This is cool, first time I’ve seen someone measure out the different ways to use DBs with intel systems, they still haven’t gotten to the best design yet though, so if anyone reading that post feels really adventurous, you can get everything you need with sub 10ms response times by persisting memory as manifolds using safetensor format. The best part is getting to skip serde, the hardest part is needing GPUs to build the structure
-- {no channel}