Radeon PRO W7900
48 GB GDDR6 ECC · 864 GB/s · 295 W
Radeon PRO W7900 48 GB · 64 GB RAM → Llama 3.3 70B Instruct
GPU · AMD
Radeon PRO W7900 for local LLMs: which models run?
With 48 GB of VRAM and 864 GB/s of bandwidth, the Radeon PRO W7900 runs models up to about Llama 3.3 70B Instruct entirely on the GPU.
- VRAM48 GB GDDR6 ECC
- Bandwidth864 GB/s
- Board power295 W
- ArchitectureRDNA 3
Specifications
- VRAM
- 48 GB
- Memory type
- GDDR6 ECC
- Memory bus
- 384-bit
- Bandwidth
- 864 GB/s
- Board power
- 295 W
- Backends
- ROCm, Vulkan
- Interface
- PCIe 4.0 x16
- Released
- April 2023
- Segment
- Workstation
- FP16 compute
- Not available
Official specData updated 24 days agoSource: AMD specifications
What the Radeon PRO W7900 runs
| Mixtral 8x7B InstructQ6_K · 8K~39.2 GB · 34–52 tok/s estimate | Excellent |
|---|---|
| Qwen3 32BQ8_0 · 8K~37.1 GB · 16–21 tok/s estimate | Excellent |
| DeepSeek R1 Distill Qwen 32BQ8_0 · 8K~37.1 GB · 16–21 tok/s estimate | Excellent |
| Qwen2.5 32B InstructQ8_0 · 8K~37.1 GB · 16–21 tok/s estimate | Excellent |
| Qwen2.5 Coder 32B InstructQ8_0 · 8K~37.1 GB · 16–21 tok/s estimate | Excellent |
| QwQ 32BQ8_0 · 8K~37.1 GB · 16–21 tok/s estimate | Excellent |
| Qwen3 30B-A3BQ8_0 · 8K~33.4 GB · 62–106 tok/s estimate | Excellent |
| Qwen3 Coder 30B-A3B InstructQ8_0 · 8K~33.4 GB · 62–106 tok/s estimate | Excellent |
| Gemma 3 27B ITQ8_0 · 8K~30.3 GB · 19–26 tok/s estimate | Excellent |
| Gemma 2 27B ITQ8_0 · 8K~31.7 GB · 19–26 tok/s estimate | Excellent |
| Mistral Small 3.2 24B InstructQ8_0 · 8K~26.7 GB · 22–30 tok/s estimate | Excellent |
| gpt-oss-20bMXFP4 · 8K~13 GB · 131–181 tok/s estimate | Excellent |
| Qwen3 14BQ8_0 · 8K~17.6 GB · 33–46 tok/s estimate | Excellent |
| DeepSeek R1 Distill Qwen 14BQ8_0 · 8K~17.8 GB · 31–45 tok/s estimate | Excellent |
| Qwen2.5 14B InstructQ8_0 · 8K~17.8 GB · 31–45 tok/s estimate | Excellent |
| Phi-4 14BQ8_0 · 8K~17.7 GB · 32–46 tok/s estimate | Excellent |
| Mistral Nemo 12B InstructQ8_0 · 8K~14.9 GB · 38–54 tok/s estimate | Excellent |
| Gemma 3 12B ITQ8_0 · 8K~14.2 GB · 39–56 tok/s estimate | Excellent |
| Gemma 2 9B ITQ8_0 · 8K~12.8 GB · 46–67 tok/s estimate | Excellent |
| Qwen3 8BQ8_0 · 8K~10.6 GB · 53–78 tok/s estimate | Excellent |
| Llama 3.1 8B InstructQ8_0 · 8K~10.2 GB · 55–81 tok/s estimate | Excellent |
| DeepSeek R1 Distill Llama 8BQ8_0 · 8K~10.2 GB · 55–81 tok/s estimate | Excellent |
| Qwen2.5 7B InstructQ8_0 · 8K~9.3 GB · 61–88 tok/s estimate | Excellent |
| Mistral 7B Instruct v0.3Q8_0 · 8K~9.2 GB · 60–88 tok/s estimate | Excellent |
| Gemma 3 4B ITQ8_0 · 8K~5.5 GB · 96–151 tok/s estimate | Excellent |
| Qwen3 4BQ8_0 · 8K~6.2 GB · 86–135 tok/s estimate | Excellent |
| Phi-4 Mini Instruct 3.8BQ8_0 · 8K~6 GB · 93–144 tok/s estimate | Excellent |
| Llama 3.2 3B InstructQ8_0 · 8K~5.2 GB · 109–170 tok/s estimate | Excellent |
| Qwen3 1.7BQ8_0 · 8K~3.7 GB · 152–254 tok/s estimate | Excellent |
| Llama 3.2 1B InstructQ8_0 · 8K~2.6 GB · 251–413 tok/s estimate | Excellent |
| Qwen3 0.6BQ8_0 · 8K~2.6 GB · 217–403 tok/s estimate | Excellent |
| Qwen2.5 72B InstructIQ4_XS · 8K~42.4 GB · 14–19 tok/s estimate | Compatible |
| Llama 3.3 70B InstructQ4_K_M · 8K~45.1 GB · 13–18 tok/s estimate | Compatible |
| DeepSeek R1 Distill Llama 70BQ4_K_M · 8K~45.1 GB · 13–18 tok/s estimate | Compatible |
| gpt-oss-120bMXFP4 · 8K~63.3 GB · 19–39 tok/s estimate | CPU/GPU offload |
| Llama 4 Scout 17B-16E InstructQ3_K_M · 8K~55.7 GB · 9.5–18 tok/s estimate | CPU/GPU offload |
| DeepSeek R1 (671B)Q8_0 · 8K~699 GB · No data | Incompatible |
| Qwen3 235B-A22BQ8_0 · 8K~247 GB · No data | Incompatible |
For each model, the highest-precision quantization that runs entirely in fast memory; otherwise the one with the best result. Speeds are ranges; "estimate" unless a benchmark exists for this exact hardware.
EstimateData updated 24 days agoSource: Run LLM Locally engine
Limits in plain words
48 GB of VRAM holds the weights and KV cache of models up to about Llama 3.3 70B Instruct at Q4_K_M. Bigger models spill into system RAM and slow down sharply: generation speed is then limited by RAM bandwidth, not by the GPU.
Where to buy a Radeon PRO W7900
Prices and links follow the market you choose.
No current offer imported for this market: the link opens a marketplace search.
Frequently asked questions
Which LLMs can the Radeon PRO W7900 run?
Entirely on the GPU: Qwen2.5 72B Instruct (IQ4_XS), Llama 3.3 70B Instruct (Q4_K_M), DeepSeek R1 Distill Llama 70B (Q4_K_M), Mixtral 8x7B Instruct (Q6_K), Qwen3 32B (Q8_0), DeepSeek R1 Distill Qwen 32B (Q8_0), Qwen2.5 32B Instruct (Q8_0), Qwen2.5 Coder 32B Instruct (Q8_0). Larger models need CPU/GPU offload.
Can the Radeon PRO W7900 run a 70B model?
Compatible: a 70B model at Q4_K_M needs about 45.1 GB; the Radeon PRO W7900 has 48 GB.
How is the speed estimated?
Token generation reads every active weight once per token, so it is limited by memory bandwidth (864 GB/s here). We apply an efficiency range per backend and label the result as an estimate until real benchmarks exist.