Catalogue · 26 models
How much VRAM for each LLM?
Memory needed in Q4, Q8 and FP16 with an 8k-token context, and the cheapest machine that runs it. Click a model for details by context and by card.
Dense · DeepSeek2025-01
DeepSeek R1 Distill 32B33 B · dense23 GBQ439 GBQ871 GBFP16
From: 2 × RTX 3060 12 GB · €1,460
MoE · DeepSeek2025-01
DeepSeek R1 671B671 B · 37 B active424 GBQ4748 GBQ81,410 GBFP16
From: 6 × A100 80 GB · €72,000
MoE · DeepSeek2025-08
DeepSeek V3.1 671B671 B · 37 B active424 GBQ4748 GBQ81,410 GBFP16
From: 6 × A100 80 GB · €72,000
No model found.