3× DGX Spark
51tok/sEveryday
1 request, prose prompt, with MTP/EAGLE (spekulativ, ~59 ms/step) (per recipe) Source
Intelligence40with thinkingno thinking 25
- Custom kernel required
Local AI · Recipe · 3× DGX Spark
Kimi-K3 UD-IQ1_S with SparkInfer on 3× DGX Spark
by vcruz305
Intelligence (original model) · Artificial Analysis44with thinkingOriginal model's value. Heavily compressed (~0.95 bit), intelligence is likely much lower.
≤ 3 bit: quantization may cost quality
Engine
- Engine
- SparkInfer sparkinfer-k3 + Patches 0001–0026
- Quantization
- GGUF UD-IQ1_S (330 GB)
- Model family
- Kimi-K3
- Context
- —
- Parameters
- 2.78T (HF-Card vcruz305/Kimi-K3-GGUF; HF-API moonshotai/Kimi-K3 safetensors total 2 779 931 837 184)
- Creator
- vcruz305
- GitHub stars
- 9
- Repo updated
- Sep 13, 2026
Measurements
Every sourced value of this recipe, each with its condition and source. Bars relative to the largest value in the group.
No sourced measurements for this recipe.
What you need
- Hardware
- 3 × NVIDIA DGX Spark (GB10)
- Weights
- Main weightsvcruz305/Kimi-K3-GGUF other
- othermoonshotai/Kimi-K3 other
- Engine
- SparkInfer sparkinfer-k3 + Patches 0001–0026
Notes
What matters before you rebuild it.
- ≤ 3 bit: quantization may cost qualityAt 3 bit and below the model may answer noticeably worse than the original. The intelligence number refers to the original.
- Experimental
- Very heavy quantization
- Multiple nodes
- License unclear
Sources
Related recipes
3× DGX Spark
38.8tok/sPeak
1 request, best value reported by source Source
Intelligence25with thinking
- Superseded
3× DGX Spark
38.2tok/sPeak
1 request, mixed prompt set Source
Intelligence25with thinking
- Experimental
- Custom kernel required
3× DGX Spark
10.5tok/sPeak
1 request, best value reported by source Source
Intelligence29with thinking
- License unclear
Local AI in your company?
In a workshop we work out which models and which hardware fit your tasks, and build the first agent on your infrastructure.