Hi, I’m using an RTX5060Ti, which forces me to use Pytorch Nightly.
I successfully compiled the latest xformers from the main branch, but when I try to run
“memory_efficient_attention“, I get the following error:
NotImplementedError: … requires device with capability <= (9. 0) but your GPU has capability (12, 0) (too new)
Does xformers not support the new sm_120(RTX 50-series) architecture yet?Thanks!
Hello!
xFormers doesn’t yet support the new RTX 50‑series sm_120 architecture, which is why memory_efficient_attention fails even though your build succeeds. Current kernels only go up to sm_90, so you’ll EZPassRI login need either a community‑patched build or wait for official updates to add sm_120 support.
Hello!
Thanks for your advice.
Which RTX-series is compatibole with the current xformers.
I’ll buy it.