Hello everybody,
This is my first post on here :)
I tried to run HPL-2.0 CUDA on six P100. I’ve compiled it and set HPL.dat / run_linpack. When I tried to run the benchmark, it didn’t want to detect any GPU (although I have checked via nvidia-smi that all GPUs presented). Did I make mistake on compiling part or anywhere else?
And also I noticed from DELL report that HPL 2.1 (or higher) is needed to maximize HPL run on Tesla Pascal / Volta. Where I can get it? Should I contact NVIDIA by myself? I’m preparing for upcoming student cluster competition and this is my very first time to work with huge GPU cluster.
Thanks so much!
This is more or less expected behavior. The HPL 2.0 that was published a long time ago is no longer suitable for use on recent GPUs (anything beyond Fermi).
NVIDIA does have a newer HPL, but it is only distributed in certain circumstances, for instance if you will submit for Top500. If you don’t plan to do something like that, we don’t provide or support HPL for general usage.
If you purchased a large enough cluster of GPUs to make it onto the Top500 list, or you are part of some other limited sets of circumstances, e.g. a team working on the Student Cluster Competition for SuperComputing show, then you will already have a way to contact NVIDIA (apart from these forums) to request assistance with HPL.
Thank you so much to reply my question.
Yes, I’m part of team preparing for Student Cluster Competition. Which email or link to contact NVIDIA for HPL request assistance? Thanks.
They will contact you, when we are ready to support those teams.