Linux Driver Developer At Valve Preps More Patches For Improving AMD GCN 1.0 GPUs
([Radeon] 58 Minutes Ago
AMDGPU + GCN 1.0 Southern Islands)
- Reference: 0001580850
- News link: https://www.phoronix.com/news/AMDGPU-More-GCN-1.0-SI
- Source link:
Thirteen years after the AMD GCN 1.0 "Southern Islands" GPUs initially launched as the Radeon HD 7000 series, recently there has been an effort to improve the support for both GCN 1.0 and the GCN 1.1 graphics processors with their open-source Linux driver stack. This recent effort has been led by one of the developers on Valve's Linux graphics team.
As noted back in July, a Valve Linux engineer has been working on big improvements for old AMD Radeon GPUs from the GCN 1.0 and GCN 1.1 generations. By default those GPUs use the older "Radeon" Linux kernel graphics driver rather than the newer "AMDGPU" kernel graphics driver used by GCN 1.2 and all newer AMD Radeon/Instinct hardware.
Using the AMDGPU kernel driver is important for RADV Vulkan, various performance optimizations, and all around better support compared to the legacy Radeon driver that rarely sees any significant improvements these days. But for making GCN 1.0/1.1 more suitable for AMDGPU, Valve contractor Timur Kristóf has been working to enhance the AMDGPU DC display support for these GPUs and other fixes/adjustments.
Some of those patches from Timur are queued up for merging to the Linux 6.18 kernel while on Friday some additional work was posted. The new patches are to address some issues Timur observed recently with AMDGPU on GCN 1.0 Southern Islands hardware. He commented on [1]the patch series :
"This series has a few minor patches to address some SI issues.
When a 4K 60Hz display is connected to Tahiti or Pitcairn there is a slight flickering near the bottom of the display. Disabling MCLK switching fixes that. (Other SI parts are likely affected too, but I didn't test them thoroughly enough to say.)
When enabling ASPM on Zen 4 with Tahiti and Oland, there are random hangs when the GPU usage is low. Disabling ASPM fixes that. At the moment I don't know if this is a platform-specific or GPU-specific issue and I don't think we can reasonably determine that without spending more time than we have. (Other SI parts may be affected, but I didn't test them for a long enough time to judge that.)
Finally, there is a DC patch to change the minimum PLL dividers to the same value as the legacy non-DC display code. This doesn't fix any visible issue but I think it's still good to have just in case." The code is now on the mailing list awaiting review.
[1] https://lists.freedesktop.org/archives/amd-gfx/2025-September/131198.html
As noted back in July, a Valve Linux engineer has been working on big improvements for old AMD Radeon GPUs from the GCN 1.0 and GCN 1.1 generations. By default those GPUs use the older "Radeon" Linux kernel graphics driver rather than the newer "AMDGPU" kernel graphics driver used by GCN 1.2 and all newer AMD Radeon/Instinct hardware.
Using the AMDGPU kernel driver is important for RADV Vulkan, various performance optimizations, and all around better support compared to the legacy Radeon driver that rarely sees any significant improvements these days. But for making GCN 1.0/1.1 more suitable for AMDGPU, Valve contractor Timur Kristóf has been working to enhance the AMDGPU DC display support for these GPUs and other fixes/adjustments.
Some of those patches from Timur are queued up for merging to the Linux 6.18 kernel while on Friday some additional work was posted. The new patches are to address some issues Timur observed recently with AMDGPU on GCN 1.0 Southern Islands hardware. He commented on [1]the patch series :
"This series has a few minor patches to address some SI issues.
When a 4K 60Hz display is connected to Tahiti or Pitcairn there is a slight flickering near the bottom of the display. Disabling MCLK switching fixes that. (Other SI parts are likely affected too, but I didn't test them thoroughly enough to say.)
When enabling ASPM on Zen 4 with Tahiti and Oland, there are random hangs when the GPU usage is low. Disabling ASPM fixes that. At the moment I don't know if this is a platform-specific or GPU-specific issue and I don't think we can reasonably determine that without spending more time than we have. (Other SI parts may be affected, but I didn't test them for a long enough time to judge that.)
Finally, there is a DC patch to change the minimum PLL dividers to the same value as the legacy non-DC display code. This doesn't fix any visible issue but I think it's still good to have just in case." The code is now on the mailing list awaiting review.
[1] https://lists.freedesktop.org/archives/amd-gfx/2025-September/131198.html
darkbasic