I’ve not seen freely available software to split such a workload across multiple machines, but realistically if I did I’d be looking at less performance that if I just got a single used datacenter card (like one of the Nvidia Tesla cards) off eBay for the same price and popped it into a computer, or if I got a single much more modern server.

I can however cluster them in fun ways for redundancy! Most hypervisors support clustering so that VMs can be migrated to another host if one needs to be taken offline for anything, or if one unexpectedly powes off the others will continue the workload. Or clustered storage where it spreads the storage across multiple hosts for redundancy as well as speed. I definitely want to get some of those old 25GB or even some of those 40GB infiniband cards and run a glusterFS or Ceph cluster to really see what clustered storage can do (I ran a Ceph cluster in a lab in college but it was over a gigabit network so everything was painfully slow)

But those projects will account for only about a dozen computers at most, so I have to find more projects and more willing people to have these systems foisted upon