Could anyone with a RTX 3080 please benchmark Tensorflow, Pytorch, Keras etc??
Highly needed. Thanks!
Sorta off-topic but smart move by nvidia to give the 3080 only 10gig of vram and no nvlink, so serious ML people will be tempted to the get the 3090 to enjoy the 24gigs of ram and nvlink.
Nvlink doesnt pool memory and isn’t needed for ML. You can leverage more than one card without it.
nvlink does slightly improve stuff, i think i’ve seen the benefits on puget/lambda system benchmarks. But nvidia does definitely make a big deal about nvlink and nvswitch on their dgx systems.
This topic was automatically closed 273 days after the last reply. New replies are no longer allowed.