Recently, I’ve run into a Tesla V100 PCIe 32GB card due to university activities so the first thing I tried to do is run it in my desktop system. I ordered a blower-style cooler off amazon (model was the Evercool Fox-2 SB-F2) and duct-taped it to the inlet on the V100, to try and keep it cool. My system specs are as follows:
- Ryzen 7 1700x w/ Wraith Max Cooler
- Gigabyte B450 Aorus M
- 32GB Corsair Vengeance DDR4 3200Mhz CL16
- EVGA Supernova 1000 P2
- 1TB Inland M.2 NVMe SSD
- NZXT H500
- Zotac GTX 1660 gaming
- Tesla V100 PCIe 32GB??? + Evercool Fox-2
When I first tried just plugging in the Tesla, my computer wouldn’t even post (debug LED was stuck on VGA). I fixed this by removing the GPU, enabling the “Above 4G Decoding” option in the BIOS, setting the default display output to PCIe Slot 2 (slot 1 has the Tesla, slot 2 has the 1660). My problem then was that windows would bluescreen constantly with the GPU installed with error message:
SYSTEM THREAD EXCEPTION NOT HANDLED
What Failed: nvlddmkm.sys
So I tried reinstalling windows with no network connection. The system did not crash until I tried to install the latest Tesla Drivers for windows from a USB stick - midway though the installation I would have the same error. So I rebooted into safe mode and fully installed the driver, which worked fine. The device was recognized under display adapters, but I couldn’t get any info about it because windows was in safe mode. Once I rebooted into windows without safe mode though, it would instantly blue screen with the same error message. I do not get the error when I boot into safe mode though.
Does anyone know how to fix this? Do I need to use windows server instead? I’ve considered switching to linux but I still play games and don’t want to deal with that hassle, nor do I want to have to take out the Tesla everytime I boot into windows.
The ultimate goal would be to have the Tesla as my primary GPU and only use the 1660 as a display out, theoretically both running) games and CUDA programs (I’m involved in some GPU-based research). Is this even possible, or if it isn’t possible would I be able to run games on my 1660 and CUDA programs on the Tesla? Or is it not possible to run the Tesla at all on my system?