Forgive me if this has been asked before but I read somewhere that the Ampere architecture is designed more towards compute task vs sheer gaming compared to turing? This has something to do with how the FP32 is executed in the new architecture? Can someone please point me towards some sources for reading/info please?
As above, unsure of good sources. The base architecture itself is, of course, aiming to balance both gaming and compute tasks. That's always been true I guess, but explicitly so ever since they put tensor cores into a lot of the chips back in the last generation & started getting such big sales for deep learning etc.
It seems to do that basic goal rather well. Then its a question of what they do with it.
A100 is obviously a compute only part.
All the other parts are going to be a mixture. Its seems fair to say that A102 was designed with a non trivial focus on compute uses, with gaming as a major byproduct. A few hints there but basically if it had been designed for gaming first it'd have been a bit smaller, likely less power draw.
The huge power draw (no laptops!), there's apparently a little bit more 'raw' FP32 than it can use in games, the huge amount of vram on the 3090 etc.
The 3070 down is seemingly going into laptops as a mobile part so it might well be balanced as a much purer gaming chip. I presume the less powerful chips are just less interesting for workstations anyway.
Facebook
Twitter