Can MI300X 192GB run DeepSeek-R1-Distill-Llama-70B?

AMD · 192 GB HBM3 · 5300 GB/s bandwidth · DeepSeek · 70.6B · model ctx up to 128K

Yes — the MI300X 192GB (192 GB HBM3) can run DeepSeek-R1-Distill-Llama-70B. At Q4 with the default 8K context it needs ≈ 53.95 GB VRAM (weights 38.83 GB + KV cache 14.12 GB + 1 GB runtime overhead), which fits within 192 GB. Decode at Q4/8K: ≈ 47.8 tok/s (estimated, batch 1). FP16 (≈ 156.32 GB) also fits, at ≈ 13.1 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	46.89 GB	Yes	47.8 est.
Q4	8K default	53.95 GB	Yes	47.8 est.
Q4	32K	96.31 GB	Yes	47.8 est.
FP16	4K	149.26 GB	Yes	13.1 est.
FP16	8K default	156.32 GB	Yes	13.1 est.
FP16	32K	198.68 GB	No	35.1 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	38.83 GB	14.12 GB	1 GB	53.95 GB	192 GB
FP16	141.2 GB	14.12 GB	1 GB	156.32 GB	192 GB

Other GPUs that run DeepSeek-R1-Distill-Llama-70B

• B200 192GB SXM — ≈72.1 tok/s Q4 est.
• H200 141GB SXM5 — ≈43.3 tok/s Q4 est.
• H100 80GB SXM5 — ≈30.2 tok/s Q4 est.
• Gaudi 3 128GB — ≈29.5 tok/s Q4 est.
• A100 80GB SXM4 — ≈18.4 tok/s Q4 est.
• Mac Studio M3 Ultra 96GB — ≈7.4 tok/s Q4 est.
• Mac Studio M2 Ultra 192GB — ≈7.2 tok/s Q4 est.
• MacBook Pro M4 Max 128GB — ≈4.9 tok/s Q4 est.

All GPUs for DeepSeek-R1-Distill-Llama-70B →

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 →