Training for a while (random time) the just stops with errors. second try to get som answers

[Erlandsson]

I have so much problems. I thought i setup a new system but this time with windows 11. completely CLEAN with just firefox installed.

i9, 64Gb of RAM, rtx4080super with latest drivers.

Installed all requrements as per instructions.
gitcloned a new kohya, set it up the same way as usual.

It installs with no errors, and browser window start. I set up training as i did a large number of times when it DID work on my win10 (i successfully trained several loras, from 15 minutes to 10 hours, no problems).

I use standard constant train and adamw or adafactor, no strange settings. tried 5 repeats, 10, 20 etc.
3 epochs, 5, 10 etc. no difference.
All these new clean installs, train for a while, sometimes 5 minutes, sometimes 2 hours, but ultimately fails with these errors in pic attached.
Cudas is working 90-100% while training.

deleting env and restart does not help.

Training while writing here, AND IT STOPPED.. same training, but got a little longer this time. 38% this time.

[Erlandsson]

Problem is solved. Everything works now (automatic1111,Forge ui, Kohya)

I found out it was INTEL CPU BOOST !!!!. I noticed that when running kohya, automatic etc, that cpu frequency spiked around 5.5-5.8Ghz (i9 1490k 3.2Ghz), and began to think it if was that somehow files read back and forth just got corrupted in that speed?
Even though NO OTHER program i run have or show any problems. Like Blender for example. But otn the other hand, it does not access drives that much)..

So i tried turning off INTEL TURBO BOOST in the bios. That helped RIGHT AWAY. No problems at all installing kohya, no more failed files, no more failed downloads etc. I have now run several 4-5 hour trains with no hickup.

Now i just need to see if i can get windows OWN cpu limiter to work , maybe i can try my way up to a higher clock where it is stable.