LLM Core uses 4.5 GB of VRAM

English AI
, ,
1

Hello everyone,

For the past few days, I’ve been finding that my Ollama + Open WebUI wasn’t working well, and that by searching a little, my inference on Ollama was in CPU/GPU mix, whereas it had been working very well for months on the GPU only.

By disabling Ollama, I was able to see that more than 4.5GB of VRAM was occupied, the fault: LLM Core.

Since QSirch, I have deleted the local models. But nothing changes.
By disabling it, the VRAM drops to 2 MB.
Would you know why?

It’s quite annoying that it nibbles at my VRAM.

QuTS hero: h6.0.3656
LLM Core: version 1.1.1
Qsirch: version 7.1.1.2
QNAP AI Core: version 3.8.1
Container Station: 3.1.4.2673

If needed, I can open a ticket, and eventually too, test with the new 6.1 firmware.

2

Thanks for the report! We’ll look into this internally. Could you share your NAS model, how much RAM is installed, and your graphics card model (including its VRAM size)? Thanks!

3

Hello,

Indeed, you’re right, sorry.

I have a TVS-Aih1688ATX with 96GB of RAM, an RTX Pro 4000 Blackwell (24GB of VRAM).

I only noticed this problem very recently, and had put it on an addition of elements in the RAG of Open WebUI, but finally, it wasn’t like that.