Linux 6.11 Features Many Exciting Updates For AMD Hardware & More

[phoronix]

Phoronix: Linux 6.11 Features Many Exciting Updates For AMD Hardware & More

It's expected to be the Linux 6.11 release day! We are just hours away from hopefully seeing Linux 6.11 stable christened as the kernel set to power the likes of Ubuntu 24.10 and Fedora 41. Here's a reminder of some of the most interesting new features and changes to look forward to with Linux 6.11...

https://www.phoronix.com/news/Linux-6.11-Feature-Recap

[Kjell]

AMDGPU users playing Cyberpunk 2077 might be stuck with 6.9 since the new revision of Clear Page Tracking in 6.11* introduces a high latency penalty thus lag spikes in certain games

This should be resolved with new DRM Buddy patches in the future

[Keith Myers]

And still no fix for the missing RAPL power monitoring for Zen 5 processors. Come on . . . a one line fix and no developer or maintainer could have merged that into the 6.11 kernel by now.

[perpetually high]

AMDGPU users playing Cyberpunk 2077 might be stuck with 6.9 since the new revision of Clear Page Tracking in 6.10+ introduces a high latency penalty thus lag spikes in certain games

This should be resolved with new DRM Buddy patches in the future

This is fixed btw in the official release of 6.11

$ sed -n '351,352p' drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
/* always clear VRAM */
flags |= AMDGPU_GEM_CREATE_VRAM_CLEARED;

[Kjell]

This is fixed btw in the official release of 6.11

$ sed -n '351,352p' drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
/* always clear VRAM */
flags |= AMDGPU_GEM_CREATE_VRAM_CLEARED;

Not quite

This issue started in 5.19 with drm/ttm patches which intended to resolve screen corruption when clearing memory by ordering the flushes correctly, as explained by the dev here:

The original commit actually fixed a rather bad bug. The stuttering is just the immediate effect of having the ordering now correctly done. So we absolutely can't revert that in any way possible.

the bug was that the kernel driver didn't fully blocked out access to freshly allocated memory which is still cleared by the hardware.
So what could potentially happen without that patch is screen corruption where parts of the rendering are just overwritten by black.

memset() is by far the absolute worst option.

This introduced stuttering in games which saturate VRAM bandwidth like Cyberpunk 2077 and was only partially mitigated by orginal Clear Page Tracking + DRM Buddy patches added to 6.9-rc4. This left us with minor latency issues/lag spikes when clearing the memory.

Down the road V2 revision of these patches were introduced in 6.10.2 which unfortunately regressed throughput/FPS with certain games like Dota 2

The FPS regression was mitigated in 6.11 by AMDGPU_GEM_CREATE_VRAM_CLEARED (which you've mentioned) at the cost of introducing higher latency that requires additional patches to DRM Buddy

[Kjell]

This is fixed btw in the official release of 6.11

I posted a fully detailed explanation with sources but Phoronix anti spam detection didn't like that (it might surface in a few hours)

In short, the changes in 6.11 fixed the FPS regression at cost of increasing latency meaning there's now more lag spikes in Cyberpunk 2077 (check links in my comment above).

v2: always set AMDGPU_GEM_CREATE_VRAM_CLEARED (Christian)

This adds allocation latency, but aligns better with user expectations. The latency should improve with the drm buddy clearing patches that Arun has been working on. In addition this fixes the high CPU spikes seen when doing wipe on release.

[justinkb]

For those on zen 4 hardware, is there any action that needs to be taken to take advantage of the new stuff or is it all just "enabled" by default?

[Keith Myers]

For those on zen 4 hardware, is there any action that needs to be taken to take advantage of the new stuff or is it all just "enabled" by default?

No, nothing needs to be done for Zen 4 or Zen 5. Everything is enabled by default.

[LinuxNoob]

Finally...

- ASUS EC Sensors driver support for the AMD-based ProArt X670E CREATOR WIFI motherboard.

[Keith Myers]

Also discovered that the SPD temp sensors in DDR5 DIMMs are reporting now.