🔌
New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
// RUN — FITS
Can the NVIDIA B200 run Kosmos-2?
Yes — here's how fast and up to what context, with real measured numbers.
Kosmos-2 on the NVIDIA B200
| Quant | Fits? | Max context | Speed | Beyond context |
|---|---|---|---|---|
| FP16 | yes | 2624k | 1825-2175 t/s | ~9.1-10.9 (slow) |
| Q8 | yes | 2648k | 3650-4350 t/s | ~18.3-21.8 (slow) |
| Q4 | yes | 2659k | 6517.9-7767.9 t/s | ~32.6-38.8 (slow) |
Estimate (memory-bound). Beyond "max context" the KV cache spills to system RAM offload (depends on your system RAM) and speed drops to the "beyond" figure.
FAQ
Can the NVIDIA B200 run Kosmos-2?▶
Yes. At Q4 it generates ~6517.9-7767.9 tok/s and fits up to 2659k context; beyond that it spills to system RAM offload (depends on your system RAM) and slows to ~32.6-38.8 tok/s.
How much context fits?▶
At Q4, up to about 2659k tokens stay in fast memory; longer context spills to system RAM offload (depends on your system RAM) and throughput collapses.