NVIDIA · 48 GB GDDR6 · 768 GB/s bandwidth ·
Mistral · 7.2B · model ctx up to 32K
Yes — the RTX A6000 48GB (48 GB GDDR6) can run Mistral-7B-Instruct-v0.3. At Q4 with the default 8K context it needs ≈ 6.4 GB VRAM (weights 3.96 GB + KV cache 1.44 GB + 1 GB runtime overhead), which fits within 48 GB. Decode at Q4/8K: ≈ 67.9 tok/s (estimated, batch 1). FP16 (≈ 16.84 GB) also fits, at ≈ 18.7 tok/s (estimated). Estimates come from memory-bandwidth math; rows tagged measured override estimates. Relative ranking is more reliable than absolute tok/s. Methodology.
Fit & speed by quant and context
Quant
Context
VRAM needed
Fits?
tok/s (decode)
Q4
4K
5.68 GB
Yes
67.9est.
Q4
8K default
6.4 GB
Yes
67.9est.
Q4
32K
10.72 GB
Yes
67.9est.
FP16
4K
16.12 GB
Yes
18.7est.
FP16
8K default
16.84 GB
Yes
18.7est.
FP16
32K
21.16 GB
Yes
18.7est.
Fits = weights + KV(ctx) + 1 GB overhead ≤ 95% of VRAM. Measured
anchors are context-agnostic; the fit verdict is recomputed per context. A missing tok/s
means the model is far beyond this card (offload-only territory).