Intel · 128 GB HBM2e · 3276 GB/s bandwidth ·
Alibaba · 79.7B (active 3B) · MoE · model ctx up to 256K
Yes — the Gaudi 3 128GB (128 GB HBM2e) can run Qwen3-Coder-Next. At Q4 with the default 8K context it needs ≈ 45.44 GB VRAM (weights 43.84 GB + KV cache 0.6 GB + 1 GB runtime overhead), which fits within 128 GB. Decode at Q4/8K: ≈ 694.9 tok/s (estimated, batch 1). FP16 needs ≈ 161.0 GB — does not fit on this card. 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
45.14 GB
Yes
694.9est.
Q4
8K default
45.44 GB
Yes
694.9est.
Q4
32K
47.24 GB
Yes
694.9est.
FP16
4K
160.7 GB
No
346.4est.
FP16
8K default
161.0 GB
No
345.1est.
FP16
32K
162.8 GB
No
337.5est.
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).