Intel · 128 GB HBM2e · 3276 GB/s bandwidth ·
Mistral · 24.0B · model ctx up to 128K
Yes — the Gaudi 3 128GB (128 GB HBM2e) can run Magistral-Small-2509. At Q4 with the default 8K context it needs ≈ 19.0 GB VRAM (weights 13.2 GB + KV cache 4.8 GB + 1 GB runtime overhead), which fits within 128 GB. Decode at Q4/8K: ≈ 86.9 tok/s (estimated, batch 1). FP16 (≈ 53.8 GB) also fits, at ≈ 23.9 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
16.6 GB
Yes
86.9est.
Q4
8K default
19.0 GB
Yes
86.9est.
Q4
32K
33.4 GB
Yes
86.9est.
FP16
4K
51.4 GB
Yes
23.9est.
FP16
8K default
53.8 GB
Yes
23.9est.
FP16
32K
68.2 GB
Yes
23.9est.
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).