Ubuntu + Radeon VII? AI/ML benchmarks?
Has anyone tried the Radeon VII under Linux, specifically Ubuntu? If so, and it works, would you mind running a quick benchmark for me? I was thinking of getting a RTX 2080ti or maybe even an RTX titan for the vram, but the RVII looks like a nice compromise on price vs performance.
1. install plaidML (google it), but running the following should work: pip install plaidml-keras plaidbench
2. go to the bottom of the GitHub and
copy the email example code into a txt file with the extension .py (ie, filename.py).
3. In the 3rd from the bottom line, change “range(10)” to “range(100)”.
4. Save the file.
5. Run the file by typing “python filename.py”
6. Copy results here!
I’m happy to help if my instructions left anything out, were unclear, or confusing.
I wouldn't derive anything from Radeon VII desktop benchmarks for eGPU. I remember seeing great desktop scores for Radeon VII for Shadow of the Tomb Raider. In my eGPU setup they were...meeeh.. There are more cases like this.
I get that there are issues with the VII. That's why I specifically want to see how it performs in an eGPU setting under linux for compute tasks. Maybe it's fine!
Raw computing numbers (gpgpu) are very good. Especially the FP64 performance of Radeon VII is amazing. Scientific calculations profit most of this, standard software doesn't.
@OliverB Have you tried installing the CUDA and cuDNN libraries in Mojave using the kepler cards?
I had CUDA installed on my older macbook. After the update to Mojave, they didn't work anymore. I haven't tried to solve this yet. I never tried cuDNN. I think it is safe to say that they won't work unless explcitely said.