Intel · 128 GB HBM2e · 3276 GB/s bandwidth ·
Microsoft · 14.7B · model ctx up to 32K
Yes — the Gaudi 3 128GB (128 GB HBM2e) can run Phi-4-reasoning. At Q4 with the default 8K context it needs ≈ 12.03 GB VRAM (weights 8.09 GB + KV cache 2.94 GB + 1 GB runtime overhead), which fits within 128 GB. Decode at Q4/8K: ≈ 141.8 tok/s (estimated, batch 1). FP16 (≈ 33.34 GB) also fits, at ≈ 39.0 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
10.56 GB
Yes
141.8est.
Q4
8K default
12.03 GB
Yes
141.8est.
Q4
32K
20.84 GB
Yes
141.8est.
FP16
4K
31.87 GB
Yes
39.0est.
FP16
8K default
33.34 GB
Yes
39.0est.
FP16
32K
42.16 GB
Yes
39.0est.
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).