Mesa 24.1 To Raise Limit Supporting More Than Eight Vulkan GPUs Per System
Should you be running nine or more GPUs per system, the Mesa 24.1 release next quarter will raise the limit of 8 DRM devices for the Vulkan API per system to now allow a theoretical 256 GPUs per system.
The Mesa Vulkan run-time code has carried a hard-coded limit of supporting eight DRM devices... This limit is now apparently being hit for at least some niche users. Valve contractor Friedrich Vock raised the 8 GPU/DRM device limit up to now 256 per system. The actual code change is just a one-liner for adjusting the size of the array for the DRM device pointers to now allow up to 256 elements.
Vock wrote in the commit making the change:
There's no links to any public bug reports or other communications around the users affected by this that are running 9 or more GPUs in their system and wanting to make use of Vulkan support. It's possibly for some crypto-mining type setups, HPC accelerators (though sadly Vulkan isn't too common here...), continuous integration (CI) / continuous delivery (CD) type systems testing across many different GPUs models, or other niche configurations where there may be a valid use for having nine or more GPUs / Direct Rendering Manager devices potentially connected to the same system... Not to mention PCIe bandwidth limitations.
In any event this change is now in place for Mesa 24.1 to allow up to 256 Direct Rendering Manager devices per system in the Vulkan runtime code up from the current limit of eight.
The Mesa Vulkan run-time code has carried a hard-coded limit of supporting eight DRM devices... This limit is now apparently being hit for at least some niche users. Valve contractor Friedrich Vock raised the 8 GPU/DRM device limit up to now 256 per system. The actual code change is just a one-liner for adjusting the size of the array for the DRM device pointers to now allow up to 256 elements.
Vock wrote in the commit making the change:
"Some people seem to have systems with more than 8 GPUs installed at once. 256 is the maximum number of devices returned by libdrm, so using this seems like a good choice for now."
There's no links to any public bug reports or other communications around the users affected by this that are running 9 or more GPUs in their system and wanting to make use of Vulkan support. It's possibly for some crypto-mining type setups, HPC accelerators (though sadly Vulkan isn't too common here...), continuous integration (CI) / continuous delivery (CD) type systems testing across many different GPUs models, or other niche configurations where there may be a valid use for having nine or more GPUs / Direct Rendering Manager devices potentially connected to the same system... Not to mention PCIe bandwidth limitations.
In any event this change is now in place for Mesa 24.1 to allow up to 256 Direct Rendering Manager devices per system in the Vulkan runtime code up from the current limit of eight.
7 Comments