I’m consulting for a company, and we are currently moving away from Github hosted runners to self hosted runners. The tests we are going to be running are various, but the particular workload I’m curious about is Cypress tests. Each worker will run one web server, and cypress tests against this server and they max out at a 3 core utilization.
The workload is interesting, because it is literally about being fastest at browsing the web using chrome, and hosting webservers (with MOCK data) in parallel.
At the moment I’ve moved the workload to 2 machines with a 13500k cpu and 64GB ddr4 memory each. We have 22 jobs to do in total, and currently have 16 workers running on the two machines, so there’s room for adding more workers, but not really on the existing machines due to memory headroom, and CPU bottleneck. But since we are to acquire some new machines, I’m not sure, if I should go for a single 9950x machine or a 285k. Most likely the AMD chip is faster than the intel, but the intel has more real cores. Or maybe we should continue with two or more machines as we currently have. If we keep the workflow design pattern, we will be capable of just adding machines to the pool as the test suite grows.
Threadrippers and the like don’t really make sense from a price to performance standpoint - although would be interesting to see how fast the fastest piece of silicon can solve this task.