DGX Spark (GB10 Grace Blackwell): 128 GB, what it runs and what it costs
Specs checked · Rental prices from 18m ago · Aliteq Team · How we compute this
The short answer
The NVIDIA DGX Spark (GB10 Grace Blackwell) has 128 GB of VRAM and 273 GB/s of memory bandwidth (2025). At Q4_K_M quality or better with an 8K context it runs 41 of the open models we track; the largest that fits with room to spare is Llama 4 Scout 17B-16E (MoE). The cheapest rental right now is $0.133/hr (Vast.ai, interruptible).
- VRAM
- 128 GB
- Bandwidth
- 273 GB/s
- Board power
- 140 W
- Launched
- 2025
Source: NVIDIA specification · bandwidth not yet confirmed by a second source, so we don't estimate speed from it
Buy it or rent it
| Provider | Type | From, per hour | Rent |
|---|---|---|---|
| Vast.ai | Spot (can be interrupted) | $0.133/hr | Vast.ai |
| Vast.ai | On demand | $0.240/hr | Vast.ai |
Lowest price per provider in our hourly snapshot (18m ago). Rent buttons are referral links. Every card side by side: cloud GPU prices.
Which AI models run on the DGX Spark (GB10 Grace Blackwell)
Computed by our VRAM engine for an 8K context: the best quality each model fits at in 128 GB. Speeds are upper-bound estimates from memory bandwidth, not benchmarks.
| Model | Best fit | VRAM used | ~tok/s |
|---|---|---|---|
| Qwen3.8-Flash-Next (MoE) | Q5_K_M | 120 GB94% · tight | — |
| Qwen3.5 122B-A10B (MoE) | Q8_0 | 125 GB97% · tight | — |
| NVIDIA Nemotron 3 Super 120B-A12B (MoE) | Q8_0 | 123 GB96% · tight | — |
| gpt-oss 120B (MoE) | Q8_0 | 117 GB91% · tight | — |
| Llama 4 Scout 17B-16E (MoE) | Q8_0 | 109 GB85% | — |
| GLM-4.5-Air (MoE) | Q8_0 | 107 GB84% | — |
| Qwen3-Next 80B-A3B (MoE) | Q8_0 | 80 GB63% | — |
| Qwen3-Coder-Next 80B-A3B (MoE) | Q8_0 | 80 GB62% | — |
| DeepSeek R1 Distill 70B | Q8_0 | 73 GB57% | — |
| Mixtral 8x7B (MoE) | F16 | 89 GB69% | — |
| Qwen3.6 35B-A3B (MoE) | F16 | 68 GB53% | — |
| Ornith-1.5 35B-A3B (MoE) | F16 | 68 GB53% | — |
Fits only below Q4 quality: Qwen3 235B-A22B (MoE) (Q4_0), MiniMax-M2.5 (MoE) (Q4_0).
Too big for 128 GB (rent a bigger card instead): what it costs to run MiniMax-M3 (MoE), what it costs to run GLM-5.3 (MoE), what it costs to run DeepSeek V4.1 Flash (MoE), what it costs to run DeepSeek V4 Pro (MoE).
Check a model and context of your own in the VRAM calculator.
Our guides on the DGX Spark (GB10 Grace Blackwell)
Cards to compare it with
- AMD Ryzen AI Max+ 395 (Strix Halo,) · 128 GB
- Apple M4 Max (16-core CPU / 40-core GPU) · 128 GB
- Apple M2 Max · 96 GB
- RTX PRO 6000 Blackwell Max-Q Workstation Edition · 96 GB
- RTX PRO 6000 Blackwell Workstation Edition · 96 GB
- Apple M4 Pro · 64 GB
Every card by VRAM tier: all GPUs for gaming and local AI.