← sources

What hardware for local Deepseek and GLM flash?

source: reddit (reddit:LocalLLaMA)published 2026-11-18 10:01ingested 2026-11-19 02:39

retained text

I guess its not just me but many people here always want more hardware and better models, so...

What hardware is needed to deploy Deepseek V4.1 Flash or GLM 5.3 Flash locally?   
By deploy I mean, to be able to run multiple parallel coding agents, at least 256k context each.   
I know you can "run" any of these MoE models on anything bigger than a regular PC but usually prompt processing struggles a lot more than generation so its not usable for agents.

To put that in perspective, I have a workstation with 512GB ddr4 ram, so DS v4.1 "almost" fits at Q4. What extra GPUs would I need to actually run it at speed with multiple sessions? Anyone with some hands on experience?

Been using Qwen 3.8 FN as my daily driver for a while, but always "why not go bigger" :)

Any ideas?

document fetched 2026-11-19 00:43 · canonical https://reddit.com/r/LocalLLaMA/comments/1xzbh80/what_hardware_for_local_deepseek_and_glm_flash