Intel Announces 12th Gen Core "Alder Lake" CPUs, Linux Tests Forthcoming

[phoronix]

Phoronix: Intel Announces 12th Gen Core "Alder Lake" CPUs, Linux Tests Forthcoming

Intel is using their inaugural Intel Innovation virtual event today to formally announce the highly-anticipated 12th Gen Core "Alder Lake" processors. These first desktop processors built on their "Intel 7" process and employ a hybrid architecture will be available in retail channels next week. Today we can talk more about Alder Lake specifications and features while our Linux performance benchmarks and support analysis will come once the Alder Lake review embargo expires next week.

https://www.phoronix.com/vr.php?view=30645

[bug77]

Based on what leaked so far, I'm pretty sure the performance is up there. But I'm curious to see the power draw.

[birdie]

Hello Michael

1. Would be nice if could run CPU specific tests after the Alder Lake CPUs settled on their actual TDP whatever it is, thus excluding the first minute or so when they run at much higher wattages (PL1/PL2, whatever)

2. Would be nice if you could pit Ryzen 5000, Rocket Lake and Alder Lake against each other when all are set to the same TDP to see what their instructions per watt are.

These would be quite unusual but very interesting benchmarks.

[Setif]

Good to see DDR5 and PCIe 5.0, but No AVX512 That's sad.
I think Intel is restricting AVX512 to HPC-like usage (ie Xeon CPUs), doing what GPU-makers did years before when they reduced fp64 capablities in their gaming GPUs and restricted them to HPC GPUs (Nvidia Tesla and AMD Instinct).

[birdie]

Good to see DDR5 and PCI 5.0, but No AVX512 That's sad.
I think Intel is restricting AVX512 to HPC-like usage (ie Xeon CPUs), doing what GPU-making did years before when they reduced fp64 capablities in their gaming GPUs and restricted them to HPC GPUs (Nvidia Tesla and AMD Instinct).

AVX512 is present on die of P cores but it's factory disabled because it would make migrating tasks between E and P cores impossible. Intel mistakenly claimed that the CPUs containing only P cores would come with AVX512 but later they clarified that no customer Alder Lake CPUs would unlock it. Maybe they want to reserve AVX512 for their enterprise customers who can pay premium for the feature.

[bug77]

Hello Michael

2. Would be nice if you could pit Ryzen 5000, Rocket Lake and Alder Lake against each other when all are set to the same TDP to see what their instructions per watt are.

These would be quite unusual but very interesting benchmarks.

Tricky feat, seeing how Alder Lake sports two IPCs...

[chimpy]

Interesting but it looks like Intel being shady again when it comes to benchmarks:


"Intel shows showing the 12900K being anywhere from slightly slower than the Ryzen 9 5950X, to being 30 percent faster. However, Intel does admit that these benchmarks were captured on Windows 11 before the performance patch for AMD CPUs was available, so the results aren’t as meaningful as they would have been had they tested the 5950X in its best performing mode, such as using Windows 10 or waiting for the patch to be available. Intel also claims that the 12900K would still be the “world’s fastest gaming CPU” if they used DDR4 memory, though they didn’t show the data to back this up.

It should also be noted that Intel tested with Windows 11’s virtualization-based security feature enabled. We recently showed in our Windows 11 vs Windows 10 benchmark test that VBS does hurt performance on the Core i9-11900K in games, though the degree to which performance is impacted depends on the game. If the 12900K is superior at managing or accelerating VBS, then the gains shown versus 11th-gen may be greater than they would have been if both platforms were tested with VBS disabled. That’s something we’ll have to explore in our review."

That's from techspot's coverage.

[ms178]

Maybe they want to reserve AVX512 for their enterprise customers who can pay premium for the feature.

While that is a plausible motive, that would be bad for a wider adoption and as AVX-512 is better suited for general-purpose use, this also would be short-sighted technically as well. I guess the lack of AVX-512 it is more likely a result of an engineering disaster and not part of a conscious product segmentation strategy. A future generation will bring AVX-512 support back, either due to a better hardware thread scheduler that actually can place these tasks on the P-cores automatically or with little cores gaining these vector capabilities. Their power cost and cost per area has to become low enough to justify their presence though with the latter solution. Hence my bet is that a future iteration improves the hardware scheduler and that the P cores will get AVX-512 enabled. As Raptor Lake is meant to come out before Zen 4, AMD would not even profit from this mishap as AVX-512 is still absent in Zen 3D and only Zen 4 is supposed to change that.

[birdie]

Interesting but it looks like Intel being shady again when it comes to benchmarks:


"Intel shows showing the 12900K being anywhere from slightly slower than the Ryzen 9 5950X, to being 30 percent faster. However, Intel does admit that these benchmarks were captured on Windows 11 before the performance patch for AMD CPUs was available, so the results aren’t as meaningful as they would have been had they tested the 5950X in its best performing mode, such as using Windows 10 or waiting for the patch to be available. Intel also claims that the 12900K would still be the “world’s fastest gaming CPU” if they used DDR4 memory, though they didn’t show the data to back this up.

It should also be noted that Intel tested with Windows 11’s virtualization-based security feature enabled. We recently showed in our Windows 11 vs Windows 10 benchmark test that VBS does hurt performance on the Core i9-11900K in games, though the degree to which performance is impacted depends on the game. If the 12900K is superior at managing or accelerating VBS, then the gains shown versus 11th-gen may be greater than they would have been if both platforms were tested with VBS disabled. That’s something we’ll have to explore in our review."

That's from techspot's coverage.

What's shady about having HW acceleration for new important security features? Or should Intel have purposely made their new CPUs slower when using VBS? After all the feature was one of the reasons Microsoft severely limited HW compatibility with older devices.

[brucethemoose]

While that is a plausible motive, that would be bad for a wider adoption and as AVX-512 is better suited for general-purpose use, this also would be short-sighted technically as well. I guess the lack of AVX-512 it is more likely a result of an engineering disaster and not part of a conscious product segmentation strategy. A future generation will bring AVX-512 support back, either due to a better hardware thread scheduler that actually can place these tasks on the P-cores automatically or with little cores gaining these vector capabilities. Their power cost and cost per area has to become low enough to justify their presence though with the latter solution. Hence my bet is that a future iteration improves the hardware scheduler and that the P cores will get AVX-512 enabled. As Raptor Lake is meant to come out before Zen 4, AMD would not even profit from this mishap as AVX-512 is still absent in Zen 3D and only Zen 4 is supposed to change that.

What Intel should have done is add AVX512 to the atom cores, even if they can't run it at full speed. AMD did something similar with Zen 1, though I've been told that AVX512 support would be a bit trickier than AVX2 on AVX hardware.

Anyway, I think any kind of widespread AVX512 adoption is pretty much doomed for the near future. At the very least, Zen 3 and Alder Lake will have to get cycled out of the market, which will take a looong time.