Which models are you using with Metis? Or want to use.

r1-1776:70B
codellama
codegemma

want to use ...

Would it be possible to run larger LLMs, such as Qwen/Qwen3-30B-A3B-Instruct-2507, Qwen/Qwen3-VL-32B-Instruct, or mistralai/Ministral-3-14B-Instruct-2512, on the larger PCIe boards (with 4 quad-core Metis AIPUs)? The memory and raw compute performance seem to be on paar with regular consumer GPUs, so I suppose it should run. Am I missing something?

​@akreal what Consumer GPU is in the Same Format and has 64GB vram <— is this the Part what you missing?

Sorry, I meant that the specs are at least as large as on the GPUs that could run those models.

Is it correct that the 4 quad-core Metis AIPUs PCIe board can run a LLM of such size (e.g. 30B parameters)?

I for the Moment Can run a 70B Model on a minisforum AI X1 Pro within 128GB ram and this Model will fit in the 64GB Metis Card if it will compiled for the Metis Card 

in other words: the 16GB Metis Card is for Video Applications and the 64GB Metis Card is for SLM/LLM

Perfect, thank you for your answer!

56GB Space on vram or ram is needed for the 70B Model 

the Parameter has nothing to do what size is the Model