Intel · 128 GB HBM2e · 3276 GB/s bandwidth ·
Alibaba · 8.95B · model ctx up to 256K
Yes — the Gaudi 3 128GB (128 GB HBM2e) can run Qwen3.5-9B. At Q4 with the default 8K context it needs ≈ 7.71 GB VRAM (weights 4.92 GB + KV cache 1.79 GB + 1 GB runtime overhead), which fits within 128 GB. Decode at Q4/8K: ≈ 232.9 tok/s (estimated, batch 1). FP16 (≈ 20.69 GB) also fits, at ≈ 64.1 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
6.82 GB
Yes
232.9est.
Q4
8K default
7.71 GB
Yes
232.9est.
Q4
32K
13.08 GB
Yes
232.9est.
FP16
4K
19.79 GB
Yes
64.1est.
FP16
8K default
20.69 GB
Yes
64.1est.
FP16
32K
26.06 GB
Yes
64.1est.
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).