Persistent GPU Issues and Video Cuts During LocalColabFold Execution

[marcarletti]

Hello!
I've been running localcolabfold on my PC (OS: Ubuntu 22.04, CUDA V11.8.89, NVIDIA driver 525) with a GeForce RTX 3060 12GB (Asus) graphics card, processing fasta files with up to 1000 sequences using the following prompt:

colabfold_batch --num-recycle 5 --msa-mode mmseqs2_uniref --pair-mode unpaired_paired --model-type alphafold2_ptm input.fasta output_folder/

I've experienced sudden video cuts and the fans going to full power. This issue occurred multiple times, and eventually, my PC stopped detecting the GPU. I then switched to a GeForce RTX 3060 12GB (EVGA) graphics card. In the first run of localcolabfold with 168 sequences using the prompt:

colabfold_batch --amber --num-relax 1 --use-gpu-relax --num-recycle 6 --model-type alphafold2_ptm input.a3m output_folder/

After 6 hours of normal operation, I experienced another cut. Both graphics cards were tested for temperature and fan speed stability while running the program, and they remained steady at normal levels (approximately 65ºC). Additionally, the technical service performed stress tests and benchmarks on the latest GPU, and it passed without issues. I wanted to ask if anyone has experienced a similar problem or if you have any ideas about where the issue might be.
Thank you in advance (: