you can write pretty performant code based on your textbook knowledge and accumulated wisdom...but writing high-performance code has always, and will i think always, be an empirical process.

you *have* to measure

you *have* to experiment

you *have* to tinker

you *have* to remember making sense isn't a prerequisite

compilers are weird. hardware is fallible. developers are myopic.

#Coding #GPULife #hpc #scientificcomputing

fp8 is here for #machinelearning use, with a collab spec from nvidia, arm, and intel.

Maybe now I can push for fp16 in more of the cisTEM codebase since it isn't the new kid on the block anymore : )

#GPULife #CryoEM

https://developer.nvidia.com/blog/nvidia-arm-and-intel-publish-fp8-specification-for-standardization-as-an-interchange-format-for-ai/