You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Yes, I have a unusual system when it comes to AI, but a few changes would make life much easier to a lot of similar people:
I do have an "AMD RX 7800 XT / AMD Ryzen 7 3700X" combination running Ubuntu Linux in German.
While the benchmark does run, it identifies both my CPU and GPU as "unknown", although ollama has no problem running the models in the GPU and the benchmark obviously detects the 16GB VRAM correctly when it comes to loading the models.
The usual, but wrong assumption is, that calls to "lshw" or "lscpu" automatically return their output in english. Unfortunately(?) mine are translated and return proper German. So neither the "Model name" nor the "product" entries are recognized - they are named "Modellname" and "Produkt" instead.
But there is an easy fix for all non-english Linux users: Just force the subprocess.run() to use the desired language, like
or just quickly r2 = subprocess.run(['lshw','-C','display'],capture_output=True,text=True,env={**os.environ, 'LANG': 'C'})
Next, there is a fallback(?) to utilize rocminfo, in case it is installed. This would be a much better result when identifying the precise AMD card instead of a generic "Navi 32 [Radeon RX 7700 XT / 7800 XT]". However the fallback will not get used, since lshw is not expected to fail.
It would help, if there is some hint, that this tool should be installed, when running the benchmark on an AMD GPU under Linux.
And since the rocminfo is not utilized, even if installed, there is no "there_is_amd_gpu" info either, just the "No NVIDIA GPU detected" which is just half the truth.
I patched the sysmain.py with the env and now get a useful reading. :-)
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
-
Yes, I have a unusual system when it comes to AI, but a few changes would make life much easier to a lot of similar people:
I do have an "AMD RX 7800 XT / AMD Ryzen 7 3700X" combination running Ubuntu Linux in German.
While the benchmark does run, it identifies both my CPU and GPU as "unknown", although ollama has no problem running the models in the GPU and the benchmark obviously detects the 16GB VRAM correctly when it comes to loading the models.
The usual, but wrong assumption is, that calls to "lshw" or "lscpu" automatically return their output in english. Unfortunately(?) mine are translated and return proper German. So neither the "Model name" nor the "product" entries are recognized - they are named "Modellname" and "Produkt" instead.
But there is an easy fix for all non-english Linux users: Just force the
subprocess.run()to use the desired language, likeor just quickly
r2 = subprocess.run(['lshw','-C','display'],capture_output=True,text=True,env={**os.environ, 'LANG': 'C'})Next, there is a fallback(?) to utilize
rocminfo, in case it is installed. This would be a much better result when identifying the precise AMD card instead of a generic "Navi 32 [Radeon RX 7700 XT / 7800 XT]". However the fallback will not get used, sincelshwis not expected to fail.It would help, if there is some hint, that this tool should be installed, when running the benchmark on an AMD GPU under Linux.
And since the rocminfo is not utilized, even if installed, there is no "there_is_amd_gpu" info either, just the "No NVIDIA GPU detected" which is just half the truth.
I patched the
sysmain.pywith theenvand now get a useful reading. :-)Beta Was this translation helpful? Give feedback.
All reactions