Hacker News new | ask | show | jobs
by epistasis 2555 days ago
> It seems to be based on this benchmark[1] which encodes a huge repeated int32 stuffed with random numbers

However, if your entire world consists of pushing around images to GPUs/TPUs, that's pretty much the only workload you ever need to optimize from your larger ops system.

Edit: not sure that this particular use case is exactly what's going on in that benchmark, now that I look more closely. Also, I wouldn't want to assert that it's a use case that matters a ton.

2 comments

Encoding an image as a repeated sequence of variable-length integers would obviously be a poor choice.
Except.. that's pretty much how raw images are represented. Substituting "integers" for "bytes" if you want channel resolution, or leaving integers if you have 32bit pixel depth.

https://github.com/nothings/stb/blob/master/stb_image.h#L120

https://wiki.libsdl.org/SDL_Surface

You overlooking the critical term: variable-length.
is anybody talking to GPU through memcached GETs/PUTs?