AI desktop
NVIDIA DGX Spark
128 GB of unified memory · 273 GB/s
Lowest tracked new price
$7,394View at Amazon Compare all retailer offersWe may earn a commission from retailer links. Affiliate disclosure
Where to buy
- $7,394View at Amazon
Amazonnew
NVIDIA DGX Spark™ - Personal AI Desktop Supercomputer – Desktop GB10 Grace Blackwell Chip
Price read
Some links are affiliate links. Prices change constantly, so confirm the exact configuration, condition, shipping, and final total at the retailer before buying.
Recorded catalog price
$7,394
$58 per GB of memory
0.04 GB/s per dollar
Recorded . In stock at that check. Retailer offers above may differ by condition or configuration.
Specs
- Memory
- 128 GB LPDDR5X
- Memory bandwidth
- 273 GB/s
- FP16 compute
- Not published
- INT8 compute
- Not published
- Whole-machine power
- 240 W
- MSRP
- Not published
- Memory architecture
- Unified with the CPU
Fields marked "not published" are ones the vendor does not state. We leave them blank rather than estimate.
Board power covers the GPU card alone; whole-machine power covers the complete computer. A card's rating excludes the rest of the system. See our power, cooling, and noise guide for power-supply planning.
What it can run
Whether a model's weights fit in 128 GB, with a 20 percent allowance for the KV cache, activations, and the runtime. This is arithmetic on the quantization format, not a benchmark. Long contexts and large batches need more.
At Q4, the smallest format below, 20 of the 25 models we size fit on this machine.
| Model | Q4 (4-bit) ? | Q5 (5-bit) | Q8 (8-bit) | FP16 |
|---|---|---|---|---|
| Qwen3.5-0.8B (0.8B) | Fits | Fits | Fits | Fits |
| LFM2.5-1.2B-Thinking (1.17B) | Fits | Fits | Fits | Fits |
| Qwen3.5-2B (2B) | Fits | Fits | Fits | Fits |
| SmolLM3-3B (3B) | Fits | Fits | Fits | Fits |
| Qwen3.5-4B (4B) | Fits | Fits | Fits | Fits |
| Gemma 4 E4B (8B) | Fits | Fits | Fits | Fits |
| Qwen3-8B (8.2B) | Fits | Fits | Fits | Fits |
| Qwen3.5-9B (9B) | Fits | Fits | Fits | Fits |
| Gemma 4 12B (11.95B) | Fits | Fits | Fits | Fits |
| gpt-oss-20b (21B) | Fits | Fits | Fits | Fits |
| Mistral Small 3.2 24B (24B) | Fits | Fits | Fits | Fits |
| Gemma 4 26B-A4B (25.2B) | Fits | Fits | Fits | Fits |
| Gemma 3 27B (27B) | Fits | Fits | Fits | Fits |
| Qwen3.5-27B (27B) | Fits | Fits | Fits | Fits |
| Qwen3.8-27B (27B) | Fits | Fits | Fits | Fits |
| Gemma 4 31B (30.7B) | Fits | Fits | Fits | Fits |
| Qwen3.5-35B-A3B (35B) | Fits | Fits | Fits | Fits |
| Llama 3.3 70B Instruct (70B) | Fits | Fits | Fits | No |
| gpt-oss-120b (117B) | Fits | Fits | No | No |
| Qwen3.5-122B-A10B (122B) | Fits | Fits | No | No |
| Qwen3-235B-A22B (235B) | No | No | No | No |
| DeepSeek-V4-Flash-0731 (284B) | No | No | No | No |
| DeepSeek-V3 (671B) | No | No | No | No |
| GLM-5.1 (744B) | No | No | No | No |
| Kimi K3 (2800B) | No | No | No | No |
Price history
31 recorded checks. Low $4,000, high $7,394.
Recent checks
- 2026-10-01 $7,394
- 2026-09-30 $6,501
- 2026-09-29 $6,500
- 2026-09-28 $6,500
- 2026-09-27 $6,500
- 2026-09-26 $6,500
- 2026-09-25 $6,500
- 2026-09-24 $6,500
- 2026-09-23 $6,500
- 2026-09-21 $5,000
Related hardware
Closest memory capacity to this one, which is what decides whether the same models load.
- Apple Mac Studio (M4 Max, 128GB) — 128 GB
- Apple Mac Studio (M5 Max, 128GB) — 128 GB
- Framework Desktop (Ryzen AI Max+ 395, 128GB) — 128 GB
Sources
Every spec here is recorded from a vendor page we fetched. See our methodology for how prices and specs are sourced.
- NVIDIA DGX Spark product page — NVIDIA , fetched