Windows 11 Better Than Linux Right Now For Intel Alder Lake Performance

[mphuZ]

Stay tuned for such article next week, already found some interesting discoveries on that front

offtop: Why is there still no news about AMDGPU-Pro?

https://www.amd.com/en/support/kb/release-notes/rn-amdgpu-unified-linux-21-40-1

[bridgman]

offtop: Why is there still no news about AMDGPU-Pro?
https://www.amd.com/en/support/kb/re...-linux-21-40-1

I don't think we have told anyone about it yet...

But yeah, there's this new 21.40 and it's the next step in unifying ROCm and AMDGPU-PRO stacks. In theory at least we now have the same ROCm code in 4.5 and 21.40 and that common code has gone through CL+GL testing on real world applications.

What I don't know yet is whether our graphics stack testing on apps like Blender and Resolve is sufficient to catch the kind of bugs that our users have been seeing, but at least we have the stack unification and "partially integrated" testing in place now so hopefully we will be able to close any remaining gaps more quickly.

Note - the installation procedure has changed significantly from previous AMDGPU-PRO releases, so people won't think you're a wuss if you read the instructions. I'm not sure the new procedure is necessarily better than the old one, but that wasn't the goal for this step.

[F.Ultra]

That is because the Microsoft Development Center gets early dies of new CPU's from Intel on a regular basis. And why would Red Hat care about a desktop CPU since that is not the core market for RHEL?

Unless an OEM (ie: Lenovo) did it themselves and submitted the patches or requested it from the OS vendor like Canonical, no one would do much until Michael shows it to everyone.

High chance that there will be Xeons based on the Alderlake architecture at some point and RHEL does have a user base among workstations and not only servers.

[bug77]

I don't disagree with all of your points, but I can't imagine buying Alder Lake and not going DDR5. It just seems like a really bonehead move. I went with AMD's X570S Gigabyte AORUS Master. $379 or so. Paid a lot, but I know the motherboard is important to me. The Intel version of that was $750+

I don't disagree that Zen had its own growing pains, but I'm talking about where everything stands in Q4 2021. I didn't *have* to buy the 5800X, but I know for $299, that's a buy that's going to last me many years. And on top of that, I have an upgrade path with the new Zen4 that comes out.

I have a Haswell, so I'm not an Intel hater. Their CPU has been going strong, 7+ years after I bought it, and I did plenty of overclocking. So I'm glad Intel is making a strong comeback, which as a result, also allowed me to get a comparable (or better) 5800X for an outstanding price.

On the contrary, I see no reason to go DDR5, especially since benchmarks show improvements only in select workloads.
And yes, if you have to buy right now, AMD wins on the motherboard front. But if you can wait a little, you won't have to pay the early adopter tax. And if you can wait a little longer, you'll probably be able to get roughly the same features (apparently sans PCIe5) on H670.

I am actually in that boat. Like you, I also believe there's no point in buying good components only to stick them in a cheap motherboard. So while my CPU is on its way, I actually have no idea when I'll buy the motherboard. I'm still looking for the right board (I need debug LEDs, QFlash and whatnot). If I can't find a decent deal, I am willing to wait for H670.

[agd5f]

I suspect windows CPU freq governor does a better job than the current ones on Linux. At least on x86 CPUs, it's probably better to let the platform handle it autonomously rather than trying to constantly determine the best frequency and walking through a continuous range of frequencies every time it checks. This probably also explains the poor performance behaviour a lot of people have seen with the AMD pstate driver (and possibly the intel pstate driver as well). With CPPC, you have a continuous scale of frequencies while with the old ACPI pstate interface you only had three, so you tend not to fall into continuously changing trying to pick the best frequency.

[Monsterovich]

On the contrary, I see no reason to go DDR5, especially since benchmarks show improvements only in select workloads.

+7% performance
+100% price

[Linuxxx]

I suspect windows CPU freq governor does a better job than the current ones on Linux. At least on x86 CPUs, it's probably better to let the platform handle it autonomously rather than trying to constantly determine the best frequency and walking through a continuous range of frequencies every time it checks. This probably also explains the poor performance behaviour a lot of people have seen with the AMD pstate driver (and possibly the intel pstate driver as well). With CPPC, you have a continuous scale of frequencies while with the old ACPI pstate interface you only had three, so you tend not to fall into continuously changing trying to pick the best frequency.

Honestly, comments like this make me wonder why there apparently isn't a proper communication channel inside AMD between different divisions?

I mean, why then will the Steam Deck ship with the AMD p-state CPU driver + schedutil?
You say that the decision-making logic should be left up to the hardware to decide, while schedutil proponents argue that the hardware can't possibly have a clue about OS run-time queues of all the different threads interacting with each other.

So, which one is it?
And who has the final say on what is best for the upcoming Steam Deck?

[Mario Junior]

Butbutbut… Windows is supposed to be the slow one. This is not fair!!!

Aside from the "muuhh spy" issue, it's the same thing or WORSE for the Linux side.

The only desktop system that is managing to save itself yet (I don't know how long) is MacOS.

[rabcor]

So the alder lake drivers for linux are gimped? good to know.

[sophisticles]

Rotflmao!