Radeon AI Pro R9700 vs RTX 5090
Best sourced output tok/s per model (single-stream generation) — engine and version, quantization, context and a source link on every row. Hover a value for the quant.
Radeon AI Pro R970014 sourced runs · best 969.4 tok/s
RTX 509023 verified runs · best 446.7 tok/s
Specs and evidence
| Radeon AI Pro R9700 | RTX 5090 | |
|---|---|---|
| VRAM | 32 GB | 32 GB |
| Sourced runs on the board | 14 | 23 |
| Models measured | 14 | 16 |
| Best output tok/s | 969.4 tok/s | 446.7 tok/s |
| Highest measured peak VRAM | 92.2 GB | 31.3 GB |
| Street price (latest snapshot) | — | — |
Head-to-head
On the 3 models measured on both cards, Radeon AI Pro R9700 leads on 2, RTX 5090 on 1. The widest gap is on Qwen3.8-27B: Radeon AI Pro R9700 64.3 tok/s vs RTX 5090 435.1 tok/s (RTX 5090 577% faster).
The measured matrix
| Model | Radeon AI Pro R9700 | RTX 5090 |
|---|---|---|
| Qwen3.6-35B-A3B | 969.4 | — |
| Qwen3.5-0.8B | 554.5 | — |
| LFM2.5-230M | 520.0 | — |
| LFM2.5-350M | 466.0 | — |
| NVIDIA-Nemotron-3.5-Lightning-30B-A3B | — | 446.7 |
| Qwen3.8-27B | 64.3 | 435.1 |
| Qwen3.5-27B | 286.6 | — |
| Ornith-1.0-35B-PrismaAURA-4.75bit-vllm-MTP | — | 283.1 |
| Kimi-Linear-48B-A3B-Instruct | — | 280.8 |
| Ornith-1.5-35B-A3B | — | 258.8 |
| North-Mini-Code-1.0 | — | 258.4 |
| LFM2.5-8B-A1B | 241.1 | — |
| Qwen3.6-35B-A3B-MTP | — | 222.9 |
| Laguna-XS-2.1 | — | 216.7 |
| KAT-Coder-V2.5-Dev-AWQ-W4A16-ASYM | — | 201.2 |
| Ornith-1.0-35B | — | 196.6 |
| gemma-4-E2B-it | 169.7 | — |
| gemma-4-E4B-it | 137.0 | — |
| Muse-Glimmer-30B | 134.6 | 124.8 |
| Qwen3.6-27B | — | 120.9 |
| Huihui-Qwen3.8-27B-abliterated | — | 117.7 |
| Qwen3.8-27B-5090-goldilocks | — | 93.5 |
| DeepSeek-V4-Flash-0731 | 59.4 | 26.4 |
| Qwen 3.8 Flash-Next Base | 57.0 | — |
| Qwen3.6-27B-MTP | — | 28.3 |
| Qwen3.8-Flash-Next | 25.6 | — |
| GLM-5.3-Flash | 17.4 | — |