AMD Radeon PRO W7900
48 GB · 864 GB/s · 295 W
In short
The AMD Radeon PRO W7900 has 48 GB of GDDR6 memory with 864 GB/s of bandwidth, which sets how fast it generates text. At 8k context it runs 78 of the 125 open models we track fully in memory; the most capable is DeepSeek V4 Flash at UD-Q2_K_XL, at about 38 tok/s (estimated). We have no current price for the AMD Radeon PRO W7900 in the US.
Updated . Speeds are estimates; how they are worked out.
Prices
Every retailer we track in the US, cheapest first. Buy links may earn us a commission.
No retailer in this market lists it right now.
What two of them run
Every model in the catalogue, with the best quant for this hardware. Estimates, not measurements.
| Model | Quant | Fits | Memory | Max context | Decode @8k | Prefill | Good for |
|---|---|---|---|---|---|---|---|
| DeepSeek V4 Flash DeepSeek | UD-Q2_K_XL | fits | 92.2 GB | 32k57k with q8 cache | 38tok/sest. | 671 tok/s | chat, coding |
| Qwen 3.8 Flash Next Alibaba Qwen | UD-Q3_K_XL | fits | 85.4 GB | 256k | 9tok/sest. | 345 tok/s | – |
| MiniMax M2.7 MiniMax | UD-Q2_K_XL | fits | 73.4 GB | 92k170k with q8 cache | 50tok/sest. | 1516 tok/s | chat, coding |
| Qwen 3.8 27B Alibaba Qwen | Q8_0 | fits | 29.0 GB | 256k | 16tok/sest. | 751 tok/s | chat |
| Qwen 3.5 122B-A10B Alibaba Qwen | Q5_K_M | fits | 86.9 GB | 256k | 37tok/sest. | 1986 tok/s | chat, coding, agents, long documents |
| Qwen 3.6 27B Alibaba Qwen | Q8_0 | fits | 28.6 GB | 256k | 16tok/sest. | 751 tok/s | chat |
| gpt-oss 120B OpenAI | MXFP4 | fits | 60.7 GB | 128k | 83tok/sest. | 3366 tok/s | chat, coding, agents, long documents |
| Qwen3 Coder Next Alibaba Qwen | Q8_0 | fits | 80.6 GB | 256k | 65tok/sest. | 5016 tok/s | chat, coding, agents, long documents |
| Nemotron 3 Super 120B-A12B NVIDIA | Q5_K_M | fits | 91.5 GB | 217k256k with q8 cache | 30tok/sest. | 1749 tok/s | chat, coding, long documents |
| Gemma 4 31B | Q8_0 | fits | 33.1 GB | 256k | 14tok/sest. | 655 tok/s | – |
| Mistral Small 4 119B Mistral AI | UD-Q6_K | fits | 94.1 GB | 17k28k with q8 cache | 47tok/sest. | 2595 tok/s | chat, coding, agents, long documents |
| Qwen 3.6 35B-A3B Alibaba Qwen | Q8_0 | fits | 35.9 GB | 256k | 86tok/sest. | 6432 tok/s | chat, coding, agents, long documents |
| Qwen 3.5 27B Alibaba Qwen | Q8_0 | fits | 28.6 GB | 256k | 16tok/sest. | 751 tok/s | chat |
| Gemma 4 26B-A4B | Q8_0 | fits | 26.7 GB | 256k | 63tok/sest. | 4901 tok/s | chat, coding, agents, long documents |
| GLM 4.5 Air Z.ai | Q5_K_M | fits | 80.6 GB | 82k128k with q8 cache | 24tok/sest. | 1183 tok/s | chat |
| Qwen 3.5 35B-A3B Alibaba Qwen | Q8_0 | fits | 35.9 GB | 256k | 86tok/sest. | 6432 tok/s | chat, coding, agents, long documents |
| Qwen3 Next 80B-A3B Alibaba Qwen | Q8_0 | fits | 80.6 GB | 256k | 84tok/sest. | 6305 tok/s | chat, coding, agents, long documents |
| GLM 4.7 Flash Z.ai | Q8_0 | fits | 31.4 GB | 198k | 50tok/sest. | 3600 tok/s | chat, coding, agents, long documents |
| Devstral Small 2 24B Mistral AI | Q8_0 | fits | 25.9 GB | 384k | 18tok/sest. | 846 tok/s | chat |
| Nemotron 3.5 Lightning 30B-A3B NVIDIA | UD-Q6_K_XL | fits | 34.0 GB | 256k | 83tok/sest. | 6702 tok/s | chat, coding, agents, long documents |
| gpt-oss 20B OpenAI | MXFP4 | fits | 12.8 GB | 128k | 108tok/sest. | 4656 tok/s | chat, coding, agents, long documents |
| Kimi Linear 48B-A3B Moonshot AI | Q8_0 | fits | 50.1 GB | 1024k | 86tok/sest. | 6113 tok/s | chat, coding, long documents |
| Qwen3 Coder 30B-A3B Alibaba Qwen | Q8_0 | fits | 32.3 GB | 256k | 71tok/sest. | 4653 tok/s | chat, coding, agents, long documents |
| Qwen3 32B Alibaba Qwen | Q8_0 | fits | 35.8 GB | 40k | 13tok/sest. | 579 tok/s | – |
| Gemma 4 12B | Q8_0 | fits | 13.6 GB | 256k | 36tok/sest. | 1708 tok/s | chat, coding, long documents |
| Qwen 3.5 9B Alibaba Qwen | Q8_0 | fits | 10.5 GB | 256k | 48tok/sest. | 2163 tok/s | chat, coding, long documents |
| Mistral Small 3.2 24B Mistral AI | Q8_0 | fits | 25.9 GB | 128k | 18tok/sest. | 846 tok/s | chat |
| Gemma 3 27B | Q8_0 | fits | 29.1 GB | 128k | 16tok/sest. | 761 tok/s | chat |
| DeepSeek R1 Distill Qwen 32B DeepSeek | Q8_0 | fits | 35.8 GB | 128k | 13tok/sest. | 605 tok/s | – |
| Llama 3.3 70B Meta | Q8_0 | fits | 73.7 GB | 73k128k with q8 cache | 6tok/sest. | 283 tok/s | – |
| Ministral 3 14B Mistral AI | Q8_0 | fits | 16.0 GB | 256k | 30tok/sest. | 1404 tok/s | chat, coding |
| Qwen3 14B Alibaba Qwen | Q8_0 | fits | 17.2 GB | 40k | 28tok/sest. | 1305 tok/s | chat, coding |
| DeepSeek R1 0528 Qwen3 8B DeepSeek | Q8_0 | fits | 10.6 GB | 128k | 48tok/sest. | 2283 tok/s | chat, coding, long documents |
| Qwen3 8B Alibaba Qwen | Q8_0 | fits | 10.6 GB | 40k | 48tok/sest. | 2283 tok/s | chat, coding, long documents |
| Phi-4 14B Microsoft | Q8_0 | fits | 17.4 GB | 16k | 28tok/sest. | 1313 tok/s | chat, coding |
| Qwen 3.5 4B Alibaba Qwen | Q8_0 | fits | 5.8 GB | 256k | 99tok/sest. | 4353 tok/s | chat, coding, long documents |
| Ministral 3 8B Mistral AI | Q8_0 | fits | 10.8 GB | 256k | 47tok/sest. | 2133 tok/s | chat, coding, long documents |
| Granite 4.2 8B IBM | Q8_0 | fits | 11.3 GB | 128k | 45tok/sest. | 2117 tok/s | chat, coding, agents, long documents |
| Gemma 4 E4B | Q8_0 | fits | 9.2 GB | 128k | 56tok/sest. | 2621 tok/s | chat, coding, long documents |
| LFM2.5 8B-A1B Liquid AI | Q8_0 | fits | 10.1 GB | 125k | 194tok/sest. | 15296 tok/s | chat, coding, long documents |
| Llama 3.1 8B Meta | Q6_K | fits | 8.5 GB | 128k | 62tok/sest. | 2357 tok/s | chat, coding, long documents |
| Qwen2.5 0.5B Alibaba Qwen | Q8_0 | fits | 1.9 GB | 32k | 752tok/sest. | 32013 tok/s | chat, coding, long documents |
| Qwen3.5 0.8B Alibaba Qwen | Q8_0 | fits | 2.2 GB | 256k | 510tok/sest. | 22025 tok/s | chat, coding, agents, long documents |
| Qwen3.5 2B Alibaba Qwen | Q5_K_M | fits | 2.8 GB | 256k | 306tok/sest. | 9035 tok/s | chat, coding, agents, long documents |
| DeepSeek R1 Distill Qwen 1.5B DeepSeek | Q8_0 | fits | 3.3 GB | 128k | 223tok/sest. | 10076 tok/s | chat, coding, long documents |
| MiniCPM5 2B openbmb | Q8_0 | fits | 4.2 GB | 128k | 157tok/sest. | 6660 tok/s | chat, coding, agents, long documents |
| LFM2.5 2.6B Liquid AI | Q8_0 | fits | 4.5 GB | 128k | 141tok/sest. | 6704 tok/s | chat, coding, agents, long documents |
| LFM2.5 VL 3B Liquid AI | Q8_0 | fits | 4.5 GB | 32k | 141tok/sest. | 5916 tok/s | chat, coding, agents, long documents |
| Spark X2.5 4B XHToken | mlx-4bit | fits | 4.5 GB | 1024k | 140tok/sest. | 4739 tok/s | chat, coding, agents, long documents |
| granite vision 4.1 4b IBM | Q8_0 | fits | 5.3 GB | 128k | 111tok/sest. | 4437 tok/s | chat, coding, long documents |
| Ministral 3 3B Instruct 2512 Mistral AI | Q8_0 | fits | 5.6 GB | 256k | 105tok/sest. | 4544 tok/s | chat, coding, agents, long documents |
| granite 4.2 3b IBM | Q8_0 | fits | 5.6 GB | 128k | 104tok/sest. | 4769 tok/s | chat, coding, agents, long documents |
| LFM2.5 24B A2B Liquid AI | Q8_0 | fits | 25.7 GB | 128k | 102tok/sest. | 8032 tok/s | chat, coding, agents, long documents |
| Mellum2.1 12B A2.5B Thinking JetBrains | Q8_0 | fits | 13.5 GB | 128k | 101tok/sest. | 7370 tok/s | chat, coding, agents, long documents |
| FrogNano 4B 2609 Microsoft | Q8_0 | fits | 5.9 GB | 256k | 96tok/sest. | 4353 tok/s | chat, coding, agents, long documents |
| Kimi VL A3B Thinking 2506 Moonshot AI | Q8_0 | fits | 17.4 GB | 128k | 85tok/sest. | 6214 tok/s | chat, coding, long documents |
| Kimi VL A3B Moonshot AI | mlx-8bit | fits | 18.2 GB | 128k | 81tok/sest. | 6214 tok/s | chat, coding, agents, long documents |
| DeepSeek R1 Distill Qwen 7B DeepSeek | Q8_0 | fits | 9.3 GB | 128k | 56tok/sest. | 2543 tok/s | chat, coding, long documents |
| GLM 4 9B 0414 Z.ai | mlx-5bit | fits | 9.3 GB | 32k | 55tok/sest. | 1997 tok/s | chat, coding, agents, long documents |
| DeepSeek R1 Distill Llama 8B DeepSeek | Q8_0 | fits | 10.3 GB | 128k | 49tok/sest. | 2357 tok/s | chat, coding, long documents |
| Ministral 3 8B Reasoning 2512 Mistral AI | Q8_0 | fits | 10.8 GB | 256k | 47tok/sest. | 2133 tok/s | chat, coding, agents, long documents |
| GLM 4.6V Flash Z.ai | Q8_0 | fits | 11.0 GB | 128k | 46tok/sest. | 1844 tok/s | chat, coding, agents, long documents |
| GLM 4.1V 9B Thinking Z.ai | mlx-8bit | fits | 12.6 GB | 64k | 39tok/sest. | 1844 tok/s | chat, coding, long documents |
| Ministral 3 14B Reasoning 2512 Mistral AI | Q8_0 | fits | 16.0 GB | 256k | 30tok/sest. | 1404 tok/s | chat, coding |
| GLM 4.5V Z.ai | Q4_K_M | fits | 62.0 GB | 64k | 29tok/sest. | 1134 tok/s | chat |
| DeepSeek R1 Distill Qwen 14B DeepSeek | Q8_0 | fits | 17.5 GB | 128k | 27tok/sest. | 1278 tok/s | chat, coding |
| GLM 4.6V Z.ai | Q4_K_M | fits | 68.4 GB | 128k | 26tok/sest. | 1134 tok/s | chat |
| Qwen3.8 27B OBLITERATED OBLITERATUS | mlx-4bit | fits | 21.2 GB | 256k | 22tok/sest. | 751 tok/s | chat, coding, long documents |
| Devstral Small 2507 Mistral AI | Q8_0 | fits | 25.9 GB | 128k | 18tok/sest. | 861 tok/s | chat |
| Magistral Small 2506 Mistral AI | Q8_0 | fits | 25.9 GB | 40k | 18tok/sest. | 861 tok/s | chat |
| Magistral Small 2507 Mistral AI | Q8_0 | fits | 25.9 GB | 40k | 18tok/sest. | 861 tok/s | chat |
| Magistral Small 2509 Mistral AI | Q8_0 | fits | 25.9 GB | 128k | 18tok/sest. | 846 tok/s | chat |
| Muse Glimmer 30B meta-models | Q8_0 | fits | 31.6 GB | 128k | 15tok/sest. | 695 tok/s | – |
| granite 4.2 30b IBM | Q8_0 | fits | 32.3 GB | 128k | 14tok/sest. | 683 tok/s | – |
| DeepSeek R1 Distill Llama 70B DeepSeek | Q8_0 | fits | 73.7 GB | 73k128k with q8 cache | 6tok/sest. | 283 tok/s | – |
| Kimi Dev 72B Moonshot AI | Q8_0 | fits | 75.8 GB | 66k123k with q8 cache | 6tok/sest. | 275 tok/s | – |
| Devstral 2 123B Instruct 2512 Mistral AI | Q5_K_M | fits | 86.3 GB | 31k57k with q8 cache | 5tok/sest. | 160 tok/s | – |
| Mistral Medium 3.5 128B Mistral AI | Q5_K_M | fits | 86.3 GB | 31k57k with q8 cache | 5tok/sest. | 157 tok/s | – |
| GLM 5.3 Flash Z.ai | UD-Q2_K_XL | offload | 102.8 GB | – | 20tok/sest. | 649 tok/s | chat |
| MiniMax M2.5 MiniMax | MXFP4 | offload | 118.6 GB | – | 14tok/sest. | 1327 tok/s | – |
| DeepSeek R1 DeepSeek | Q4_K_M | won't run | 378.5 GB | – | – | – | – |
| DeepSeek R1 0528 DeepSeek | Q4_K_M | won't run | 379.0 GB | – | – | – | – |
| DeepSeek R1 Zero DeepSeek | Q4_K_M | won't run | 378.5 GB | – | – | – | – |
| DeepSeek V3 DeepSeek | Q4_K_M | won't run | 378.5 GB | – | – | – | – |
| DeepSeek V3 0324 DeepSeek | Q4_K_M | won't run | 378.5 GB | – | – | – | – |
| DeepSeek V3.1 DeepSeek | Q4_K_M | won't run | 379.4 GB | – | – | – | – |
| DeepSeek V3.1 Terminus DeepSeek | Q4_K_M | won't run | 379.4 GB | – | – | – | – |
| DeepSeek V3.2 DeepSeek | Q4_K_M | won't run | 379.4 GB | – | – | – | – |
| DeepSeek V4 Flash 0731 DeepSeek | MXFP4 | won't run | 147.7 GB | – | – | – | – |
| DeepSeek V4 Flash Vision Exp DeepSeek | MXFP4 | won't run | 156.4 GB | – | – | – | – |
| GLM 4.5 Z.ai | Q4_K_M | won't run | 205.8 GB | – | – | – | – |
| GLM 4.6 Z.ai | Q4_K_M | won't run | 205.0 GB | – | – | – | – |
| GLM 4.7 Z.ai | Q4_K_M | won't run | 205.8 GB | – | – | – | – |
| GLM 5 Z.ai | MXFP4 | won't run | 384.4 GB | – | – | – | – |
| GLM 5.1 Z.ai | MXFP4 | won't run | 422.0 GB | – | – | – | – |
| GLM 5.2 Z.ai | UD-Q4_K_M | won't run | 435.9 GB | – | – | – | – |
| GLM 5.3 Z.ai | Q8_0 | won't run | 748.4 GB | – | – | – | – |
| Kimi K2 Moonshot AI | Q4_K_M | won't run | 580.0 GB | – | – | – | – |
| Kimi K2 Instruct 0905 Moonshot AI | Q4_K_M | won't run | 580.5 GB | – | – | – | – |
| Kimi K2 Thinking Moonshot AI | Q4_K_M | won't run | 580.5 GB | – | – | – | – |
| Kimi K2.5 Moonshot AI | Q4_K_M | won't run | 580.5 GB | – | – | – | – |
| MiniMax M2 MiniMax | Q4_K_M | won't run | 132.1 GB | – | – | – | – |
| MiniMax M2.1 MiniMax | Q4_K_M | won't run | 132.1 GB | – | – | – | – |
| MiniMax M3 MiniMax | MXFP4 | won't run | 242.0 GB | – | – | – | – |
Specifications
From the manufacturer's published specifications; sources are linked where we have them.
- GPU
- Radeon PRO W7900
- Memory
- 48 GB GDDR6
- Memory bus
- 384-bit
- Memory bandwidth
- 864 GB/s
- FP16 compute
- 122.6 TFLOPS
- Software
- ROCm
- Board power
- 295 W
- PCIe generation
- Gen 4
- PCIe lanes
- x16
Compare with
Questions
- How much memory does the AMD Radeon PRO W7900 have for AI models?
- The AMD Radeon PRO W7900 has 48 GB of GDDR6 memory with 864 GB/s of bandwidth, which sets how fast it generates text.
- What is the best model the AMD Radeon PRO W7900 can run locally?
- DeepSeek V4 Flash (UD-Q2_K_XL) fits fully in its memory at 8k context and generates at about 38 tok/s, an estimate from its memory bandwidth. 78 of the 125 models we track fit at that context.
- Can the AMD Radeon PRO W7900 run GLM 5.3 Flash?
- Not fully in memory: at UD-Q2_K_XL and 8k context GLM 5.3 Flash needs about 103 GB, so part of it runs from system RAM and it slows to about 20 tok/s (estimated).
- How much does the AMD Radeon PRO W7900 cost in the US?
- We have no current price for the AMD Radeon PRO W7900 in the US.
Speeds on this page are for 8k context. See the best hardware for local AI at every budget.