no just nvidia
no just nvidia
20 Replies
So I followed your advice and made a clean nvidia driver install and now it doesn't crash but it takes forever. I let over night and I'm at 5%
good
that means you are ussing too much VRAM
before starting training
which config you tried?
The 24gb config
ok then do this
increase number of block swap by 1
try
and increase by 1 and try
until you find your best speed
Ok going to try that
Thanks again for your time
Maybe should I go for the 16Gb config ?
12 gb works better than 16
but my strategy should work better
This part ?
Or this part ?
How long usually before we can see the first 1% coming ?
Ok so the farthest I can get to is 35, at 36 it crashes.
I can get around 40s/it
that is so wrong
rtx 3090 can do like 7-8 second / it
if you are gold member i can connect your pc and debug reason
No, just bronze supporter
What could make it slow like that ?
I've closed all unnecessary running app and services.
I've followed every steps of the tutorial.
Launched the downgrade pytorch.bat file
Is there a log, screenshot or command that can help figuring the issue ?
sadly nothing comes my mind. how much vram you are using before training?
Around 400Mo
it should work great in that case
400 mb is very low
What do you suggest me ?
uninstalling everything and starting on a fresh install ?
probably
maybe even format windows