Radeon AI Pro R9700 vs RTX 3090
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 309018 verified runs · best 709.5 tok/s
Specs and evidence
| Radeon AI Pro R9700 | RTX 3090 | |
|---|---|---|
| VRAM | 32 GB | 24 GB |
| Sourced runs on the board | 14 | 18 |
| Models measured | 14 | 11 |
| Best output tok/s | 969.4 tok/s | 709.5 tok/s |
| Highest measured peak VRAM | 92.2 GB | 22.0 GB |
| Street price (latest snapshot) | — | $1,720 |
Head-to-head
On the 3 models measured on both cards, Radeon AI Pro R9700 leads on 0, RTX 3090 on 3. The widest gap is on Qwen3.8-27B: Radeon AI Pro R9700 64.3 tok/s vs RTX 3090 164.6 tok/s (RTX 3090 156% faster).
The measured matrix
| Model | Radeon AI Pro R9700 | RTX 3090 |
|---|---|---|
| Qwen3.6-35B-A3B | 969.4 | — |
| LFM2.5-1.2B-Instruct | — | 709.5 |
| Qwen3.5-0.8B | 554.5 | — |
| LFM2.5-230M | 520.0 | — |
| Qwen3.5-0.8B-MTP | — | 476.1 |
| LFM2.5-350M | 466.0 | — |
| NVIDIA-Nemotron-3.5-Lightning-30B-A3B | — | 353.7 |
| LFM2.5-2.6B | — | 305.6 |
| Qwen3.5-27B | 286.6 | — |
| LFM2.5-8B-A1B | 241.1 | — |
| Qwen3.5-4B-MTP | — | 240.4 |
| gemma-4-E2B-it | 169.7 | — |
| Qwen3.8-27B | 64.3 | 164.6 |
| Qwen3.5-9B-MTP | — | 164.3 |
| gemma-4-E4B-it | 137.0 | — |
| Muse-Glimmer-30B | 134.6 | — |
| DeepSeek-V4-Flash-0731 | 59.4 | — |
| Qwen 3.8 Flash-Next Base | 57.0 | — |
| Qwen3.8-Flash-Next | 25.6 | 38.6 |
| Ornstein3.6-27B-MTP-NSC-ACE-SABER | — | 32.9 |
| GLM-5.3-Flash | 17.4 | 28.1 |
| DeepSeek-V4.1-Flash | — | 26.0 |