Took 42 hours to quantize on 4xA40s, at a batch size of 128. I could've went higher, but hindsight.
At that batch size, it was using about 25-30 GiB per GPU, utilization remained at 100%.