DGX Spark vs DGX Spark x2
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
DGX Spark x233 verified runs · best 474.0 tok/s
Specs and evidence
| DGX Spark | DGX Spark x2 | |
|---|---|---|
| VRAM | 128 GB | 128 GB |
| Sourced runs on the board | 52 | 33 |
| Models measured | 9 | 7 |
| Best output tok/s | 86.8 tok/s | 474.0 tok/s |
| Highest measured peak VRAM | — | — |
| Street price (latest snapshot) | — | — |
Head-to-head
On the 4 models measured on both cards, DGX Spark leads on 1, DGX Spark x2 on 3. The widest gap is on Qwen3.8-Flash-Next-hibrid48: DGX Spark 60.0 tok/s vs DGX Spark x2 474.0 tok/s (DGX Spark x2 690% faster).
The measured matrix
| Model | DGX Spark | DGX Spark x2 |
|---|---|---|
| Qwen3.8-Flash-Next-hibrid48 | 60.0 | 474.0 |
| DeepSeek-V4.1-Flash | 47.0 | 154.0 |
| Qwen3.8-Flash-Next | 75.7 | 93.0 |
| DeepSeek-V4.1-Flash-1017 | — | 92.8 |
| Qwen 3.8 Flash | 86.8 | — |
| GLM-5.3-Flash | 64.1 | 46.3 |
| CYBER-FROST-3.8 | 60.7 | — |
| Beam-1.0 | — | 58.9 |
| Qwen3.8-27B | 56.0 | — |
| NVIDIA-Nemotron-3.5-Super-VL-120B-A12B | 48.0 | — |
| MiniMax M3.1 Flash | — | 39.4 |
| Underdog-Saluki-27B | 22.6 | — |