DGX Spark vs Radeon AI Pro R9700 x3
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.
DGX Spark52 sourced runs · best 86.8 tok/s
Radeon AI Pro R9700 x39 verified runs · best 223.4 tok/s
Specs and evidence
| DGX Spark | Radeon AI Pro R9700 x3 | |
|---|---|---|
| VRAM | 128 GB | 32 GB |
| Sourced runs on the board | 52 | 9 |
| Models measured | 9 | 9 |
| Best output tok/s | 86.8 tok/s | 223.4 tok/s |
| Highest measured peak VRAM | — | 25.8 GB |
| Street price (latest snapshot) | — | — |
Head-to-head
Only one model has been measured on both cards (Qwen3.8-Flash-Next). On that single data point DGX Spark records 75.7 tok/s and Radeon AI Pro R9700 x3 records 28.1 tok/s — treat it as one data point, not a ranking.
The measured matrix
| Model | DGX Spark | Radeon AI Pro R9700 x3 |
|---|---|---|
| gpt-oss-120b | — | 223.4 |
| Nemotron-Cascade-2-30B-A3B | — | 95.4 |
| Qwen 3.8 Flash | 86.8 | — |
| Qwen3.8-Flash-Next | 75.7 | 28.1 |
| GLM-5.3-Flash | 64.1 | — |
| CYBER-FROST-3.8 | 60.7 | — |
| Qwen3.8-Flash-Next-hibrid48 | 60.0 | — |
| Qwen3-VL-8B-Instruct | — | 58.8 |
| Qwen3.8-27B | 56.0 | — |
| NVIDIA-Nemotron-3.5-Super-VL-120B-A12B | 48.0 | — |
| Muse-Glimmer-30B | — | 47.8 |
| DeepSeek-V4.1-Flash | 47.0 | — |
| Ling-3.0-flash | — | 39.3 |
| Ornith-1.5-35B-A3B | — | 35.4 |
| Ornith-1.0-9B | — | 31.2 |
| gemma-4-31B-it | — | 23.9 |
| Underdog-Saluki-27B | 22.6 | — |