← sources

Monolithic ddr3 server

source: reddit (reddit:LocalLLaMA)published 2026-10-25 20:34ingested 2026-10-26 02:40

retained text

Hi all. 

Do any of you guys run a server like this for inference of big models. Given deepseek is down to 8/16b active I'm curious as to whether this could start to make sense if I was to add a gpu. This server is local for $550 USD.

Dell PowerEdge R820 4x CPU E5-4640 768GB RAM 2x 146GB HDD Server

768GB RAM EEC 16GB EACH.

4 x INTEL E5-4640 2.40GHZ - 2.80GHZ CPUS.

Thanks!

Edit: my interest in this is mostly driven by all the new projects that claim to run big moe models on consumer desktops super fast.

document fetched 2026-10-26 00:38 · canonical https://reddit.com/r/LocalLLaMA/comments/1xfd1nk/monolithic_ddr3_server