FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

Intel's quest to move from "trusted by default / the reference" to "check for scam" is getting worse every release. And it's 100% self inflicted. How weird.

Re: FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

#6

Post by bede » Fri, Jul 11, 2025, 11:15 AM UTC

KomoD wrote: post #2

actual link: https://github.com/triton-lang/triton/pull/7298

Thank you, perhaps the parent can be edited to use this URL instead

Re: FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

#7

Post by brightmood » Fri, Jul 11, 2025, 11:15 AM UTC

koakuma-chan wrote: post #4

is 100 tflops a lot?

yea

Re: FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

#8

Post by hofrogs » Fri, Jul 11, 2025, 11:15 AM UTC

orlp wrote: post #3

GenuineIntel moment.

I'm interested in that story, what are you referring to with "GenuineIntel"?

Re: FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

#9

Post by orlp » Fri, Jul 11, 2025, 11:17 AM UTC

hofrogs wrote: post #8

orlp wrote: post #3

GenuineIntel moment.

I'm interested in that story, what are you referring to with "GenuineIntel"?

Intel's C++ compiler is known to add branches in its generated code checking if the CPU is "GenuineIntel" and if not use a worse routine: https://en.wikipedia.org/wiki/Intel_C%2B%2B_Compiler#Support....

Re: FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

#10

Post by hvenev » Fri, Jul 11, 2025, 11:19 AM UTC

In `libnvidia-nvvm.so` the string `cutlass` appears right after `Memory Dependence Analysis` and `memdep`. Perhaps it acts as an optimization attribute of some sort, where the compiler is allowed to make assumptions about the kernel's behavior that are not valid in general?