Radeon PRO V710 holds 357 of the models in our catalogue and is, in practice, a IQ4_XS card — the largest it takes is Flint at IQ4_XS.
Neighbours in memory rather than in price: memory decides whether a card can do the job at all, so two cards of the same size at different prices is the comparison you are making. Ordered by memory, then bandwidth — not by the value column, which is worked out from bandwidth and price alone and therefore rewards a cheap card whatever its software stack does to that bandwidth in practice. Read it as one input, not as a ranking.
AMD Radeon PRO V710 — 28 GB VRAM.
- Brand
- AMD
- VRAM
- 28 GB GDDR6
- Bandwidth
- 504 GB/s
- FP16 Compute
- 55.3 TFLOPS
- FP32 Compute
- 27.6 TFLOPS
- Stream Processors
- 3,456
- TDP
- 158 W
- Architecture
- RDNA 3.0
With 28 GB VRAM and 504 GB/s bandwidth, this GPU handles models up to 36B parameters.
Speed ≈ bandwidth / model_size × efficiency. A 7B model at Q4 runs at ~58 tok/s.
| Model | Size | VRAM Q4 | tok/s | Avg |
|---|---|---|---|---|
| Seed-OSS 36B Instruct | 36B | 22.5 GB | 12 | 54.4 |
| MiniCPM-V 4.7 35B-A3B | 35.21B | 22.0 GB | 149 | 76.4 |
| c4ai-command-r-v01 35B | 35B | 21.9 GB | 13 | 27.5 |
| Qwen 3.5 35B A3B | 35B | 21.9 GB | 149 | 56.1 |
| Qwen 3.6 35B A3B | 35B | 21.9 GB | 149 | 53.3 |
| K2 Horizon 32B | 34.78B | 21.7 GB | 13 | 54.0 |
| Nous Capybara 34B | 34.4B | 21.5 GB | 13 | 42.0 |
| Yi-1.5 34B | 34.4B | 21.5 GB | 13 | 45.3 |
| Falcon-H1 34B | 34B | 21.3 GB | 13 | 58.6 |
| CodeLlama 34B | 34B | 21.3 GB | 13 | 25.4 |
| Nous Hermes 2 34B | 34B | 21.3 GB | 13 | 47.0 |
| Phind CodeLlama 34B | 34B | 21.3 GB | 13 | 68.1 |
| LLaVA-1.6 Yi 34B | 34B | 21.3 GB | 13 | 47.4 |
| WizardCoder Python 34B | 34B | 21.3 GB | 13 | 73.2 |
| Yi 34B | 34B | 21.3 GB | 13 | 33.4 |
| Qwen3-VL 32B Instruct | 33.36B | 20.9 GB | 13 | 44.6 |
| Nanbeige4.3 33B | 33.12B | 20.7 GB | 14 | 92.4 |
| DeepSeek Coder 33B | 33B | 20.7 GB | 14 | 26.0 |
| Vicuna 33B | 33B | 20.7 GB | 14 | 17.2 |
| LLaMA 1 30B | 33B | 20.7 GB | 14 | 17.8 |