Page 1 of 1

How to exceed 10 gpu limit per rig

Posted: Wed Apr 01, 2020 12:44 pm
by Mustung
Hi all,

through gpu servers i would be able to fold on more than 10 gpus. Since it is a known bug/limitation that the folding software will only read the first digit of gpu index, we can not fold on the 11th gpu as the gpu index "11" would be read as "1".

Is there any way/hack to exceed this limit? I know change the limitation of gpu index is not of priority and meaning less for most people, so i guess a work around would be a better idea.

Container? Virtual machine gpu passthroug?

Re: How to exceed 10 gpu limit per rig

Posted: Wed Apr 01, 2020 1:44 pm
by ipkh
VM passthrough might work.
The real fix would require a client change and the extreme edge case of 10 or more gpus in a system means that's unlikely to happen.

Re: How to exceed 10 gpu limit per rig

Posted: Wed Apr 01, 2020 2:27 pm
by Mustung
nah dont think dev would have time for this. I believe it has been there for a while on github.

how bad is VM passthroguh performance wise? Less than 10% should acceptable. Not a linux expert here, but does dokcer has a better change maintaining gpu performance?