do what Consumer Reports does and go to the open market and buy the generally available product for the full retail price, and then evaluate it. That is a requirement to be an honest broker not influenced by a vendor, which is why Consumer Reports does it. So make up your mind as to whether you want to be a journalist that is independent or whether you are willing to compromise and do P.R. work for apple in exchange for early access.
This is just a list of issues and articles that came to my mind. Every of these sites also criticized the thermal of the last intel macbook pros and all hated the touchbar and all of them are still reporting on apple on exactly the day the embargo lifts. wtf are YOU on about?
If you don't cherry pick there are enough critical articles and critical comments in every review (You said there aren't in your first comment.
You are totally right they must have a giant blacklist if blacklisting is for you not to send demo units of unreleased products to every outlet. THE BLACKLIST IS GIGANTIC DUDE.
If you ever look for help call here The National Alliance on Mental Illness (NAMI): 1-800-950-6264
We've banned this account for breaking the site guidelines egregiously and repeatedly. You can't do that here, regardless of how wrong someone is or you feel they are.
If you don't want to be banned, you're welcome to email hn@ycombinator.com and give us reason to believe that you'll follow the rules in the future. They're here: https://news.ycombinator.com/newsguidelines.html.
Your comment is not in the spirit of HN and I hope it is deleted. You are asserting someone has mental illness just because they are claiming something which is well known.
Gamers nexus, which is a top reviewer, have mentioned they dont get some products due to their critical take. It's not an unreasonable or outlandish view at all. Its ridiculous to be so aggressive against another person just to protect a brand. Do you perhaps work for or have shares in Apple? I cannot fathom why someone would be so desperate to protect the image of a billion dollar company otherwise.
No one is entitled to get early access to non released hardware. The outlets that do get access usually sign NDAs and they are not allowed to take demo units apart and you know what GAMERNEXUS also says in their videos that they don't sign these NDAs. If he had the chance to sign it it seems he decided to not do it not the other way around. The spirit of HN was that it's usually pretty good moderated against trolls and flamebaits like the guy above.
You also broke the site guidelines egregiously, and you've posted flamewar comments before this too. I'm not going to ban you right now because you've also posted good things, but if you continue to break the site guidelines, we will ban you. Please review https://news.ycombinator.com/newsguidelines.html and stick to the rules when posting here.
That means thoughtful, substantive comments, curious conversation, and respect towards other commenters, no matter how wrong they are or you feel they are.
On pure performance metrics Intel will be the winner every cycle they bring one "enthusiast" CPU that tops the benchmark in gaming or creator CPU loads. they are expensive they are power hogs but they are the fastest. I don't get why you are so adjetated by it.
>they are expensive they are power hogs but they are the fastest.
That is absolutely false and a lie for multicore processes, real world usage, and programs that I use. This is an enormous lie, unless your workload only consists of synthetic benchmarks that are Intel optimized and ignore the gains from other hardware like premium DDR5 they use to set these sythetic benchmarks.
As I said in my other comments you are delusional. have a good one. You see a conspiracy theory without proofing it and you disregard the facts from several high regarded sources like anandtech, phoronix and everyone who has published their benchmarks today. It's the literal definition of a delusion.
You think every benchmarking site, even the sites that use open source software to benchmark is getting payed by Intel? Phoronix and Anandtech have open benchmark suites you can go and check the results yourself. You are arguing that commercial benchmark providers, open source software developers, tech journalists and enthusiasts and hobbiest are in a conspiracy to favor Intel in performance benchmarks without providing a single source stating your acquisitions except yourself. Insane.
This is just a list of issues and articles that came to my mind. Every of these sites also criticized the thermal of the last intel macbook pros and all hated the touchbar and all of them are still reporting on apple on exactly the day the embargo lifts. wtf are YOU on about?
Critical articles like “iPad mini review (2021): The best small tablet gets a facelift”?
https://www.engadget.com/apple-ipad-mini-6th-generation-2021...
You’re grasping at straws since ifixit doesn’t get Apple press invites, hardware and you linked to searches of Apple praise since you can’t find evidence to prove your point. Thanks you proved my point for me and linked to articles that prove I am right.
Its very good to see three viable, great emerging CPU architectures. Intel is important, and them getting their CPU node up is critical. AMD has saved x86 from irrelevency and stagnation, and has been sitting on their next node while Intel caught up, but now has to release it to keep their spot on top. Apple is remaking the game with the M1 / Pro and Max variants - and we still have Jade 2c and Jade 4c coming.
Things happen when someone wants to work on them. Running Linux on Apple Silicon machines is a lot more exciting than running Linux on a generic Intel machine with an Apple logo, so there's been a lot more development effort on the former.
You can already get a usable desktop on an M1 Mac, if you really want to. The CPU is fast enough that you don't absolutely need graphics acceleration, and Wifi, USB, and display output all work.
The big missing piece is the GPU, but Alyssa has a fully-custom user-space implementation for macOS that largely works.
Saying Apple has remade the processor game is like saying the PS4 remade the gaming market. Yeah, it's true. But only for people that use consoles/macs. That kind of hardware only supports the configuration it's sold in and not full spec. If you try to do things that 99% of people won't do it won't work (ie, attempting to boot off an external drive when the internal one breaks).
I disagree. Sure primarily faster more power efficient Apple desktops/laptops benefit Apple users. But it also helps people realize what arm designs can offer. The M1 Max is a marvel, decent GPU performance compared to discrete, and amazing GPU performance for integrated. Amazing perf/watt, and also amazing memory bandwidth. Desktops and Laptops are primarily in the 50-70GB/sec range, M1 Max is at 400 GB/sec.
Suddenly Intel and AMD have to keep an eye not just on each other, but also on the various ARM designs targeting microcontrollers up to supercomputers.
I don't get why people are pouncing on the memory BW thing. There are very few applications which are actually bound by memory bandwidth (most of these will never be used on a laptop) and I've seen nothing so far to suggest that the CPU cores in the M1P/M1M can actually use it. The M1P/M parts are basically a midrange GPU and CPU in the same package and use a wide, high-speed memory interface (LPDDR is conceptually more similar to GDDR than standard DDR) to get GPU-like bandwidth because there's an actual GPU in there - which is actually rather impressive.
For reference, in Zen 2/3 a chiplet (of eight cores) is limited to around 25 GB/s write, 50 GB/s read.
Edit: The anandtech test has memory B/W numbers and gives 100 GB/s for a single core and around 220 GB/s for all cores, which is extremely high, but also not the full memory bandwidth.
One clear benefit is decent GPU performance, without the cost, physical size, and power/cooling for a discrete GPU. It's way faster than Intel or AMD's integrated graphics. You can see this difference when you unplug a MBP vs any of the laptops with a discrete GPU.
Marketing brags about memory bandwidth based on clockspeed * bus width = 400GB/sec. Getting 60% of peak on some memory bandwidth benchmark is pretty common. Try McCalpin's stream benchmark on your platform of choice to verify. I suspect you'll find similar on Intel or AMD on desktops, laptops, or servers.
Not sure I buy the memory bound argument, sure peak performance will not change, but worst case performance can be caused by cache misses. So I expect that the new MBPs will be more evenly fast than the x86-64 competition, even while doing stressful things. I don't currently need a $3k laptop, but am hoping they ship a mini with a M1 Max or Pro.
To quote anandtech:
The fp2017 suite has more workloads that are more memory-bound, and it’s here where the M1 Max
is absolutely absurd. The workloads that put the most memory pressure and stress the DRAM the
most, such as 503.bwaves, 519.lbm, 549.fotonik3d and 554.roms, have all multiple factors of
performance advantages compared to the best Intel and AMD have to offer.
So an apple M1 Max is 19% faster on SpecFP (floating point applications) as a Ryzen 5950x which has twice as many fast cores (16 vs 8) and runs at 105 watts TDP. That's pretty amazing in my book.
Memory bandwidth per core, per clock cycle is actually very limited on many systems. So you end up being bound by memory bandwidth any time you're running compute-intensive apps on multiple cores, unless thermal constraints hit sooner - which is quite unlikely on Apple silicon, and ARM silicon more generally.
A Dell Latitude 5520 Laptop, 15inch, 4k, 2TB SSD is $4,099.00 form Dell (after a 35% off coupon). A Apple M1 MAx 16 inch, with a far better display both in quality and display, better keyboard, same storage is also $4099.
Hmm. I don't know that model, but I have bought two Dell XPS15s in the past year. One had 4K, 64G RAM, 2T SSD and 8 cores. It was $2500. The other was slightly lower spec and $2400.
PC laptop pricing fluctuates so much and I don't know why. You can kit out a ThinkPad P15 with these specs for ~2500 right now. Next month the prices will be flipped. Go figure. At least Apple is consistently priced high.
ThinkPads are made of shitty plastic, have stupid keyboard layouts, worse trackpads, and come preloaded with malware/bloatware. I say this an exclusive x86 user. ThinkPads are average quality tools, MacBooks are delightful, beautifully designed devices. Really tempted to switch
But if my workload matches what the M1 is good at, I’d rather have a super quiet, cool laptop that can give me max performance off the battery for the whole day.
I am curious how much of the M1 gains can be realized by just slapping 16 gigs of cache on your die. I know that's a gross oversimplification, and I'm 100% just going off intuition here but it seems to me that memory access efficiency is what carries the brunt of the M1's gains.
Very wide in the sense of memory bandwidth particularly, or is there another kind of wideness at play here?
"has a big reorder buffer", I'm interpreting this as "one of the notable advantages of the M1 is it's ability to be more clever about processing instructions out of order to maximize resource utilization". Is that about right?
Width in the sense of how many instructions can be executed at the same time.
If you’re really interested in this it might be worth finding a copy of the Patterson and Hennessy book. It’s a big read and expensive (but older versions are on the internet archive [1]) and covers all these design issues in quite a lot of detail.
Gotcha, I wasn't really aware of instruction level parallelism
previously. I think the pieces of the puzzle are a lot clearer to me now, thanks for the replies.
At some point the physical distance to parts of cache would mean you'd have rapidly diminishing returns on adding more. For 16GB it'd need to be some kind of tiered thing with nearer cache segments being quicker. Maybe you could have it present itself as a giant unified cache... sort of like what they did in the new IBM mainframes.
The M1 parts actually have way worse memory latency (i.e. random accesses) than both AMD and Intel. A finely tuned Intel system has around 3x lower memory latency than an M1P/X.
All of this is SDRAM, the S stands for synchronous and means that the memory is driven according to fixed timings in relation to a fixed bus clock. All LPDDR4X-4266 parts with the same timings perform exactly the same, whether they are soldered to the interposer or are 5 cm away on the board.
That's not at all what I'm seeing. Can you be specific as to what chips you are looking at and what benchmark you are using to draw this conclusion?
I'll go first.
Personally when talking about memory latency, I want to measure time to main memory, without TLB thrashing. Not that TLB latency isn't interesting, but it should be separately quantified.
I've written a few microbenchmarks in this area, but I get very similar numbers to the Anandtech R per RV prange. Which puts the latency to main memory at around 35ns looking like it might flatten out at 40ns. Yes, full TLB thrashing is up at 110ns or so, but that's not a usual use case. If it is, at least under linux, you can switch to 1GB pages if that's important to you.
R per RV prange numbers on the Ryzen 9 5950x is 65ns or so.
So sure the TLB worst case is higher latency on the M1, but then you have to figure out how big the TLB is, and how often that's an issue if you want to know the real world performance impact.
The i9-12900K with DDR5 gets 30ns on the R per RV prange, and 92ns on full random (tlb thrashing).
Even assuming the worst case TLB behavior the m1 max is 111ns and the 5950x is 79ns or 1.4x higher. On the Intel side is 111ns vs 92n, 1.2x higher.
Finding it hard to find any numbers that make the M1X look like 3x higher memory latency.
I feel like R per R isn't quite the perfect match for the pointer-chasy workloads I have in mind where you end up going through more-or-less random selections of relatively small (<< page size) objects, so R per R amortizes the randomization a little much. I still think having the overall totally random access for a system is a good number precisely because it contains all the penalties and latencies that can add up. Embarrassing as it may be, I also compared the wrong numbers across parts here (and in the past too, I think) - thanks for calling that out. I was kinda irritated how exactly they would be able to get the performance they were getting across all workloads with such poor latency but failed to double check.
Ah, sensible. Generally with the TLB numbers and the memory latency you can get a good idea of the performance by interoplating between the two number for any number of TLB misses (from 0% to 100%).
The M1 Max also has a crazy number of memory channels, even the older M1 has 4 channels, pro has 8 channels, and max has 16. So you can have many more cache misses in flight, this is part of why the m1 max is 20% faster than the 5950x with twice as many cores.
I have not seen TLB size reports for M1/M1 Pro/Max, but Anandtech has reported following figures for the A14:
«The L1 TLB has been doubled from 128 pages to 256 pages, and the L2 TLB goes up from 2048 pages to 3072 pages. On today’s iPhones this is an absolutely overkill change as the page size is 16KB, which means that the L2 TLB covers 48MB which is well beyond the cache capacity of even the A14» [0].
For the default page size on A14 of 16kb, that gives the 4Mb L1 cache coverage (16kb * 256), and 48Mb (16kb * 3072) L2 cache coverage. Off the top of my head only POWER 9/10 (and maybe POWER11) come close with such large TLB's. aarch64 page sizes come as fixed presets, e.g.:
- 4kb, 2Mb, and 1Gb
- 16kb, 32Mb
- 64kb, 512Mb
The OS can choose either of the three, and OS X defaults to the 16kb / 32Mb preset, although I have not seen whether it can simultaneously handle both, 16kb and 32Mb, page sizes.
A garbage collected runtime that has been optimised to make use of large TLB's or large page sizes can exploit the full advantage of the increased TLB depth. Azul JVM comes to mind with their ZGC garbage collector having been heavily optimised for terrabyte scale Java memory workloads.
I am now very curious to see the M1 Max vivisection results that would reveal whether the TLB size in it is even larger (or not).
Honestly it's kinda crazy what's possible these days. Couple years ago you'd easily burn 30+ watts on just the memory chips to get this kind of bandwidth.
Indeed, 64GB ram, 400GB/sec, decent GPU, ML acceleration, 10 cores, etc all in a small package that gives good battery life to a relatively thin laptop.
Here's hoping they put the same in the mac mini. Anyone interested in a linux port join the Marcan patreon, I'm kicking in a few $ a month.
Interesting, I didn't know that the memory latency was worse. The bandwidth is still much higher right? In your estimation how much of the perf gains of the M1 chip are due to the increased memory bandwidth vs other optimizations/advancements?
I don't think you can single out one aspect when comparing designs which are so different, but I'd say most of the CPU performance is down to the core. There you have the M1 core which is (iirc) 10 pipes wide, much wider than both AMD and Intel, and a much fatter frontend as well to feed the beast. I think that's the main advantage the M1 has. It also has a larger L1 cache (twice as big as x86). These two are one of the few areas where x86 has an innate disadvantage because fattening up the frontend is much more complex on x86 compared to ARM, and you can't have a VIPT L1 cache larger than 64K with 4K pages, which is a hard-to-change default on x86, while the M1 by default uses larger pages (32K or something like that).
"x86 has an innate disadvantage [...] fattening up the frontend" this is inherent largely because of ISA complexity and variable width instructions?
"M1 core which is [...] much wider than both AMD and Intel"
The core width you're referring to is the decoder width yeah? As another poster pointed out the M1 has a large reorder buffer as well. Combined with the ability to index a larger L1 cache what I'm getting here is that the M1 can be a lot better about scheduling instructions (and perhaps even running non-interfering instructions in parallel on a given core (is that a thing?)) because the frontend has more power and space to do so.
I guess that efficiency is then a big part of the puzzle on why the increased bandwidth to ram makes such an impact?
Many things don't push the memory bandwidth and are cache friendly. However GPUs are bandwidth limited and the M1 Max does quite well against any other integrated graphics from Intel or AMD.
Even on the CPU side it can be a big win, in particular on SpecFPRate (a collection of heavy floating point real world codes, not microbenchmarks) Anand has this to say: The fp2017 suite has more workloads that are more memory-bound, and it’s here where the M1 Max is absolutely absurd. The workloads that put the most memory pressure and stress the DRAM the most, such as 503.bwaves, 519.lbm, 549.fotonik3d and 554.roms, have all multiple factors of performance advantages compared to the best Intel and AMD have to offer.
To drive this home compare the Spec2017 FP Rate, the M1 Max gets 81.07, the Ryzen 5950x (high end desktop with twice as many fast cores and a 105 watt TDP) gets 62.27.
So a low power M1 Max with half as many cores and much lower power is 30% faster than AMD's highest end desktop chip. Instead of a desktop size/volume/power, you can get it in a laptop that's 2/3rd of an inch thick.
Jade 2/4c are additional processors being developed by Apple for their <strike>Macbook Air line of notebooks</strike> Mac Pro line for anyone who is like me and didn't know.
Edit: first thing I read was wrong, turns out that these are Mac Pro level processors, see below comments for more details. Thanks everyone for the clarification.
All this die chopping really sounds like the consequence of the Intel architects Apple hired away. Over the years lots of Intel products segmented SKUs by laying them out in a way that facilitated simple chops post wafersort.
My understanding is that they are rumored, code names.
But they are expected be even higher scaled iterations of Apple Silicon. In other words, while M1 had up to 4 performance (P) cores, 4 efficiency (E) cores and 8 GPU (G) cores, the M1 Pro and Max scaled up to as high as 8P, 2E and 32G.
The Jade 2/4c designation is rumored to go even higher on P and G cores, which means it's much more likely to end up in a Mac Pro than the Macbook Air.
That has been the case with every site I looked at, its full of Intel shilling, they compare a CPU that is many times as expensive as AMD, see its got a slight edge with way higher power usage and declare the Intel processor the winner.
CPU prices are usually much closer to MSRP except perhaps right at launch. You can put a 12700K or 12600K in your bag at Newegg right now at ~MSRP+$20, and you were able to do that for the 12900K Newegg pre-order last week. The 12700K at $449 on Newegg right now is some excellent value you have to admit (same P cores as 12900K, 4 less efficiency cores) any way you look at it, and the 12600K even more so.
Claiming that the Intel CPUs are "many times more expensive" than the AMD comparables is just wrong.
If you want that CPU you’ll have to pay much more than MSRP, and the MSRP is higher than the real world price you can get the AMD for. Using historical prices with pre order prices that nobody can get now is a dishonest way to reflect cost. MSRP is a useless representation in this case and for GPUs.
Today is launch date. It's natural that there are less stocks (and only scalpers remain). I tend to see Ryzen scalped than Intel, partially thanks to Intel makes massive amount of CPUs.
Watching several detailed YouTube reviews (Gamers Nexus, Hardware Unboxed), the performance difference is 0-20% above a 5950x. Long duration (above 8 min) render workloads seem to push AMD back to break even, but memory dependant mutlithreaded bursty workloads heavily favor 12900k. Of course halo products are going to halo. The interesting story may be the 12600k, for it's price to performance considerations. Cooling the chips is not simple, air cooling is not enough, one reviewer recommending 360mm AIO as a requirement.
Edit: Gaming benchmarks are all over the place reviewer to reviewer. Some showing huge jumps, others not, game to game.
Looks like the i5-12600K is the real value play for Intel's 12th gen. It essentially beats Ryzen 5600X at similar power consumption. LTT review: https://www.youtube.com/watch?v=-EogHCFd7w0
That said, the cost of ownership is likely much higher for Intel vs. AMD - due to requiring DDR5 RAM and new motherboard.
You don't need DDR5 RAM. DDR5 is a halo feature at this point, only for the people for whom money is no issue. There's zero reason to use DDR5 RAM with the 12600k (considering you could be served better by spending the extra money on a better CPU or more capacity). The performance benefit is varied and probably not more than 5% at best.
The increased price of Z690 vs B550 and even X570 is significant though.
Twice as many channels so twice as many cache misses in flight. The Anandtech shows significant gains on multi core workloads. So the I9-12900K with DDR4 shows 61.33 int, 59.55 FP. With DDR5 it gets 80.53 Intel and 81.85. Some of that is increased bandwidth, but having more memory requests in flight helps as well.
Imagine all the cores generating cache misses and you have 2 channel of memory. When a channel handles a transaction it's unavailable for 40 ns (keep in mind that @ 3GHz with 2 instructions per cycle that's 240 instructions per core).
Now with 4 channels you can have twice as many transactions in flight and your cores spend less time idle and waiting on memory.
So over a 30% increase in performance when going from DDR4 to DDR5, of course this doesn't matter if your application has a high rate of cache hits.
DDR5 has more future promise, whether that's lower power, higher bandwidth, higher capacity, maybe lower latency too. There's something about DDR5 requiring ECC, but it's not clear if that includes actual CPU/OS reporting or if it's internally using ECC but with no error reporting.
But, I'm not sure how much of that shines through in the first DDR5 product vs mature DDR4 products. Now is the part of the cycle where the old memory is still high volume and the new memory is starting to ramp up; in not too long, we'll have the part of the cycle where they made too much of the old memory and prices drop and it's time to scoop up all the old memory you'll ever need.
DDR5 include on-die ECC. Which is very different from traditional ECC. But many vendors marketing (and many enthusiasts are talking) as if it's the same. See [0], there's some really good explanations in the replies.
Well, does anyone have any stats for how often memory errors happen inside the module vs. in transit?
My impression has been that the vast majority are the former, and those are pretty well fixed by on-die ECC.
But I will say I'm annoyed that there's no way to get datapath ECC without buying 25% extra chips and dealing with worse prices and speeds. Why can't a tiny chip on the module calculate those 8 bits on the fly?
Fake parity ram was a thing back when parity ram was also a thing. Fake ECC ram could also become a thing. It's easy enough to calculate the values on the fly.
Interesting to look at, though it would have to get really popular again to be worth it. I'll keep my fingers crossed.
And to be clear I don't want fake parity. I want it to verify the internal parity and then use those bits to calculate the data bus parity. Instead of paying 25% extra to get two layers of parity.
12% extra to go from zero parity to full parity was a much better offer than 25% extra to go from mostly-parity to mostly-double parity.
One big thing is better resistance to cosmic ray bitflips due to the on-chip error protection. (Different from ECC, all DDR5 gets it.) Hopefully rowhammer style trickery is effectively quashed now.
I asked myself the same thing when DDR4 came out. I was looking forward to skipping it, but Ryzen was able to unofficially use ECC ram so that intrigued me. DDR5 will be able to have larger sizes for ram as well if I remember correctly.
I think this is great improvement. (I own machines with intel and AMD processors). It feels like AMDs progress has lit a fire under intel to innovate. Competition seems to be back and I look forward to the progress next few years. I think the i5 not the i9 is the chip of note.
I'm using linux more and more and like open architecture PCs, so optimistic they can stay competitive with those SOC. Should be interesting how the linux scheduler deals with the different core types.
The architecture bumps to DDR5.0, and PCIe5.0 hopefully will help performance in the future too.
I predict it will be many years before linux works in a satisfactory way out of the box on these machines. The existing kernel model of task priority is not rich enough. Perhaps this will accelerate the adoption of user-space task scheduling.
This might be completely irrational, but it really annoys me when the press use 'big.little' for all heterogeneous CPUs. big.LITTLE is an ARM trademark for ARM's implementation of heterogenous designs. Intel have clearly decided heterogeneous is a marketing no-go and have gone with 'hybrid' architecture, which is fair enough. But these are absolutely not big.LITTLE CPUs.
"big.little" is like "kleenex" or "ziploc". Yes, technically they refer to a specific implementation of a thing, heterogenous CPU core size, tissues, plastic storage bags, but in reality, they were so perfectly branded that they just replaced the item name. It's not a big deal.
And I'm sure that people are going to call the Google Tensor "big.little" when it is really "big.medium.little" (it has 3 different types of ARM cores)!
No need to look that far. Mainstream mobile processors in everyone's pocket have already been using 3 types of cores for the last few years (cortex A55, A7* and X1).
Unless we get something better sounding than "heterogeneous CPU topology", I'm going to go with big.LITTLE. I know it's incorrect to apply it outside of ARM processors, but it's also not 13 syllables and it's more likely that someone will know what I'm talking about.
Disappointingly, despite the big.little design adding efficiency cores, this release focuses on performance, and is still far behind AMD (and of course Apple) on efficiency.
Yeah, but this is also about "enthusiast" CPUs, which is very much about eye popping top perf numbers, independent of efficiency, etc. We're talking sports cars, not commuter options.
It is. Depending on the consumer, these are just what the doctor ordered. Get a high quality motherboard, massive cooling solution, and eke out the maximum performance.
But this won't translate well into excellent laptop chips, so we'll continue to wait and see if Intel has something ready in time for their next mobile release.
And personally, I don't get enough benefit from the last few percentage points of performance to give up a quiet, cool-running, relatively power-sipping desktop system. That's just me, but that means I prefer to see advances in efficiency regardless of the market segment!
Lol how was a top-of-the-top end desktop release supposed to translate well into excellent laptop chips?
Something about your wording here seems to imply that it's not practically guaranteed that these don't translate into laptop results
-
I honestly never get why the whole "quiet and cool" thing comes up for these top of the line CPUs.
My current personal system is near-silent and already drawing something like 450W peaks for GPU usage alone, which translates to a lot more than "a few percentage points of performance"
If you can pay, you can make just about anything short of server parts quiet and cool, and the power difference won't affect your CO2 footprint in any meaningful way. And when you're looking at $600 CPUs, generally speaking you can pay...
The whole point of big.little and having efficiency cores is to save power on mobile when you don't need all that performance.
The argument is made that in the same budget (die/power/heat) of a performance core you can fit multiple efficiency cores to help with threaded workloads, but this Intel CPU uses even more power, so wouldn't they have been better off with 100% performance cores on an "enthusiast" chip?
Intel didn't use 100% performance cores because apparently they felt they could get more performance from the little cores (which are literally smaller, you can fit more in a given area on the die)
Shouldn't that be obvious, it's practically a tautology
The anandtech benchmarks show that they have better performance in single-core benchmarks (that run on the performance cores) and Ryzen has the upper-hand in multi-threaded benchmarks (since it is all performance cores). So it seems like if the goal was to make an enthusiast chip they would have been better off stuffing their chip with their superior performance cores.
Obviously Intel wants to compete on mobile. It's the biggest market now, and they've been the worst at it. But a new platform is expensive so in the beginning you have to start with the high-margin enthusiast chips, so even if a platform is designed to have gains on mobile you have to start them at the enthusiast level. If they weren't planning on scaling this down to mobile there would be no reason for this architecture. But it doesn't look like it will actually end up any more efficient with their tech.
You don't work at Intel I presume. Even of the people who work at Intel I doubt there's any one person who could tell you why they decided on the given arrangement.
I expect this kind of baseless speculation on some more casual forums not one where most of us are aware of how insanely complex modern CPU design is.
In the end they came up with a design that is already performing top of class in the workloads that are most common, and it's unlikely we won't see improvements over time.
> Overall, the 12th Gen Intel Core CPUs are terrific across the board, providing a large performance boost over the previous 11th Gen CPUs, and in almost every single case, handily out-performed AMD's Ryzen 5000 series. Intel's lead is larger at the i5 and i7 level, but even with the Core i9 12900K, Intel consistently came out on top.
I don't know why you're acting like they gave something up here... because their synthetic benchmarks aren't up to snuff?
> Lol how was a top-of-the-top end desktop release supposed to translate well into excellent laptop chips?
See Zen 2. Zen 3. The desktop chips are very efficient. When you throw 105W at them, they compete at the very top of the desktop performance market. Previous to this Alder Lake release, they were beating much more power hungry Intel chips. Now they remain competitive at half the power.
Cut them down to 15W-45W chips, and they work great in laptops, with ample performance and excellent battery life.
If Intel has massively more efficient chip designs that they plan to use in mobile, why aren't they making use of them? They could have competitive performance without doubling the power consumption.
It's a similar story with Apple Silicon. They have chips that can do amazing things at 10W. Crank 60W through them and they are excellent performers.
If you have no choice but to consume massive amounts of power to get top performance, so be it. But given the choice...
You seem to have strong opinions about this, while also missing the past decade of desktop and laptop CPU history.
Intel based the entire Core line on the architecture they designed with laptop chips in mind (initially, Pentium M). They found that throwing more power at efficient chips works pretty well. So they switched gears from Netburst.
And while you just ignored my comments on Zen, they still apply. AMD designed Zen for efficiency, which allowed it to be excellent in high power, high performance applications, while also being excellent in low power, medium performance applications. While the chips have differences due to being used differently, the core architecture is used from 15W chips all the way up to 125W desktop chips and even 280W workstation chips.
It's not impossible that Intel designs completely different architectures for their next laptop chips than what they revealed with Alder Lake, but it's also unlikely, and the point of pairing efficiency cores with performance cores is to allow for flexibility in how they use the architecture. If Alder Lake is really just for 300W workstations, then Intel made a mistake bothering with efficiency cores.
What we didn't see today was evidence that Alder Lake is ready for mobile, because these chips are much less efficient when compared to alternatives. How can we expect this architecture to be efficient when you throw less power at it, when it's already proving to be inefficient and requiring lots of power in order to perform?
It means that they have exponentially more knobs to turn.
-
And I stand by my point of ignoring insistence on speculating about unannounced hardware for no discernible benefit?
It's also weird that this is the second comment acting like outsiders know better than Intel when it comes to what mix of cores was best
Especially since this has proven to be an excellent workstation CPU:
> Overall, the 12th Gen Intel Core CPUs are terrific across the board, providing a large performance boost over the previous 11th Gen CPUs, and in almost every single case, handily out-performed AMD's Ryzen 5000 series. Intel's lead is larger at the i5 and i7 level, but even with the Core i9 12900K, Intel consistently came out on top.
intel is competing on 10nm (now called intel 7) against TMSC's 7nm process.
>If Intel has massively more efficient chip designs that they plan to use in mobile, why aren't they making use of them? They could have competitive performance without doubling the power consumption.
I'd guess they're aimed at making competitive enthusiast class desktop chips with their existing process size from foundries they already have made many many chips from so there's much less investment, and saving any smaller process capacity for laptops chips.
“Enthusiast” CPUs seem pointless. My CPU is never above 20% in AAA games as it’s all on the GFX card now.
Good to see CPUs going through the great decoupling that software did.
IMO Steamdeck is the future of home desktops. Both my kids are into science; I’m excited to have a drone remote, sensor base station, generic pc, etc, in a high quality package versus something like Pine phone.
Valve and Apple are pushing hardware forward. Hopefully they can obsolete needing data centers of generic CPUs and tons of Byzantine software by making hardware with the best logic for a task built in, available to home users.
Just a question about gaming. I haven't really seen any good AAA games worth playing anymore. The GPUs and CPUs have great capabliities but I don't see any good games anymore that make buying the hardware worth it anymore. Most of the games that are good I am interested in don't need good hardware. Do you feel the same trend when you play games?
I don’t know whether it applies to you, and don’t even know whether it’s true, but I think that may have less to do with new games being worse than with you being older and/or having seen more games. Getting older makes people less inclined to be obsessed with games, and having seen more games decreases the chance of a new game being an outlier, and outliers attract attention.
I think this applies to other fields, too. Watch your umptieth Super Bowl, and chances are you will think back to the ‘better’ one you saw when you were young. Twenty years from now, the kids watching their first one now will say the same about this one.
I don't completely disagee, but the point is that games haven't really gotten better, gameplay hasn't improved for many games (look at Cyberpunk 2077), there are more and more HD remakes since they aren't making new good games that excite people (SC2 was never as loved as SC1 same with Diablo 2 vs 3), and graphics have improved but gameplay has not. I think Nintendo is the most consistent with good new games but thats not really relevant to PC gaming.
"Best HD upgrades" is just because the default is upscaled, older consoles like the Dreamcast had amazing HD PC upgrades too (probably since the default resolution was 640 x 480, the emulated BoTW on the Wii U was 720p). Very interesting DLC, is it just the WiiU verison or switch version?
Yeah I think that’s a side effect of knowing how the sausage is made.
I have written my own ECS loops, rendering pipelines; all naive but after that it’s optimizing to product fit, and product emotional themes are pretty copy-paste to satisfy social memes.
They still occasionally come out but it’s rare. I haven’t been happy with an AAA game aside from Prey recently (2017 as “recent”) and Cyberpunk 2077 was all hype and no substance. I think they’re running out of new interesting games (Paradox, and Arkane are still good studios though) and many games I’m intrigued by are just remakes.
Starcraft Remastered, AoE II HD, System Shock, the Halo collection for instance, the Homeworld remake didn’t even interest me since I heard it was worst in some ways with hit boxes. They also don’t need new graphics cards. It’s so different from when PC hardware upgrades and games were so much more closely coupled.
Sony has started to port prestige first party PlayStation games like Horizon Zero Dawn to pc. If you like that sort of thing it's worth checking out...
Basically all of my favorite games get worse simulation performance (FPS or UPS) the longer you play as the simulation becomes more complex (Factorio, RimWorld, Oxygen Not Included, Kerbal Space Program, Stellaris) so the performance is absolutely desired.
There are plenty of games that actually use CPU: Microsoft Flight Simulator, Factorio, Stellaris, Total War, pretty much any city simulator game, etc. Sure, your average dumb AAA action game won't, but that doesn't mean a good CPU is worthless.
Factorio isn't stressing 8 core CPUs. Stellaris can be played on a laptop. You are confirming his conclusion that they don't need a strong CPU to play.
lol! Stellaris can be played on a laptop, but try ramping the Galaxy size up to 1000 and/or increase the habitable planet multiplier. You get a couple hundred years in and the game just crawls even on nice hardware. Its not unplayable, its a strategy game, but the pace definitely slows down a lot, and space battles aren't as fun to watch.
By the same token you can play virtually any game on a cheap gaming rig. Just put all the graphics on low, run it at 720p and be happy with 20 fps.
Most games don't need the latest or greatest hardware to run well, there's a lack of good AAA games that make the value proposition of new hardware much less appealing versus the days of wanting to build a computer to play Crysis.
With different software pipelines they could run right on a GPU
It’s all state in a machine, and ML is showing us recursion + memory accomplish a lot; why all the generic structure in x86 if we can prove our substrate works just as well with better power efficiency if it’s structured specifically?
Chips aren’t concepts, they’re coupled to physics; simplify the real geometry. I think that’s what Apple is really proving with its chips, and why Intel is trying to become a foundry; they realize their culture can only extend x86 and x86 comes from another era of manufacturing.
I got into tech designing telecom hardware for mass production in the late-90 and early-00s. I just code now but still follow manufacturing, and have friends that work in fabs all over; this is just sort of a summary of the trends we see shrug emoji
Is that a realistic goal to run all on GPU? Nvidia wants ARM to make GPU/CPUs together. The idea is as intriguing as making games that are OS independent and just run bare metal by making them with ISAs. I don’t think there’s games that do that.
When testing games, CPU reviews tend to test reduced resolutions and quality settings with the highest-end GPU they have, as a means of highlighting the differences between the CPUs.
While there aren't any nefarious intentions on behalf of the reviewer, this approach runs into the following problems:
- People buying high-end GPUs are unlikely to be running at resolutions of 1080p or below (or at lower quality settings), and won't see as much (if any) performance difference between CPUs as what reviewers show.
- People buying lower-end GPUs are going to be GPU-bottlenecked, and won't see as much (if any) performance difference between CPUs as what reviewers show.
- Each frame being rendered needs to be set up, animated, sent to the GPU for display, etc., and like all workloads, there's going to be portions that can't be effectively parallelized. As such, the higher the frame rate, the more likely the game is to be bottlenecked by single-threaded performance, which is an area where Intel CPUs have traditionally been strong relative to AMD's. However, as frames get more complex and take longer to render, the CPU has more of an opportunity to perform that work in parallel, and raw computational throughput is an area where AMD's modern CPUs have been strong relative to Intel's. So just because a CPU has leading performance in games today, doesn't necessarily mean that will hold in the future as game worlds become more complex (and reviewers revisiting the performance of 2017-era AMD Zen 1 vs. Intel Kaby Lake in recently-released titles have already started seeing this).
In short, the way that reviewers test CPU performance in games results in the tests being artificial and not really reflective of what most end users would actually experience.
After all, a graph showing nearly identical CPU performance across the lineup and the reviewer concluding, "yep, still GPU-limited," doesn't make for an interesting article/video.
I don't think it is possible for OSS projects to utilize this big.little system: I tried to look at the instruction level details of the thread director but could find very little. If that's the case, the future of linux desktop/laptop is even dimmer than before.
Android has downstream (partially upstream?) EAS scheduler to handle this kind of architecture [1], but I'm not sure there is any move to take this up to the intel platform. As a linux laptop user, I hope someone does something.
Linux has supported big.LITTLE style systems for a long time, and has been improving scheduler support for them for about 10 years.
(The main gap you tend to see in the Linux world is integration further up the stack. For example, allowing app developers to nominate a core preference by configuration. That's partly due to the assumption of the distribution model.)
Would love to know more. Would you mind dropping some starting point? My viewpoint is heavily biased to Android so it's great if I can learn what more standard Linux is doing. Thanks!
Eneargy Aware Scheduler is from ARM/Android.
I wasn't aware of Capacity Aware Scheduler, but it's also from ARM. I'm glad to see these being upstreamed. Thanks for the pointer!
Intel does a surprisingly fantastic job of open-sourcing things. They're one of the biggest contributors to the Linux kernel, and the official drivers for their integrated GPUs are in the mainline kernel. I wouldn't be surprised if we soon see patches submitted from Intel adding Alder Lake support to Linux's Capacity Aware Scheduling [1], or if there already have been patches submitted under the radar.
EDIT: Oh, and Linux support for their NICs is fantastic as well in my experience, both wired and wireless. That's something Broadcom can learn from - I end up replacing Broadcom wireless NICs in my laptops with Intel ones because it's such a pain to work with Broadcom on Linux.
Isn’t AMD’s graphics also in mainline? I’m very happy with using Intel support on Linux, the only reason I had seeked Broadcom was for the wireless injection intel couldn’t do.
They finally have some (nonstandard) Wayland support, after the standards were set and they had every chance to make changes, and its shitty that their open source linux drivers are locked at a low clock unless you have an old GPU. Horrible Linux experience, Linus needs to flip them off again. Its not bad with the propietary drivers, and its unfortunately the only way to run a CUDA for ML.
As a Linux desktop user I am still a bit disappointed about AMD. My Ryzen 7 does not run 100% reliably (freezes many times a year, althoug typically not more than once a month. The net is full of similar stories, but nothing seems to help). Independently but equally annoying the temperature sensors are not well supported, obviously AMD does not share documentation or contribute to the kernel. Don't remember the details now. Gave up frustrated 2 years ago or so.
YMMV... I bought a 5950X in late January for my main workstation (Arch) and haven't had any problems with it at all. FWIW my last Intel CPU (2700k IIRC) crashed my machine regularly whenever the iGPU fired up for some tasks (e.g. I could crash Blender in < 10 seconds...)
FYI temperature sensors are fixed in the latest Linux kernel. The only occasional bug I get is from the amdgpu driver (I use an APU) but even that seems to have been resolved.
I would highly recommend making sure you are using the latest BIOS and kernel.
Many times since the machine exists (> 3 years). But it's "only" Ubuntu HWE, not Arch or mainline.
In some phase I thought it happens mostly when there is "something big" happening on the screen. So I swapped the Nvidia GPU against an AMD one. But that did not change anything. And the last 1.5 years I have been using the machine mostly remotely over ssh and the freezes still occur occasionally.
So if it's not the CPU proper it could well be the motherboard. Memory tests have never shown anything, but of course I did not run them for 6-8 weeks what it takes at average to get the next freeze.
That sucks, you think its an ubuntu problem then? It sounds like a software issue, in my experience AMD's drivers in Linux are not as high quality as Intel's and if its not logged its hard to diagnose the issues. I wonder if debian would have the same issues.
My Threadripper runs rock solid under Linux and freezes under Windows 10... The only time it was shutting down in Linux was when the famous Enermax water pump was failing...
While this is exciting, the thing I'm most looking forward to is intel entering the dedicated GPU market next year. Arc is looking really promising and god knows nVidia and AMD need some disrupting.
Intel's Linux drivers for their integrated graphics chipsets are pretty good, at least compared to Nvidia's, so hopefully the same effort will be spent on Linux support for Arc. It would be nice to have another option than AMD on the table for Linux-compatible GPUs.
Intel and cheap performance product just doesn’t seem like something I’d combine and since they didn’t announce the prices I’m dubious it’ll be well priced, and I can only judge them from their history, like the DG1. I don’t want to hype something to a feverish pitch that isn’t realistic. I expected them to be bad and the first generation wasn’t good. The second generation I can only guess will be iterative.
> Performance is expected to land between the GeForce GTX 1650 and GTX 1650 SUPER but with raytracing capabilities.
It won’t compete very well with these 2016 GPUs aside from having ray tracing but an AMD driver with RADV got ray tracing backported so it’s going to have to be better for people to it over their old GPU (I avoided ryzen for a long time since my older CPU was stronger). The tensor cores looks interesting but if you use free online tokens to run ML models it’s not worth getting for that. For 1060 6GB/480 8GB owners (which I fall under) it’s not much different, do you have a mid range 2016 card? If you do, what’s the benefit you see? Nvidia is still my only option to run CUDA.
“ Jim Salter
We typically consider Cinebench to be the gold standard for general-purpose CPU tests—and for the first time in years, Intel trounces AMD's best offerings here.”
What nonsense look at the graph this comment is referencing. The new i9 barely edges ahead of Ryzen and they use the term trounces.
Give me a break … price per performance still clearly in AMD’s favor.
I'm American and I don't know what it means either. I assume burning a barn is bad? I expect not a lot of overlap between the audience of the article and people knowledgeable about agriculture idioms.
It's a sports colloquialism. It means 'exciting'. Someone may say "the game was a real barn burner", meaning that it was exciting and competitive. Ars is saying that Alder Lake is an excitingly competitive CPU.
my understanding is that a number of (already deluxe expensive fancy) motherboards have add-on chips to further jack up the cost i mean add thunderbolt4 support. i haven't seen any indicators that the kick ass readily available on chip 4x 40 Gbps of connectivity that intel mobile chips offer on integrated tb4 is available on desktop.
this, to me, constrains the desktop from being a good movipe partner to the great laptops available. at least usb4 mandates host-to-host networking, but all these great laptops with desktops lagging so far behind, having such lower standards, is excessively sad, to me.
atm phones are way far behind. no usb4 (with host-to-host networkingl nor tb4 are avialbable. i have a hard time imagining phones remaining market segmented out, not participating, in the good, not meeting the new bar.
The UHD 770 is in fact current-gen integrated graphics. It has the same modern features like DP 1.4a, HDMI 2.1, etc., which allow higher resolutions/refresh rates compared to older products like UHD 630. The main difference is fewer execution units compared to the Iris Xe-branded graphics.
Also, so far they've only put Iris Xe on 4-core/8-thread laptop CPUs; all their high-core-count laptop CPUs have the same reduced-execution-unit pattern as the desktop ones. To me, there seems to be some tradeoff between integrated graphics and CPU performance.
Intel is leeting the CPU stay on turbo boost forever, AMD seems to be more focused on Efficiency. But, if you only enable PBO curve optmizer, so AMD Ryzen 5950x can stay more time in hgher clocks, ADL 12900K is spanked in multi-thread. People achieves 11500 in CB20 and 31000 in CB23 consuming 235W. So no, ADL may great for Gaming but not for heavily mutli-thread apps. For me ADL is Intel's desperation in become the better CPU Maker. They are not anymore I guess.
The prices do look very compelling, though Intel motherboards do tend to be more expensive than AMD so spec out the whole system before deciding which is the better deal.
It's great to have competition, even if it's comping out a year behind the equivalent AMD systems.
"In order to get to this point, Intel had to cut down some of the features of its P-core, and improve some features on the E-core. The biggest thing that gets the cut is that Intel is losing AVX-512 support inside Alder Lake. When we say losing support, we mean that the AVX-512 is going to be physically fused off, so even if you ran the processor with the E-cores disabled at boot time, AVX-512 is still disabled."
"But it does mean that AVX-512 is probably dead for consumers."
"Intel isn’t even supporting AVX-512 with a dual-issue AVX2 mode over multiple operations - it simply won’t work on Alder Lake. If AMD’s Zen 4 processors plan to support some form of AVX-512 as has been theorized, even as dual-issue AVX2 operations, we might be in some dystopian processor environment where AMD is the only consumer processor on the market to support AVX-512."
It's odd, because in today's review, they have this to say:
> I have to say a side word about AVX-512 support, because we found it. If you’re prepared to disable the E-cores, and use specific motherboards, it works. After Intel spent time saying it was fused off, we dug into the story and found it still works for those that need it. It’s going to be interesting to hear how this feature will be discussed by Intel in future.
One thing that I genuinely appreciate about AMD is that I don't need to do any research to see what CPUs support what. They have a handful of SKUs, they're all straightforward, you spend what you're willing to and go from there. I feel like Intel requires me to do a bunch of research to see if the CPU I've picked checks all the boxes.
Then I buy my AMD and deal with their weird driver issues and throw my hands up.
>I feel like Intel requires me to do a bunch of research to see if the CPU I've picked checks all the boxes.
I will not be surprised if Intel follows the AMD example soon. Pat Gelsinger is a product person. And generally Product person understand these issues better. Compared to Marketing / Financial people who have zero understanding. ( I dont want to bash them too much but seriously I have never met a marketing / financial guy who is any good at products in the 20+ years. )
It's to allow easy e-core p-core thread migrations in the OS. They probably weren't intending the p-cores to be matched with e-cores when the RTL was being initially written for the p-cores, which would have been a few years ago.
Very good point. I guess they both cores have to have identical ISAs as they don’t know which instructions will be needed at the point at which they’re deciding which core to use.
My suspicious / theory is that Intel needed the hybrid design for Mobile, otherwise there is no way to keep up with AMD in terms of pref / watt. So they basically throw this hybrid design on desktop as test bed before launching it on Mobile.
It's probably part of their fab strategy. They want to offer x86 IP blocks to SoC customers, and it's nice to have options tuned for performance and ones tuned for size / efficiency.
If you're willing to accept inflation since the early 90s check out something like the 5600G instead of the top end models like the 5950X, or wait for the low end SKUs of Alder lake to come out too. The 5600G is a fantastic new 6 core CPU that will handle all but the most extreme multithreaded workloads with ease while staying under 100 Watts.
Pentium G6400 is available <$100, is rated at 58W TDP, and is "decent" by some measure. For example it is dramatically faster than RPi4 for almost all tasks:
Considering that I'm typing this comment on a laptop that is probably fair bit slower than that G6400, I do believe it is decent for basic desktop usage.
Right now, AMD is selling every chip they make, so they're not making much for the low end. Intel's got quad core comet lake chips around that price, but they've also not been selling low end versions of their newer stuff either.
If Alder Lake's release signals Intel 10nm (aka Intel 7) finally working for desktop chips, then we may see products addressing the low end market again.
You should probably be able to limit power to 100W though; power targeting is definitely a firmware feature on AMD, and I'd be surprised if it's not available on Intel as well. If nothing else, you can just provide 100W worth of cooling, and thermal management should throttle back the power.
Inflation is a thing that exists, so decent $100 CPUs are going to be hard to come by if you expect to run modern applications in 2021.
However, there's plenty of competitive CPUs that can operate below 100W. CPU reviews tend to focus on the top-end products which throw efficiency out the window for higher clock speeds, but lower tier SKUs target 95W, 65W, and 35W power points, while still maintaining quite reasonable performance.
While I'm not as familiar with the power management capabilities of modern Intel CPUs, AMD Ryzen CPUs can operate in an "Eco mode" where the power target is lowered, so you can optimize for power efficiency without having to go out of your way to buy a special "low-power" SKU.
It's hard to cool 307W over extended periods without your computer becoming really loud, even with a water cooling setup.
Now, if they had that kind of cooling available, they could have easily gotten a lot more performance out of the Ryzen, while the i9 is already operating at the limits of what consumer grade cooling can sustain.
There's no way that Intel processor is beating a 5950X that is also drawing 300W. At that point some Ryzen's will be operating at close to 6GHz per core.
The most interesting takeaway is that Intel apparently has no qualms selling a processor with a stock configuration that draws 300W.
Also their Ryzen numbers are beyond suspicious. On a silent configuration (peak power draw ~209W, 175W sustained, some undervolting) my Ryzen 5900X hits a Cinebench R20 score of 8835 vs their 8390. I suspect my rig is worse since I cheaped out on the RAM, but I wouldn't know because they omit information about the Ryzen test rig.
Their Ryzen numbers don't add up with other benchmarks around the internet either: The 5950X should hit around 10400, and the 5900X should hit 8600 in their stock configurations with mid-to-high-grade RAM[1]. Their scores are 400 points and 200 points lower respectively.
To get numbers as bad as theirs, I have to mismatch my FCLK and RAM intentionally. But again, we don't know what their setup is because they didn't tell us.
We however do know that they used DDR5 RAM for the Intel processor, which the current gen Ryzen don't support. Likely some performance was gained there as well, but obviously they didn't test the Intel processor with DDR4 RAM to get an idea of how much comes down to faster RAM. It's still fair to use DDR5, since the Intel processor supports it, but it's hard to say much about the Intel processor performance (development) itself without some additional information. Without this information we don't know whether we can expect the next generation of Ryzen processors with DDR5 support to immediately leave Intel in the dust again, or whether DDR4/DDR5 doesn't really matter.
[1]: https://www.guru3d.com/articles_pages/amd_ryzen_9_5900x_and_...