medium10 pts · first answer countsgpu-computinghardware

Memory bandwidth by GPU generation

Decode-heavy inference is memory-bandwidth-bound. About how much more HBM bandwidth does an H100 SXM have than an A100 80GB SXM?

A.About 10× (20 vs 2 TB/s)B.About 1.6× (3.35 vs about 2.0 TB/s)C.The same, about 2 TB/s eachD.About 1.1× (2.2 vs 2.0 TB/s)

Discussion 0

Join free to post in the discussion.