Can MI300X 192GB run Qwen3-Next-80B-A3B-Instruct?

AMD · 192 GB HBM3 · 5300 GB/s bandwidth · Alibaba · 81.3B (active 3B) · MoE · model ctx up to 256K

Yes — the MI300X 192GB (192 GB HBM3) can run Qwen3-Next-80B-A3B-Instruct. At Q4 with the default 8K context it needs ≈ 46.32 GB VRAM (weights 44.72 GB + KV cache 0.6 GB + 1 GB runtime overhead), which fits within 192 GB. Decode at Q4/8K: ≈ 1124.2 tok/s (estimated, batch 1). FP16 (≈ 164.2 GB) also fits, at ≈ 309.2 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	46.02 GB	Yes	1124.2 est.
Q4	8K default	46.32 GB	Yes	1124.2 est.
Q4	32K	48.12 GB	Yes	1124.2 est.
FP16	4K	163.9 GB	Yes	309.2 est.
FP16	8K default	164.2 GB	Yes	309.2 est.
FP16	32K	166.0 GB	Yes	309.2 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	44.72 GB	0.6 GB	1 GB	46.32 GB	192 GB
FP16	162.6 GB	0.6 GB	1 GB	164.2 GB	192 GB

Other GPUs that run Qwen3-Next-80B-A3B-Instruct

• B200 192GB SXM — ≈1697.0 tok/s Q4 est.
• H200 141GB SXM5 — ≈1018.2 tok/s Q4 est.
• H100 80GB SXM5 — ≈710.6 tok/s Q4 est.
• Gaudi 3 128GB — ≈694.9 tok/s Q4 est.
• A100 80GB SXM4 — ≈432.5 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — ≈173.7 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — ≈169.7 tok/s Q4 est.
• MacBook Pro M4 Max 128GB — ≈115.8 tok/s Q4 est.

All GPUs for Qwen3-Next-80B-A3B-Instruct →

Other models for the MI300X 192GB

• Qwen3-0.6B — ≈4215.9 tok/s Q4 est.
• gemma-3-1b-it — ≈3372.7 tok/s Q4 est.
• Llama-3.2-1B-Instruct — ≈2810.6 tok/s Q4 est.
• Qwen3-1.7B — ≈1686.4 tok/s Q4 est.
• Qwen3-30B-A3B — ≈1124.2 tok/s Q4 est.
• Qwen3-Coder-30B-A3B-Instruct — ≈1124.2 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.

All models on the MI300X 192GB →