Can MI300X 192GB run gemma-4-26B-A4B-it?

AMD · 192 GB HBM3 · 5300 GB/s bandwidth · Google · 25.2B (active 3.8B) · MoE · model ctx up to 256K

Yes — the MI300X 192GB (192 GB HBM3) can run gemma-4-26B-A4B-it. At Q4 with the default 8K context it needs ≈ 15.62 GB VRAM (weights 13.86 GB + KV cache 0.76 GB + 1 GB runtime overhead), which fits within 192 GB. Decode at Q4/8K: ≈ 887.6 tok/s (estimated, batch 1). FP16 (≈ 52.16 GB) also fits, at ≈ 244.1 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	15.24 GB	Yes	887.6 est.
Q4	8K default	15.62 GB	Yes	887.6 est.
Q4	32K	17.9 GB	Yes	887.6 est.
FP16	4K	51.78 GB	Yes	244.1 est.
FP16	8K default	52.16 GB	Yes	244.1 est.
FP16	32K	54.44 GB	Yes	244.1 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	MI300X 192GB VRAM
Q4	13.86 GB	0.76 GB	1 GB	15.62 GB	192 GB
FP16	50.4 GB	0.76 GB	1 GB	52.16 GB	192 GB

Other GPUs that run gemma-4-26B-A4B-it

• B200 192GB SXM — ≈1339.7 tok/s Q4 est.
• H200 141GB SXM5 — ≈803.8 tok/s Q4 est.
• H100 80GB SXM5 — ≈561.0 tok/s Q4 est.
• Gaudi 3 128GB — ≈548.6 tok/s Q4 est.
• A100 80GB SXM4 — ≈341.5 tok/s Q4 est.
• RTX 5090 32GB — ≈300.1 tok/s Q4 est.
• RTX 4090 24GB — ≈168.8 tok/s Q4 est.
• RTX 4090D 24GB (China) — ≈168.8 tok/s Q4 est.

All GPUs for gemma-4-26B-A4B-it →

Other models for the MI300X 192GB

• MiniCPM5-1B — ≈3122.9 tok/s Q4 est.
• Qwen3.5-2B — ≈1686.4 tok/s Q4 est.
• GLM-4.7-Flash — ≈1124.2 tok/s Q4 est.
• Qwen3-30B-A3B-Instruct-2507 — ≈1124.2 tok/s Q4 est.
• Qwen3-Next-80B-A3B-Instruct — ≈1124.2 tok/s Q4 est.
• Qwen3-Coder-Next — ≈1124.2 tok/s Q4 est.
• Qwen3.5-35B-A3B — ≈1124.2 tok/s Q4 est.
• Qwen3.6-35B-A3B — ≈1124.2 tok/s Q4 est.

All models on the MI300X 192GB →