Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

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

twitter.com

Post reply on HN
Print view View original

111–114 of 114 posts

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

#111

Post by bosco_mcnasty »

MBCook wrote: post #52
jdright wrote: post #37

Earlier quoted context omitted.

yes, that is a very usual way (known practices) of vendors applying specific optimizations for known things. It is also part of the benchmarks game they play against each other.

The link is long dead and the Wayback machine doesn’t have a copy. But in 2001 ATI was caught applying optimizations to Quake 3 when someone realized if you renamed the executable from “quake” to “quack” the score dropped a ton. It was a big scandal. I know that’s common now but that wasn’t a thing that was done at the time.

the etiology of this hack is pretty obvious with a simple google search:

https://docs.nvidia.com/cutlass/index.html

it presumably makes various assumptions and speedups for NVIDIA's matrix multiplication library... called cutlass

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

#112

Post by hinkley »

withinboredom wrote: post #106
hamburglar wrote: post #104

Earlier quoted context omitted.

“Checking for escaped text” is the sort of nonsense that tells you you’re dealing with amateur developers.

Indeed. The rules are simple: - Unescape, sanitize or validate at all entry points. - Escape all outputs (this includes the database queries). If you follow those simple rules, you never have to check once you are past a controller. And you should fuzz your controllers to make sure no unexpected data makes it past there.

Thing about taking a job is they don’t generally let you look at the code first and nope out if it’s fucked six ways to Sunday.

Everyone has clever answers for greenfield projects and empty rhetoric for brown.

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

#114

Post by saagarjha »

zahlman wrote: post #108
saagarjha wrote: post #101

Earlier quoted context omitted.

I don't think you are a corporate shill. I do think that you immediately going "clearly the tweet is wrong" without doing any research whatsoever was unwarranted, though. You also keep bringing up that it's 93 commits but all getting squashed you have to do is search for "cutlass" to find out what is going on. I think you're obligated to do at least that when you call it out for being wrong.

How did you get "clearly" out of "appears to be"? How did you get "without doing any research whatsoever" out of me demonstrably following the link and reading and quoting what appeared on the facing page?

I think if you spent the amount of effort you’ve used to reply to me here to skim through the PR you would’ve changed your mind. I understand that the code requires some domain knowledge to fully understand but I think even a cursory skim would be enough to disabuse me of the idea that the tweet was cherry-picking.
Post reply on HN