AMD Radeon PRO W7900
The AMD Radeon PRO W7900 has 48 GB VRAM and 864 GB/s memory bandwidth. It can run 57 of our 94 tracked models natively in VRAM at 8k context.
With 48 GB GDDR6, the AMD Radeon PRO W7900 is a workstation-tier GPU that can run 57 models natively. It handles 70B-class models at Q4 quantization.
The AMD Radeon PRO W7900 is AMD's flagship RDNA 3 workstation GPU, delivering 48GB ECC-capable GDDR6 on a 384-bit bus at 864 GB/s with 96MB of Infinity Cache and 6,144 stream processors. It can hold 34B models at Q8_0 and 70B models at Q4_K_M in VRAM, matching the NVIDIA RTX 6000 Ada on capacity at comparable bandwidth, making it AMD's most capable single-GPU workstation inference platform. ROCm applies on Linux; use the Vulkan backend on Windows.
AMD Radeon PRO W7900: 2023 RDNA 3 flagship workstation GPU with 48GB ECC-capable GDDR6 on a 384-bit bus at 864 GB/s, 96MB Infinity Cache, 6,144 stream processors.
70B models fit at Q4_K_M; 34B models fit at Q8_0 entirely in VRAM, matching the NVIDIA RTX 6000 Ada on capacity at comparable bandwidth. ~25-35 t/s for 7B Q4.
ROCm on Linux for full acceleration; Vulkan backend on Windows. AMD's most capable single-GPU workstation option for local LLM inference.
| Vendor | AMD |
| Architecture | RDNA 3 |
| VRAM | 48 GB |
| Memory type | GDDR6 |
| Memory bandwidth | 864 GB/s |
| Compute backend | ROCM |
| Tier | Workstation |
| Released | 2023 |
| Models (native) | 57 / 94 |
| Models (offload) | 8 / 94 |
Software: ROCm is Linux-only; on Windows use the Vulkan backend instead. Requires llama.cpp compiled with ROCm support.
Popular models for this GPU
Models this GPU runs natively in VRAM (57)
- Qwen 2.5 72B Instruct72B · MMLU-Pro 71.1Q3_K_M · ~15 t/s
- Llama 3.3 70B Instruct70B · MMLU-Pro 68.9Q3_K_M · ~15.4 t/s
- DeepSeek R1 Distill Llama 70B70B · MMLU-Pro 70.0Q3_K_M · ~15.4 t/s
- Llama 3.1 70B Instruct70B · MMLU-Pro 66.4Q3_K_M · ~15.4 t/s
- Mixtral 8x7B Instruct v0.146.7B · MMLU-Pro 29.7Q6_K · ~15.4 t/s
Show 52 more
- Command-R 35B35B · MMLU-Pro 33.0Q6_K · ~14.2 t/s
- Qwen 3.5 35B-A3B (MoE)35B · MMLU-Pro 85.3Q8_0 · ~52 t/s
- Qwen 3.6 35B35B · MMLU-Pro 85.2Q8_0 · ~14.3 t/s
- Ornith 1.5 35B-A3B (MoE)35B · MMLU-Pro N/AQ8_0 · ~52 t/s
- Yi 1.5 34B Chat34.4B · MMLU-Pro 37.0Q8_0 · ~14.6 t/s
- Qwen3 32B32.8B · MMLU-Pro 65.5Q8_0 · ~15.5 t/s
- Qwen 2.5 32B Instruct32.5B · MMLU-Pro 69.0Q8_0 · ~15.3 t/s
- Qwen 2.5 Coder 32B Instruct32.5B · MMLU-Pro 62.3Q8_0 · ~15.3 t/s
- DeepSeek R1 Distill Qwen 32B32.5B · MMLU-Pro 65.0Q8_0 · ~15.3 t/s
- Nemotron 3 Nano 30B32B · MMLU-Pro 78.3Q8_0 · ~50.7 t/s
- Gemma 4 31B30.7B · MMLU-Pro 85.2Q8_0 · ~16.4 t/s
- Qwen3 30B-A3B (MoE)30B · MMLU-Pro 61.5Q8_0 · ~49.1 t/s
- Nemotron 3.5 Lightning 30B-A3B30B · MMLU-Pro 81.6Q8_0 · ~52.6 t/s
- Muse Glimmer 30B27.8B · MMLU-Pro N/AQ8_0 · ~18.9 t/s
- Gemma 2 27B Instruct27.2B · MMLU-Pro 38.0Q8_0 · ~17.5 t/s
- Gemma 3 27B Instruct27B · MMLU-Pro 67.5Q8_0 · ~18.6 t/s
- Qwen 3.6 27B27B · MMLU-Pro 86.2Q8_0 · ~19.2 t/s
- UI-Mate 27B27B · MMLU-Pro ~86.2Q8_0 · ~19.2 t/s
- Bonsai 27B27B · MMLU-Pro 81.5Ternary (Q2_0) · ~63.7 t/s
- Qwen 3.8 27B27B · MMLU-Pro N/AQ8_0 · ~19.2 t/s
- Gemma 4 26B (MoE)25.2B · MMLU-Pro 82.6Q8_0 · ~40.6 t/s
- Mistral Small 3.1 24B Instruct24B · MMLU-Pro 66.8Q8_0 · ~20.9 t/s
- Mistral Small 22B22.2B · MMLU-Pro 49.2Q8_0 · ~22 t/s
- GPT-OSS 20B21B · MMLU-Pro 67.9Q8_0 · ~43.3 t/s
- Qwen3 14B14.8B · MMLU-Pro 61.0BF16 · ~18.1 t/s
- Qwen 2.5 14B Instruct14.7B · MMLU-Pro 63.7BF16 · ~18.1 t/s
- Phi-4 14B Instruct14B · MMLU-Pro 70.4BF16 · ~19.1 t/s
- Mistral Nemo 12B Instruct12.2B · MMLU-Pro 35.6BF16 · ~21.8 t/s
- Gemma 3 12B Instruct12.2B · MMLU-Pro 60.6BF16 · ~22.1 t/s
- Gemma 4 12B (Unified)12B · MMLU-Pro 77.2BF16 · ~20.6 t/s
- Gemma 2 9B Instruct9.2B · MMLU-Pro 32.0FP32 · ~14.2 t/s
- Qwen 3.5 9B9B · MMLU-Pro 82.5FP32 · ~15.5 t/s
- Ornith 1.5 9B9B · MMLU-Pro N/AFP32 · ~15.5 t/s
- Llama 3.1 8B Instruct8B · MMLU-Pro 48.3FP32 · ~17 t/s
- DeepSeek R1 Distill Llama 8B8B · MMLU-Pro 41.0FP32 · ~17 t/s
- Qwen3 8B8B · MMLU-Pro 56.7FP32 · ~16.9 t/s
- Qwen 2.5 7B Instruct7.6B · MMLU-Pro 56.3FP32 · ~18.2 t/s
- Mistral 7B Instruct v0.37.25B · MMLU-Pro 30.0FP32 · ~18.7 t/s
- Gemma 3 4B Instruct4B · MMLU-Pro 43.6FP32 · ~34 t/s
- Gemma 4 E4B4B · MMLU-Pro 69.4FP32 · ~33 t/s
- Phi-3.5 Mini Instruct3.8B · MMLU-Pro 47.4FP32 · ~30.5 t/s
- Phi-4-mini Instruct3.8B · MMLU-Pro 67.3FP32 · ~34.5 t/s
- Llama 3.2 3B Instruct3.2B · MMLU-Pro 24.0FP32 · ~40.9 t/s
- Qwen 2.5 3B Instruct3.1B · MMLU-Pro 32.4FP32 · ~44.2 t/s
- Gemma 2 2B Instruct2.6B · MMLU-Pro 17.8FP32 · ~49.8 t/s
- Gemma 4 E2B2B · MMLU-Pro 60.0FP32 · ~66.8 t/s
- SmolLM2 1.7B Instruct1.7B · MMLU-Pro 19.0FP32 · ~66.8 t/s
- Qwen 2.5 1.5B Instruct1.5B · MMLU-Pro 16.8FP32 · ~90.1 t/s
- Llama 3.2 1B Instruct1.24B · MMLU-Pro 12.5FP32 · ~107.4 t/s
- Gemma 3 1B Instruct1B · MMLU-Pro 14.7FP32 · ~129.8 t/s
- Qwen 2.5 0.5B Instruct0.5B · MMLU-Pro 10.0FP32 · ~267.3 t/s
- SmolLM2 360M Instruct0.36B · MMLU-Pro 8.0FP32 · ~316.3 t/s
Models that fit with CPU offload (8)
These use system RAM for layers that don't fit in VRAM, so expect much slower inference.
- Mixtral 8x22B Instruct v0.1141B · MMLU-Pro 40.0Q2_K · ~2.2 t/s
- Mistral Medium 3.5 128B128B · MMLU-Pro N/AQ2_K · ~3.1 t/s
- Qwen 3.5 122B-A10B (MoE)122B · MMLU-Pro 86.7Q3_K_M · ~4.8 t/s
- Nemotron 3 Super 120B120B · MMLU-Pro 83.7Q3_K_M · ~4.9 t/s
- GPT-OSS 120B117B · MMLU-Pro 80.7Q3_K_M · ~9.6 t/s
- Llama 4 Scout 109B109B · MMLU-Pro 74.3Q3_K_M · ~4 t/s
- GLM-4.5 Air 106B106B · MMLU-Pro 81.4Q3_K_M · ~7.1 t/s
- GLM-4.6V 106B106B · MMLU-Pro 79.9Q3_K_M · ~7.1 t/s
Too large for this GPU (29)
- Llama 3.1 405B Instruct
- DeepSeek V3 671B
- DeepSeek R1 671B
- Llama 4 Maverick 400B
- Qwen3 235B-A22B (MoE)
- MiniMax M1 456B
- GLM-4.5 355B
- GLM-4.6 355B
- GLM-4.7 358B
- MiniMax M2.5 229B
- GLM-5 744B
- MiniMax M2.7 229B
- Kimi K2.6
- GLM-5.1 754B
- DeepSeek V4 Pro 1.6T
- DeepSeek V4 Flash 284B
- GLM-5.2 753B
- Nemotron 3 Ultra 550B-A55B
- Step 3.5 Flash
- Step 3.7 Flash
- MiMo V2.5 Pro
- Kimi K2.5
- MiniMax M3
- Inkling
- Kimi K3
- DeepSeek V4 Flash 0731 284B
- Qwen3.8 2.4T-A95B
- DeepSeek V4 Pro 0813 1.6T
- Ornith 1.5 397B (MoE)
Continue reading
Frequently asked questions
- How much VRAM does the AMD Radeon PRO W7900 have?
- The AMD Radeon PRO W7900 has 48 GB of GDDR6 with 864 GB/s memory bandwidth.
- What is the AMD Radeon PRO W7900 best for?
- With 48 GB of VRAM, the AMD Radeon PRO W7900 is ideal for running 70B-class models at Q4 quantization and large MoE models, a workstation sweet spot for local inference.
- What LLMs can the AMD Radeon PRO W7900 run locally?
- The AMD Radeon PRO W7900 can run 57 of the 94 open-weight models tracked by CanItRun natively in VRAM at 8k context. Top options include: Qwen 3.8 27B at Q8_0, Muse Glimmer 30B at Q8_0, Ornith 1.5 9B at FP32.
- Can the AMD Radeon PRO W7900 run Gemma 4 31B?
- Yes. The AMD Radeon PRO W7900 runs Gemma 4 31B natively in VRAM at Q8_0 quantization, achieving approximately 16.4 tokens per second.
- Can the AMD Radeon PRO W7900 run Qwen 3.6 27B?
- Yes. The AMD Radeon PRO W7900 runs Qwen 3.6 27B natively in VRAM at Q8_0 quantization, achieving approximately 19.2 tokens per second.
- Can the AMD Radeon PRO W7900 run Qwen3 8B?
- Yes. The AMD Radeon PRO W7900 runs Qwen3 8B natively in VRAM at FP32 quantization, achieving approximately 16.9 tokens per second.