News for hackers. Stuff that matters.

A comment on Fp8 runs ~100 tflops faster when the kernel name has "cutlass" in it. Parent comment

Re:Fp8 runs ~100 tflops faster when the kernel name has "cut... by surajrmal on Friday October 03, 2025 @10:57AM (#45463712)
Sadly it's built into the vulkan protocol. Even a fully userspace driver arrangement with a microkernel ends up giving the driver access to the client's information. Of course it's forgeable the way it's done though so you could opt out if you really wanted to.

[1]: https://github.com/KhronosGroup/Vulkan-Headers/blob/main/inc...

[ Parent | on HN ]
Re:Fp8 runs ~100 tflops faster when the kernel name has "cut... by doubletwoyou on Friday October 03, 2025 @11:54AM (#45464362)
I mean Khronos put that in for a reason. If the drivers didn't get explicit information about the application being run, they would do silly heuristics like quake3 to squeeze out performance.
[ Parent | on HN ]