NVIDIA · Edge AI · 2025
Robotics computer with 128 GB and Blackwell. Similar LLM speed to a DGX Spark.
from $5,499 (typical price)
Researched from manufacturer data and public sources; not tested by us.· Last reviewed Oct 11, 2026
Popular models at Q4_K_M with a 32K-token context. Memory needed from real model files; speed is the theoretical ceiling.
Model weightsContext cache (KV)RuntimeUsable limit (colour = verdict)
| Model | Verdict | Memory | Speed ceiling |
|---|---|---|---|
| Qwen3 1.7B | Runs | ≤ 246 tok/s | |
| Qwen3.5 9B | Runs | ≤ 48 tok/s | |
| Gemma 4 12B | Runs | ≤ 38 tok/s | |
| gpt-oss-20b | Runs | ≤ 136 tok/s | |
| Qwen3.6 27B | Runs | ≤ 16 tok/s | |
| Qwen3.6 35B A3B | Runs | ≤ 148 tok/s | |
| Llama 3.3 70B | Runs | ≤ 6 tok/s | |
| gpt-oss-120b | Runs | ≤ 100 tok/s | |
| Qwen3.5 122B A10B | Runs | ≤ 45 tok/s | |
| Qwen3.5 397B A17B | Not enough memory | — |
Theoretical maximum for generating text: memory bandwidth divided by the bytes read per token. Real runs usually reach about half to three quarters of it.
Prices are editorial references (launch price and a typical street price on the date shown), not live prices. In 2026 the memory shortage has pushed many prices well above launch.
Prices are shown only if updated in the last 24 hours and may change; the retailer’s price at checkout applies. We may earn a commission from these links at no cost to you. Affiliate disclosure
Best for: Robotics and physical-AI projects; a GB10 box is the better home LLM server.
For robotics. For LLMs alone, a DGX Spark or Strix Halo box costs less.
Source: Manufacturer specifications
How well the popular local-AI tools run on it today (editorial assessment).
Measured results with their source. We never invent speeds.
No measured results yet. The table above shows the theoretical ceiling instead.
DGX Spark / GB10
Largest popular model at 4-bit: Qwen3.5 122B A10B
128 GB with CUDA in a tiny box: the NVIDIA answer to Strix Halo and the Mac.
NVIDIA GPUs
Largest popular model at 4-bit: Qwen3.5 122B A10B
96 GB on one card at RTX 5090 speed: gpt-oss-120b and 122B MoE models entirely in GPU memory.
Ryzen AI
Largest popular model at 4-bit: Qwen3.5 122B A10B
Strix Halo's successor with up to 192 GB (160 GB for the GPU): Qwen3 235B in a mini PC.