AMD · RDNA 3
Best AI Models for AMD Radeon PRO W7800 (32 GB)
Runs 27–31B and 35B MoE models well; 70B doesn't fit in 32 GB.
32 GB GDDR6576 GB/s91 TFLOPS FP16260 WRDNA 3 2023
| # | Model | Speed | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 16.9 t/s | 16.9 tok/s | Q4_K_M · 21.7 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 71.7 t/s | 71.7 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 68.6 t/s | 68.6 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 16.9 t/s | 16.9 tok/s | Q4_K_M · 21.7 GB | |
| 3 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 61.0 t/s | 61.0 tok/s | Q4_K_M · 22.0 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 71.7 t/s | 71.7 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 68.6 t/s | 68.6 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 16.9 t/s | 16.9 tok/s | Q4_K_M · 21.7 GB | |
| 3 | Qwen3.5 27BAlibaba Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 71.7 t/s | 71.7 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 68.6 t/s | 68.6 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 · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 61.0 t/s | 61.0 tok/s | Q4_K_M · 22.0 GB | |
| 4 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 16.9 t/s | 16.9 tok/s | overall | Q4_K_M · 21.7 GB |
| 5 | Q4_K_M · 19.3 GB · 68.6 t/s | 68.6 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 106 t/s | 106 tok/s | Q4_K_M · 3.4 GB | |
| 2 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 81.3 t/s | 81.3 tok/s | Q4_K_M · 13.1 GB | |
| 3 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 71.7 t/s | 71.7 tok/s | Q4_K_M · 16.5 GB | |
| 4 | Q4_K_M · 19.3 GB · 68.6 t/s | 68.6 tok/s | Q4_K_M · 19.3 GB | |
| 5 | Qwen3 30B A3B Thinking 2507Alibaba Q4_K_M · 19.4 GB · 64.3 t/s | 64.3 tok/s | Q4_K_M · 19.4 GB |
| # | Model | Speed | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 19.8 t/s | 19.8 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 16.9 t/s | 16.9 tok/s | Q4_K_M · 21.7 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 71.7 t/s | 71.7 tok/s | Q4_K_M · 16.5 GB | |
| 5 | MiniCPM V 4.7 35B A3BOpenBMB Q4_K_M · 22.1 GB · 66.3 t/s | 66.3 tok/s | Q4_K_M · 22.1 GB |
See all 346 models on the PRO W7800 →Best quant that fits · speeds estimated on this card
The PRO W7800 runs 27–31B and 35B MoE models well. RDNA 3 generation, released 2023.
GenerationRDNA 3released 2023
PRO W7800 strengths and weaknesses
Strengths
- Runs 35B MoE models: Qwen3.6 35B A3B at 61.0 tok/s in 22 GB
- Room for long context: Qwen3.8 27B at Q4 uses 17.9 GB, leaving 14.1 GB free
Weaknesses
- 70B models don't fit: Llama 3.3 70B Instruct needs 51.3 GB at Q4 vs 32 GB here
- Slow prompt reading: A 4K-token prompt on Qwen3.8 27B waits 11.4 s for the first token
- ROCm, not CUDA: No FP8 weights; MXFP4 only through llama.cpp, not vLLM
Output speed depends on memory bandwidth (576 GB/s). Grey rows need more than 32 GB.
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB54.3tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB36.4tok/s
- 27–31BQwen3.8 27B · Q4_K_M · 17.9 GB19.8tok/s
- 35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB61.0tok/s
- 70BLlama 3.3 70B Instruct · not at 4-bitneeds 51.3 GB
- 120B MoEGPT OSS 120B · not at 4-bitneeds 70.8 GB
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
All 1176 models on the PRO W7800
Featured onlyBest quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
Lucie 7BChat | Q4_K_M | 6.5 GB | 75.3 t/s | 32K | S86, Runs great | |
GPT OSS Safeguard 20BChatFunctions | Q4_K_M | 13.5 GB | 74.9 t/s | 131K | S86, Runs great | |
Mellum2 12B A2.5B ThinkingChatFunctions | Q4_K_M | 7.8 GB | 99.5 t/s | 131K | S91, Runs great | |
| Q4_K_M | 0.0 GB | 34560.0 t/s | 1K | S100, Runs great | ||
Pythia 1BChat | Q4_K_M | 0.7 GB | 486.8 t/s | 2K | S100, Runs great | |
| Q4_K_M | 6.2 GB | 64.1 t/s | 8K | A84, Runs well | ||
| Q4_K_M | 7.6 GB | 62.6 t/s | 33K | A84, Runs well | ||
| Q4_K_M | 0.1 GB | 2304.0 t/s | 1K | S100, Runs great | ||
Openai GPTChat | Q4_K_M | 0.1 GB | 4320.0 t/s | 1K | S100, Runs great | |
| Q4_K_M | 1.2 GB | 323.0 t/s | 128K | S100, Runs great | ||
Phi 1 5ChatCode | Q4_K_M | 0.9 GB | 367.7 t/s | 2K | S100, Runs great | |
DialoGPT SmallChat | Q4_K_M | 0.1 GB | 2880.0 t/s | 1K | S100, Runs great | |
SmolLM3 3B BaseChat | Q4_K_M | 3.4 GB | 150.3 t/s | 66K | S97, Runs great | |
| Q4_K_M | 6.2 GB | 64.1 t/s | 8K | A84, Runs well | ||
Gemma 4 12BChat | Q4_K_M | 9.0 GB | 42.0 t/s | 262K | A76, Runs well |
Specs
- Architecture
- RDNA 3
- VRAM
- 32 GB GDDR6
- Memory bandwidth
- 576.0 GB/s
- FP16 (dense)
- 90.50 TFLOPS
- Stream processors
- 4,480
- TDP
- 260 W
- Released
- 2023-04-13
- MSRP
- $2,499
PRO W7800 FAQ
What is the largest model the PRO W7800 can run?
Qwen3.6 35B A3B (36.0B) is the largest at 4-bit or better: at Q4_K_M it uses 22.0 GB of 32 GB, and runs at about 61.0 tok/s.
Can the PRO W7800 run Qwen3.8 27B?
Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of 32 GB, and it is the best model that runs well on this card.
Can the PRO W7800 run 70B models?
No. Llama 3.3 70B Instruct needs 51.3 GB at Q4, more than the 32 GB on this card.
How many tokens per second does the PRO W7800 give?
About 19.8 tok/s on Qwen3.8 27B at Q4_K_M, with the first token after 11.4 s on a 4K-token prompt.
What power supply does the PRO W7800 need?
Plan a 550 W PSU (2× the 260 W TDP).
Is 32 GB enough for coding and agents?
Yes for coding: Qwen3.8 27B, the best of the 10 coding models that fit, runs at about 20 tok/s. For agents, not yet: no agent-scored model fits in less than 96 GB.