27-OCT-2024

1109

Finished the passthrough setup for randy, next steps are to get BTG set up w/ the 2070 I have and to run cables from the rack to my office for HDMI and DP from BTG. Then I can use looking-glass and BTG for near-native performance on the VMs.

Once these are done I’ll look towards migrating mirzakhani into this model as well; and maybe move it down to the rack as a result. Then I’ll have two hosts with plenty of GPU horsepower I can eventually MDev up.

I haven’t dug too deeply into the MDev side of things, but from the little I’ve read so far, I’m hoping I can have a setup which gangs up all these GPUs, less for the gaming side of things (if I have to pull one out of the gang to run whatever, so be it), but for the compute side it’ll be important. A quick test with hashcat got me 13GHashes/s on the P40 alone. I can’t get it to run on the ‘CPU’ on the VM, since it’s a QEMU vCPU it doesn’t see it as a ‘real’ CPU, but I suspect 13GHashes/s is going to translate to a pretty good rate of speed on Brocard. Assuming I get about the same in candidates-tested-per-second, it’s 11Y of compute time, but I have good reason to believe it’ll be faster (since MD5 is compartively more complicated than what I’m doing with brocard) and hopefully I can clear the 2e64 space somewhat more quickly without having to scale out too much.