DGX Spark 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.
DGX Spark48 sourced runs · best 86.8 tok/s
RTX 309018 verified runs · best 709.5 tok/s
Specs and evidence
| DGX Spark | RTX 3090 | |
|---|---|---|
| VRAM | 128 GB | 24 GB |
| Sourced runs on the board | 48 | 18 |
| Models measured | 9 | 11 |
| Best output tok/s | 86.8 tok/s | 709.5 tok/s |
| Highest measured peak VRAM | — | 22.0 GB |
| Street price (latest snapshot) | — | $1,720 |
Head-to-head
On the 4 models measured on both cards, DGX Spark leads on 3, RTX 3090 on 1. The widest gap is on Qwen3.8-27B: DGX Spark 56.0 tok/s vs RTX 3090 164.6 tok/s (RTX 3090 194% faster).
The measured matrix
| Model | DGX Spark | 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 |
| Qwen3.8-27B | 56.0 | 164.6 |
| Qwen3.5-9B-MTP | — | 164.3 |
| Qwen 3.8 Flash | 86.8 | — |
| Qwen3.8-Flash-Next | 75.7 | 38.6 |
| GLM-5.3-Flash | 64.1 | 28.1 |
| CYBER-FROST-3.8 | 60.7 | — |
| Qwen3.8-Flash-Next-hibrid48 | 60.0 | — |
| NVIDIA-Nemotron-3.5-Super-VL-120B-A12B | 48.0 | — |
| DeepSeek-V4.1-Flash | 47.0 | 26.0 |
| Ornstein3.6-27B-MTP-NSC-ACE-SABER | — | 32.9 |
| Underdog-Saluki-27B | 22.6 | — |