Untitled topic

Post by snklt » Thu, Nov 20, 2025, 10:44 PM UTC

charleshong wrote: post #9

gfhsad wrote: post #7

Earlier quoted context omitted.

Whenever I see '17x faster than experts,' I read 'the experts didn't actually try very hard on the baseline.'

Well, most of our results are not 17x. But still (IMO) solid across the board! Also, the 17x came from a pretty obscure fusion optimization that isn't called out anywhere in the documentation (we had to run the profiler to see what was actually going on). Wouldn't be surprised if whoever within AWS wrote the kernel didn't know about that optimization.

Post by incin1234 » Thu, Nov 20, 2025, 10:53 PM UTC

tomhow wrote: post #55

[stub]

Post by melissapan » Thu, Nov 20, 2025, 10:54 PM UTC

tomhow wrote: post #55

[stub]

Post by maltese669 » Thu, Nov 20, 2025, 11:04 PM UTC

tomhow wrote: post #55

[stub]

Post by yrh » Thu, Nov 20, 2025, 11:23 PM UTC

tomhow wrote: post #55

[stub]