Finding optimal NVME setup

I’m trying to find the optimal NVME configuration for my plotting rig (20C 40T, 128Gb RAM). I’m thinking that I have enough CPU and RAM to run at least 20 plots in parallel. I read somewhere that a single plot requires around 350MBps sustained write throughput in the temp directory, is this about right?

It seems that I would need 7,000MBps of sustained write throughput to run 20 plots in parallel on one machine?

I was curious about sustained write speed of NVME drives you can actually buy these days, so I tested “Sabrent 1TB Rocket NVMe PCIe M.2 2280 Internal SSD High Performance Solid State Drive (SB-ROCKET-1TB)” from Amazon. It appears to have a small but very fast cache. Once the cache gets full, the sustained write throughput is around 985MBps.

Test wrote 100Gb to the drive with

dd if=/dev/zero of=/dev/nvme0n1 bs=1M count=100000

and measured with iostat.

With that throughput, looks like this drive will only be able to support 3 parallel plots before it becomes a bottleneck. The 1Tb limit only allows 3 parallel plots, so this particular drive is balanced in terms of capacity and sustained write performance. I suspect that the 2Tb version of this drive will have more capacity, but the same sustained write throughput, so it will become a bottleneck if I try more to use it for more than 3 plots in parallel.