Question: Current hardware recommendation to run MMseqs2 for LocalColabFold& Co

[dirrei]

Hi,
we started to exceed the number of requests the public mmseqs2 server is suitable for and would like to set up our own internal server. I was now wondering about the current hardware recommendation? The databases need to be loaded into memory so minimum 1TB RAM , but is 1TB sufficient, or should it be 1.5TB or even 2TB? Also the number of cores are 64 sufficient or better 128? We only have a small user base, but some users tend to send 10k plus sequences daily. Running colabfold and co in a two step procedure is not really an option as some users lack the knowledge and others don't like the inconvenience of doing so. Therefore, we would like to run a small mmseqs2 server on premise.
Thank you in advance for your input!

[scschneider44]

Feel free to wait for the official recommendation, but we saw success with a 1TB machine with 64vCPUs. It may be necessary to try different setups to figure out what will work best for you.