I’m having a hard time understanding how strong this model actually is based on their benchmarks and my Spark is currently being used for some datagen so I can’t try the model out myself. Seems like this type of architecture will most likely be the future of the Spark. Anyone used this to see how strong it is? nvidia/Nemotron-TwoTower-30B-A3B-Base-BF16 · Hugging Face
I can’t get it to work yet. It wants two cuda devices, so a single dgx spark can’t seem to do it, and it wants them both on the same machine so I’m having trouble getting it working using two sparks.
If anyone has had success please share!
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 | 31 | 2426 | June 10, 2026 | |
| Nemotron-3-Super-120B-A12B-NVFP4 on single DGX Spark: 23.45 tok/s (spark-arena.com/ benhmarks) | 6 | 1121 | May 26, 2026 | |
| Help running Nemotron 3 Nano 30B-A3B-FP8 on DGX Spark (GB10) | 41 | 3542 | January 24, 2026 | |
| DGX Spark not installing CUDA, Nemotron, vllm, etc | 5 | 322 | January 30, 2026 | |
| Running nvidia/nemotron-3-super on DGX spark | 12 | 2336 | March 26, 2026 | |
| Guide for fine-tuning NVIDIA-Nemotron-3-Super-120B-A12B-NVFP4 to add new programming language support | 1 | 497 | April 29, 2026 | |
| Nemotron 3 Super: Updates Approaching Agentic Usability | 1 | 488 | April 5, 2026 | |
| [Benchmark] nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-NVFP4 | 5 | 1374 | May 1, 2026 | |
| Nemotron-3-Ultra-550B-A55B (2-bit GGUF) across 2× DGX Spark via llama.cpp RPC — it works (~5 tok/s) | 7 | 857 | June 7, 2026 | |
| DGX Spark, Nemotron3, and NVFP4: Getting to 65+ tps | 14 | 2339 | December 22, 2025 |