Radeon AI Pro R9700 x3 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 R9700 x39 sourced runs · best 223.4 tok/s
RTX 309018 verified runs · best 709.5 tok/s
Specs and evidence
| Radeon AI Pro R9700 x3 | RTX 3090 | |
|---|---|---|
| VRAM | 32 GB | 24 GB |
| Sourced runs on the board | 9 | 18 |
| Models measured | 9 | 11 |
| Best output tok/s | 223.4 tok/s | 709.5 tok/s |
| Highest measured peak VRAM | 25.8 GB | 22.0 GB |
| Street price (latest snapshot) | — | $1,720 |
Head-to-head
Only one model has been measured on both cards (Qwen3.8-Flash-Next). On that single data point Radeon AI Pro R9700 x3 records 28.1 tok/s and RTX 3090 records 38.6 tok/s — treat it as one data point, not a ranking.
The measured matrix
| Model | Radeon AI Pro R9700 x3 | RTX 3090 |
|---|---|---|
| LFM2.5-1.2B-Instruct | — | 709.5 |
| Qwen3.5-0.8B-MTP | — | 476.1 |
| NVIDIA-Nemotron-3.5-Lightning-30B-A3B | — | 353.7 |
| LFM2.5-2.6B | — | 305.6 |
| Qwen3.5-4B-MTP | — | 240.4 |
| gpt-oss-120b | 223.4 | — |
| Qwen3.8-27B | — | 164.6 |
| Qwen3.5-9B-MTP | — | 164.3 |
| Nemotron-Cascade-2-30B-A3B | 95.4 | — |
| Qwen3-VL-8B-Instruct | 58.8 | — |
| Muse-Glimmer-30B | 47.8 | — |
| Ling-3.0-flash | 39.3 | — |
| Qwen3.8-Flash-Next | 28.1 | 38.6 |
| Ornith-1.5-35B-A3B | 35.4 | — |
| Ornstein3.6-27B-MTP-NSC-ACE-SABER | — | 32.9 |
| Ornith-1.0-9B | 31.2 | — |
| GLM-5.3-Flash | — | 28.1 |
| DeepSeek-V4.1-Flash | — | 26.0 |
| gemma-4-31B-it | 23.9 | — |