# Strata with multi gpu explanation — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/9120b6b503a61bb6

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-10-19 12:55 ingested 2026-10-20 02:39

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1x9w53x/strata_with_multi_gpu_explanation/)

## retained text



```
I am mostly familiar with llama.cpp but have been using QFN - UD-Q4\_K\_XL with Strata and need some help understanding how it works with multiple gpu's.

I initially started with a single gpu and got about 70 t/s. RAM use was around 80 GB.

Then I added the second and third gpu's using the --gpus all flag and got around 100 t/s and RAM was still at around 80 GB.

I expected RAM usage to come down since all 3 gpu's VRAM was filled so I don't quite understand how Strata works here. It seems like I might as well just run it with a single gpu since 70-80 t/s is more than fine. Is it using the other gpu's for context or what?

Also, is there any way to get it to run with 64 GB of RAM?

I was hoping that adding more gpu's would lower the RAM usage.

Edit: You can offload some of the experts to disk by setting a ram budget as per u/nsfnd.

Edit the strata json file and add this argument, where 50 is your ram budget in gb.

`"--resident-budget-gib", "50"
```



document fetched 2026-10-20 00:42 · canonical https://reddit.com/r/LocalLLaMA/comments/1x9w53x/strata_with_multi_gpu_explanation
