Milot Mirdita

You need to drop all instances of `--max-seq-len` and replace `--target-cov 0.95` with `-c 0.95 --cov-mode 1`. Lines 112-114 are quite expensive. I think that could be done smarter now,...

This issue is fixed in Git, but not part of a release yet. I think I fixed it shortly after the last release. You can download precompiled binaries at mmseqs.com/latest....

Did you solve the issue with the clustering not being hierarchical? This should produce a hierarchical clustering. Foldseek is quite a bit more tricky since it uses different modules will...

This seems pretty puzzling, we are running on an older CPU model and nothing else seems out of order. MPI won't help and will only add additional latency. MMseqs2 can...

I took a closer look at the CPU flags. I think your KVM is disallowing sse41 and avx2. At least sse41 is required for reasonable performance.

Okay maybe that's not the issue since cmake claims to have found the various SIMD instruction sets to work: ``` -- Performing Test HAVE_AVX2_EXTENSIONS -- Performing Test HAVE_AVX2_EXTENSIONS - Success...

Exactly, run the prefilter command with: ``` perf record -g -F 250 -- mmseqs prefilter ....rest-of-arguments--- ``` This will generate a `perf.data` file that contains the perfomance profiling information. It...

Sorry I was traveling. I don't see anything out of the ordinary in your screenshot. I still don't really have an idea whats going on. I repeated the UniRef search...

I don't think its a NUMA issue either. Our system should be affected worse by NUMA issues since its a 8-socket system.

Not personally. But we have used the server on two completely differently spec-ed servers and the performance is okay both time: * 1x AMD Epyc 7713P / 2TB RAM /...