Can Gaudi 3 128GB run gemma-3-27b-it?

Intel · 128 GB HBM2e · 3276 GB/s bandwidth · google · 27.4B · model ctx up to NoneK

ONEAPI

Yes — the Gaudi 3 128GB (128 GB HBM2e) can run gemma-3-27b-it. At Q4 with the default 8K context it needs ≈ 21.55 GB VRAM (weights 15.07 GB + KV cache 5.48 GB + 1 GB runtime overhead), which fits within 128 GB. Decode at Q4/8K: ≈ 76.1 tok/s (estimated, batch 1). FP16 (≈ 61.28 GB) also fits, at ≈ 20.9 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	18.81 GB	Yes	76.1 ± est.
Q4	8K default	21.55 GB	Yes	76.1 ± est.
Q4	32K	37.99 GB	Yes	76.1 ± est.
FP16	4K	58.54 GB	Yes	20.9 ± est.
FP16	8K default	61.28 GB	Yes	20.9 ± est.
FP16	32K	77.72 GB	Yes	20.9 ± est.

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).

VRAM breakdown at 8K context

Quant	Weights	KV cache	Overhead	Total needed	Gaudi 3 128GB VRAM
Q4	15.07 GB	5.48 GB	1 GB	21.55 GB	128 GB
FP16	54.8 GB	5.48 GB	1 GB	61.28 GB	128 GB

Other GPUs that run gemma-3-27b-it

• B200 192GB SXM — ≈185.8 tok/s Q4 est.
• MI300X 192GB — ≈123.1 tok/s Q4 est.
• H200 141GB SXM5 — ≈111.5 tok/s Q4 est.
• H100 80GB SXM5 — ≈77.8 tok/s Q4 est.
• A100 80GB SXM4 — ≈47.4 tok/s Q4 est.
• RTX 5090 32GB — ≈41.6 tok/s Q4 est.
• RTX 4090 24GB — ≈23.4 tok/s Q4 est.
• RTX 4090D 24GB (China) — ≈23.4 tok/s Q4 est.

All GPUs for gemma-3-27b-it →

Other models for the Gaudi 3 128GB

• MiniCPM5-1B — ≈1930.3 tok/s Q4 est.
• Qwen3.5-2B — ≈1042.4 tok/s Q4 est.
• DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx — ≈906.4 tok/s Q4 est.
• Qwen3.6-35B-A3B — ≈694.9 tok/s Q4 est.
• Qwen3.5-35B-A3B — ≈694.9 tok/s Q4 est.
• GLM-4.7-Flash — ≈694.9 tok/s Q4 est.
• Qwen3-30B-A3B-Instruct-2507 — ≈694.9 tok/s Q4 est.
• Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 — ≈694.9 tok/s Q4 est.

All models on the Gaudi 3 128GB →