← sources

Worth buying a DGX-1 (8× V100 32GB) for local LLM inference in 2026? ~15k €

source: reddit (reddit:LocalLLaMA)published 2026-11-09 15:28ingested 2026-11-10 02:42

retained text

I found used DGX-1 locally for 15k€

8 x V100 32GB SXM2 (256GB VRAM,NVLink), dual Xeon E5-2698 v4, 512GB RAM, 4× IB EDR, 3.5 kW

  
I am curious to see if anyone has that config to tell me real world experience? Which models do you run and what is throughput?

I plan to rent similar (runpod,vast,lambda or wherever i can find) to experiment myself as well.



On a side note -> how good it is for deep learning - training CV models?

document fetched 2026-11-10 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1xrulta/worth_buying_a_dgx1_8_v100_32gb_for_local_llm