Best AI Models for AMD Radeon PRO W7900 (48 GB)

Runs 27–31B and 35B MoE models well; 70B barely fits and is slow (12 tok/s).

48 GB GDDR6864 GB/s123 TFLOPS FP16295 WRDNA 3 2023

What LLMs can the PRO W7900 run?

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB
3	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 25.4 t/s	25.4 tok/s	Q4_K_M · 21.7 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 88.5 t/s	88.5 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 75.8 t/s	75.8 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB
2	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 25.4 t/s	25.4 tok/s	Q4_K_M · 21.7 GB
3	Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 73.6 t/s	73.6 tok/s	Q4_K_M · 22.0 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 88.5 t/s	88.5 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 75.8 t/s	75.8 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB
2	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 25.4 t/s	25.4 tok/s	Q4_K_M · 21.7 GB
3	Qwen3.5 27BAlibaba Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 88.5 t/s	88.5 tok/s	Q4_K_M · 16.5 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 75.8 t/s	75.8 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Math Score	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s		Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s		Q4_K_M · 17.9 GB
3	Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 73.6 t/s	73.6 tok/s		Q4_K_M · 22.0 GB
4	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 25.4 t/s	25.4 tok/s	overall	Q4_K_M · 21.7 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 75.8 t/s	75.8 tok/s		Q4_K_M · 19.3 GB

#	Model	Speed ↓	Quant · VRAM
1	Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 159 t/s	159 tok/s	Q4_K_M · 3.4 GB
2	GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 102 t/s	102 tok/s	Q4_K_M · 13.1 GB
3	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 88.5 t/s	88.5 tok/s	Q4_K_M · 16.5 GB
4	Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 81.5 t/s	81.5 tok/s	Q4_K_M · 6.6 GB
5	NVIDIA Nemotron 3.5 Lightning 30B A3B BF16NVIDIA Q4_K_M · 19.3 GB · 75.8 t/s	75.8 tok/s	Q4_K_M · 19.3 GB

#	Model	Speed	Quant · VRAM
1	Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB
2	Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB
3	Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 25.4 t/s	25.4 tok/s	Q4_K_M · 21.7 GB
4	Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 88.5 t/s	88.5 tok/s	Q4_K_M · 16.5 GB
5	JEV 27B VLautotrust Q4_K_M · 17.9 GB · 29.8 t/s	29.8 tok/s	Q4_K_M · 17.9 GB

See all 368 models on the PRO W7900 →Best quant that fits · speeds estimated on this card

About the PRO W7900

The PRO W7900 runs 27–31B and 35B MoE models well. RDNA 3 generation, released 2023.

GenerationRDNA 3released 2023

PRO W7900 strengths and weaknesses

Strengths

Runs 35B MoE models: Qwen3.6 35B A3B at 73.6 tok/s in 22 GB
Room for long context: Qwen3.8 27B at Q4 uses 17.9 GB, leaving 30.1 GB free

Weaknesses

120B MoE models don't fit: GPT OSS 120B needs 70.8 GB at Q4 vs 48 GB here
Slow prompt reading: A 4K-token prompt on Qwen3.8 27B waits 8.4 s for the first token
ROCm, not CUDA: No FP8 weights; MXFP4 only through llama.cpp, not vLLM

PRO W7900 tokens per second by model size

Output speed depends on memory bandwidth (864 GB/s). Grey rows need more than 48 GB.

9BQwen3.5 9B · Q4_K_M · 6.6 GB
81.5tok/s
14BQwen3 14B · Q4_K_M · 11.9 GB
54.6tok/s
27–31BQwen3.8 27B · Q4_K_M · 17.9 GB
29.8tok/s
35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB
73.6tok/s
70BLlama 3.3 70B Instruct · IQ4_XS · 46.4 GB tight · IQ4_XS
12.4tok/s
120B MoEGPT OSS 120B · not at 4-bit
needs 70.8 GB

Estimated speeds: Q4_K_M, 4K context, one user. How we estimate

All 1216 models on the PRO W7900

Featured only

Best quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works

LLM models compatible with AMD Radeon PRO W7900 — ranked by Score, recency and Fit
Model	Quant	VRAM	Speed	Context	Fit
FrogNano 4B 26094.7BChatFunctions Q4_K_M·159.0 t/s	Q4_K_M	3.4 GB7%	159.0 t/s	262K	S97, Runs great
MinerU2.5 Pro 2605 1.2B1.2BVision Q4_K_M·508.2 t/s	Q4_K_M	1.1 GB2%	508.2 t/s	8K	S100, Runs great
D1 Omni 600M587MVision Q4_K_M·720.0 t/s	Q4_K_M	1.2 GB2%	720.0 t/s	128K	S100, Runs great
Holo4 35B A3B35.1BVisionFunctions Q4_K_M·79.4 t/s	Q4_K_M	21.5 GB45%	79.4 t/s	262K	S87, Runs great
Mellum2.1 12B A2.5B Thinking12.1BChatFunctions Q4_K_M·116.7 t/s	Q4_K_M	7.8 GB16%	116.7 t/s	131K	S94, Runs great
Gemma 4 26B A4B IT Qat Q4 0 Unquantized25.8BVisionFunctions Q4_K_M·88.5 t/s	Q4_K_M	16.5 GB34%	88.5 t/s	262K	S89, Runs great
WeVisOne VL 2B2.4BVisionFunctions Q4_K_M·259.2 t/s	Q4_K_M	3.6 GB8%	259.2 t/s	262K	S100, Runs great
Granite Vision 4.1 4B4.0BVisionFunctions Q4_K_M·180.6 t/s	Q4_K_M	4.0 GB8%	180.6 t/s	131K	S99, Runs great
JiRackUltra 1B1.8BChatFunctions Q4_K_M·365.1 t/s	Q4_K_M	1.8 GB4%	365.1 t/s	131K	S100, Runs great
MiniCPM V 4.6 Thinking1.3BVisionFunctions Q4_K_M·458.8 t/s	Q4_K_M	1.5 GB3%	458.8 t/s	262K	S100, Runs great
Clef Flash9.4BVisionFunctions Q4_K_M·83.5 t/s	Q4_K_M	6.5 GB14%	83.5 t/s	262K	S88, Runs great
LocateAnything 3B3.8BVisionFunctions Q4_K_M·194.2 t/s	Q4_K_M	3.2 GB7%	194.2 t/s	33K	S99, Runs great
NVIDIA Nemotron Labs Dory1.0 4B BF164.0BChatFunctions Q4_K_M·169.4 t/s	Q4_K_M	5.7 GB12%	169.4 t/s	262K	S98, Runs great
Humanizer12.0BChat Q4_K_M·63.0 t/s	Q4_K_M	9.0 GB19%	63.0 t/s	262K	A84, Runs well
North Mini Code 1.030.5BChatCodeFunctions Q4_K_M·72.9 t/s	Q4_K_M	19.4 GB40%	72.9 t/s	500K	S86, Runs great

PRO W7900 specs, power and PSU

Specs

Architecture: RDNA 3
VRAM: 48 GB GDDR6
Memory bandwidth: 864.0 GB/s
FP16 (dense): 122.60 TFLOPS
Stream processors: 6,144
TDP: 295 W
Released: 2023-04-13
MSRP: $3,999

PRO W7900 FAQ

What is the largest model the PRO W7900 can run?

Qwen2.5 72B Instruct (72.7B) is the largest at 4-bit or better: at Q4_K_S it uses 46.6 GB of 48 GB, and runs at about 12.4 tok/s.

Can the PRO W7900 run Qwen3.8 27B?

Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of 48 GB, and it is the best model that runs well on this card.

Can the PRO W7900 run 120B MoE models?

No. GPT OSS 120B needs 70.8 GB at Q4, more than the 48 GB on this card.

How many tokens per second does the PRO W7900 give?

About 29.8 tok/s on Qwen3.8 27B at Q4_K_M, with the first token after 8.4 s on a 4K-token prompt.

What power supply does the PRO W7900 need?

Plan a 600 W PSU (2× the 295 W TDP).

Is 48 GB enough for coding and agents?

Yes for coding: Qwen3.8 27B, the best of the 10 coding models that fit, runs at about 30 tok/s. For agents, not yet: no agent-scored model fits in less than 96 GB.

Best coding LLMs 48 GB VRAM GPUs with 48 GB or more