Intel Core i5 12600K / Core i9 12900K “Alder Lake” Linux Performance
phoronix.com
phoronix.com
However, I do wonder whether they are going for the wrong thing. They are not competing against AMD's Zen 3 desktops at anything but the very high end, and their power consumption numbers are pretty insanely high. In both the laptop and server market, the thing that matters most is performance per watt rather than performance per dollar, and they aren't ahead in that metric.
My sense is that Apple's M1 and other ARM-based designs are getting to the point where they are "good enough" for the vast majority of use cases, even if they are insufficient for a few high-end applications. This is classic disruptive innovation, and we are seeing it reach its logical conclusion. It's kind of interesting that Alder Lake's dominance is due to them adopting something very similar to ARM's big.LITTLE
How so? The $289 i5-12600K (or $264 for the version without integrated graphics) looks to outperform the more expensive mid-range AMD parts in many tests. It's even at the top of the Performance-per-dollar chart on page two: https://www.phoronix.com/scan.php?page=article&item=intel-12... It also leads the performance-per-dollar average in the conclusion.
The i5-12600K also has lower average power consumption than everything but the low-end Ryzen 5600X, but it beats the power power-hungry AMD processors in most benchmarks. This is really impressive. I don't understand why everyone is convinced the Alder Lake is inefficient.
> In both the laptop and server market, the thing that matters most is performance per watt rather than performance per dollar, and they aren't ahead in that metric.
The i5-12600K is the performance-per-watt leader in many (maybe most) of these benchmarks. The "Per Watt Results" chart on page 12 is confusing, but the i5-12600K is the performance/watt winner in maybe half or more of the tests: https://www.phoronix.com/scan.php?page=article&item=intel-12...
I think the mid-range i5-12600K is the real star of this show, matching the performance of the AMD 5800X but at a significantly lower price. I think people are putting too much focus on the enthusiast-oriented i9-12900K which was designed to be a no-compromise performance chip without regard to efficiency.
If Intel manages to get a lot of I5s out in the wild, then AMD could decide to drop prices. That scenario could have an impact on negotiations with ODMs and System manufacturers too. Keep in mind that most quoted OEM prices you see for CPUs mean about as much as a quoted Invoice price on a car.
https://www.reddit.com/r/buildapcsales/comments/qn4y1h/cpu_m...
Comments implying the i9, a product that exists to gain max performance at all costs, should be put down for high power consumption. At the same time the same comments act like the rest of the range can be ignored because they're not the fastest of the bunch.
As a consumer, I love all of this competition. I have an M1 Max in my laptop, a 5950X in my homelab server, and I'll probably put this i5-12600K in my personal/gaming PC. They all have different strengths and I'm benefiting from it in every domain.
I agree, though, about all the competition, even if we don't agree on power consumption. But I think it's a use-case/personal preference thing. These are tools or toys, depending on use, and if you use a system as a tool, and a high-end CPU saves you time by doing things faster, overlooking power consumption makes sense.
Personally, I like to cut costs where possible, while also avoiding high power usage and fan noise, so I like efficient systems that do enough. The i5-12600K is enough, though it doesn't make sense in my personal situation where I could upgrade to a Ryzen 5000 chip without any other changes.
I also dislike Microsoft putting their latest and greatest CPU software in Windows 11, as I currently have no wish to change my gaming PC over to that system.
But the main thing is just that Intel has been cramming high power and heat into CPUs for quite a few years now, and I'm eager to see the other end of it, where top chips can manage what they're doing without 200W or more. We know it's possible, because AMD and Apple have efficient, performant chips.
AMD and other smaller competitors seem to have a lot of "street cred" for being value-players. Strangely, so does Apple (at least on the computer front).
for cost of increased power consumption
Read the "Per Watt Results" chart on page 12: https://www.phoronix.com/scan.php?page=article&item=intel-12...
The i5-12600K leads performance-per-watt results in most tests.
They include power results for some tests, where you can even see the i5-12600K consuming less power while outperforming most of the AMD parts: https://www.phoronix.com/scan.php?page=article&item=intel-12... This doesn't hold for every test, but the power consumption issue has been greatly exaggerated.
If power consumption is a concern, do not buy the i9-12900K. It's a no-compromise chip that disregards power efficiency for the sake of peak performance. Buy the i5-12600K, which is actually decently efficient.
259 Watts max!!
It's interesting to see the i5-12600K beating the AMD Ryzen 5800X in both performance and power consumption in the benchmark on page 2: https://www.phoronix.com/scan.php?page=article&item=intel-12... .
The i5-12600K also leads in many of the performance-per-watt tests on page 12: https://www.phoronix.com/scan.php?page=article&item=intel-12...
Also interesting to see that they used a mid-tier air cooler with 92mm fans and it was fine:
> The Alder Lake processors were being tested with a Noctua NH-U9S heatsink with dual fan configuration while using the Noctua NM-i17xx-MP78 mounting kit. The Noctua U9S was also used for cooling the other processors under test. With the i9-12900K pulling close to a 200 Watt average during this demanding benchmark, the average core temperature on air cooling was 81 degrees with a peak of 98 degrees. The Core i5 12600K saw a 63 degree average and peak of 72 degrees during this NAMD run.
That 92mm Noctua cooler is good, but it performs significantly worse than the much larger 140mm fan, dual-tower coolers like the NH-D15 that enthusiasts will likely be using. Yet it worked just fine.
The CPU consumes a lot of power, but it’s not as big of a deal as many people make it out to be. If power consumption is a concern you could get the i5 version for a lower power consumption and thermals. It still performs very well and is a good value.
The thermal feedback loops are so good that you can even put something like an AMD 5950X into a small case with low airflow and it will automatically adjust the clock rate down to a level it can sustain under load.
Lower temperatures will allow for higher frequencies for longer periods, though, so cooler is better. The potential for damage from excessive temperatures has been greatly exaggerated, though. Unless you’re planning to run the CPU at full load for a decade, it doesn’t really matter. A couple hours of gaming every day (which won’t produce temperatures seen in these synthetic 100% load benchmarks) is basically negligible.
For some perspective: Laptop CPUs routinely hit 90-100C because the fan curves are tuned for quietness. Those CPUs are built on the same processes. Some newer parts like the RAM on RTX3080 and RTX3090 have even higher thermal limits up to 110C. It’s not a big deal.
In the consumer world, how often is a CPU pegged to 100% full load across all cores like in these benchmarks? It doesn't happen while gaming. Maybe someone who does video encoding or compiling large projects will spike their CPU usage to 100% for a couple hours per day at most. If a CPU only spends 1-5% of its time at these high temperatures, it doesn't really add up to much over even 5 years. And even if the CPU dies after 5 years, it's really cheap to buy a replacement at that point.
If you're running a server at 100% load 24/7, that's 1-2 orders of magnitude more time spent at the high temperatures. If you need high reliability and the most dependable lifespan for those servers, you'd want to choose a lower clocked CPU. That's one of the reasons why server parts are clocked lower than their consumer counterparts.
Yeah that's sort of my situation, I do some large data conversion (not video but same idea) tasks that do run 24/7, sometimes up to a couple weeks long, but those tasks are infrequent and the box is idle the rest of the time. I do most of it on a Hetzner server that uses an old Intel desktop cpu. CPU has been fine all this time, but they did swap out the memory sticks for me due to suspected errors a while back.
So far it sounds like only AMD supports ECC memory for the desktop processors, and imho that is significant. I'm staying away from home PC's (other than my dinky laptop) but if I were to get one, I'd want ECC.
I haven't seen any benchmarks how immune to radiation DDR-5 to DDR-4 with ECC is but DDR-5 has a kind of parity check build in.
The writing is on the wall: the day will come when Intel will not write accelerated Linux graphics drivers for their laptop GPUs and that's the day when Linux on notebooks just dies. They will do so because Windows 11 + WSL (now with wslg) provides all the Linux most need.
Intel's 3D graphics drivers are in Mesa (Vulkan and OpenGL) and have been for more than a decade. The team at Intel (which I was a member of until I left a year ago) continues to develop and improve the drivers.
Especially now when Intel is entering the dGPU arena, there will be enormous pressure to reallocate people like you from Linux to the Windows driver team I bet.
My understanding is that Intel has very strong Linux drivers for all their hardware. For their entry in the discrete GPU market they will likely provide good Linux drivers to distance themselves from Nvidia.
That's a rather strong statement about the future of AMD as a going concern, don't you think!
You're right though, servers usually have some sort of very basic GPU for basic graphics. Sometimes from Matrox and others.
The parent comment makes no sense at all for a different reason. Intel wants to enter the top dGPU market and yes, Linux dominates the server landscape and yes, Intel probably wants to sell dGPUs to all the cloud providers and would have to continue to support Linux in that case.
Parent comment is simply FUD. It's sad it's been upvoted.
Many run completely headless and have no GPU at all, either on PCIe or in the processor.
Not this tired trope again.
Azure makes a lot of money off Linux, and Microsoft has been a pretty good open source participant the last several years. But it's still a business.
Quite frankly, I prefer Gnome to Windows, but I can be happy with a Mac too.
Microsoft trying to make Linux more accessible to it's HUGE professional community of programmers that are content with their OS driver but want/need accessible Linux access. The company has a decades-long track record of positive contributions to the programming community.
focus on i5s and i7s.
There are lots of reasons for success and failure. I would not count out two enormous chip makers full of skilled people and experience.
I'll be perfectly fine for that 5-10 years.