NVIDIA DGX Spark (GB10, 128GB)
Compare NVIDIA DGX Spark (GB10, 128GB) against another machine in the Arena
Reference: Nvidia DGX on Wikipedia
- Class
- SoC
- Memory
- 128 GB LPDDR5X
- Runs (Q4_K_M)
- —
- Bandwidth
- 273 GB/s
- TDP
- 240 W
- Released
- 2025-10-15
- Price (US)
- $4699 new as of 2026-08
- Price (Canada)
- CA$8449.99 new sourced, as of 2026-08
Sourcing and disambiguation notes
273 GB/s is the binding constraint: on dense 70B-class models decode lands around 3-5 tok/s, while an MoE like gpt-oss-120b (~5.1B active) decodes 35-53 tok/s depending on the llama.cpp build -- updates during October 2025 alone raised gpt-oss-120b decode from ~38.5 to ~52.9 tok/s on unchanged hardware. Peak system power is specified at 240W (140W GB10 SoC plus up to 100W rest-of-system); users report roughly 170W under sustained load. The price is the Founders Edition, the only configuration NVIDIA sells directly; it rose from $3,999 at the October 2025 launch, which NVIDIA attributes to LPDDR5X supply constraints. Partner machines (Acer Veriton GN100, Asus Ascent GX10, Dell Pro Max GB10) start around $3,999 for a 1TB variant. The Canadian listing had one unit online and one in a single store, and sits well above the US band; NVIDIA and partner AI mini-PC pricing has been volatile in Canada this year.
Specification sources
- https://www.nvidia.com/en-us/products/workstations/dgx-spark/ — vendor, 2025-10-15
- https://www.engadget.com/ai/nvidia-starts-selling-its-3999-dgx-spark-ai-developer-pc-120034479.html — press, 2025-10-15
Measurements
Decode is token generation — the speed you feel while an answer streams. Prefill is prompt processing — the wait before it starts. Why bandwidth predicts decode speed.
No records yet for this hardware. Know of a published benchmark on it? Submit the link.
What verified, single-source and estimated mean, and the same rows with every filter and sort in the benchmarks explorer.
Speed over time
What it can run
Fit is arithmetic, not a measurement — how it is computed. Against 128 GB; models that fit are listed largest first.
No modeled quant fits in 128 GB.