NVIDIA RTX Spark Compact Desktops From ASUS, Dell and MSI Go on Sale With Up to 128GB Memory
Source: blogs.nvidia.com

Image: StockSnap on Pixabay
NVIDIA said on November 18 that compact RTX Spark desktops from ASUS, Dell and MSI are available now. In the company's announcement, NVIDIA describes them as Windows PCs for running local AI agents, image and video generation and model fine-tuning, with up to 128GB of unified memory and NVIDIA's AI software stack.
Three systems on sale, three more makers to follow
The first machines are the ASUS ProArt GR1X, the MSI EdgeMesa N AI+ and the Dell Creator Edition Desktop. Launch coverage says Acer, GIGABYTE and HP will ship their own RTX Spark mini PCs later, as will Microsoft's Surface RTX Spark Dev Box. No dates for those were given in the coverage I reviewed.
NVIDIA's pitch is that a single memory pool shared by the CPU and GPU leaves room for larger local models than a typical desktop graphics card with fixed VRAM. The same machines are also meant for creative and development work that benefits from RTX acceleration.
Hardware and prices reported at launch
Press coverage of the launch lists a 20-core Grace CPU paired with a Blackwell RTX GPU that has 6,144 CUDA cores, a rating of up to 1 petaflop of AI performance and 10Gb Ethernet. The same reports say the systems can run local models of up to 120 billion parameters. I could not confirm those figures against NVIDIA's own page, so treat them as reported rather than official.
Prices come from retail and press reports, not from NVIDIA:
- ASUS ProArt GR1X: from $4,499 with 64GB of memory and 1TB of storage.
- MSI EdgeMesa N AI+: reported at $4,399 for 64GB and $7,499 for 128GB. Some outlets list slightly different figures, so check MSI's store before ordering.
- Dell Creator Edition Desktop: no price has been published.
The 128GB ceiling is the top configuration, not the baseline. The cheapest listed models carry 64GB.
Running models on the machine
The launch announcement is about hardware, so there is no model, licence or context length to report. For software, NVIDIA said in earlier posts that its llama.cpp and vLLM optimizations are available directly and through LM Studio and Ollama. It also publishes a llama.cpp setup guide for RTX Spark, and Ollama serves its local API on port 11434 once installed. The announcement does not mention MLX, which targets Apple silicon.
Which open-weight models fit in 64GB or 128GB depends on the quantization and the context length you need. That is the sizing question buyers will have to answer for themselves, since the launch material does not list tested model configurations.
Pressure on Apple and AMD, and what remains unproven
These desktops compete for the same buyers as Mac minis and Mac Studios with large unified memory, and as AMD-based mini PCs with big shared pools. At $4,399 and up, they are priced for developers and studios, not casual users. A Windows-based system with NVIDIA's CUDA stack does have an edge for anyone whose tools already depend on it.
Several questions are still open:
- Dell's price and configurations have not been published.
- Memory bandwidth, which largely determines how fast a large model generates text, was not in the coverage I reviewed.
- No independent benchmarks exist yet. Speedup claims from NVIDIA, including up to 1.9x faster local inference from its llama.cpp and vLLM work, come from the vendor.
- Pricing and availability outside the US have not been announced.
- Dates for the Acer, GIGABYTE, HP and Surface systems are unknown.
Until reviewers run real models on retail units, the practical limits of a 64GB or 128GB RTX Spark box remain a matter of NVIDIA's claims. Buyers who need a specific model should check its memory requirement first, then confirm the configuration with the vendor.
Featured image: StockSnap on Pixabay