NVIDIA DGX Spark - new little AI supercomputer

In july 2025, soon it should be available

Page content

Nvidia is about to release NVIDIA DGX Spark - little AI supercomputer on blackwell architecture with 128+GB unified RAM and 1 PFLOPS AI performance. Nice device to run LLMs.

Nvidia Dgx Spark This image is from Nvidia site, see more internals photos and spec details there.

Overview of NVIDIA DGX Spark

The NVIDIA DGX Spark is a compact, high-performance AI supercomputer designed for desktop form factors. Powered by the NVIDIA Grace Blackwell GB10 Superchip, it integrates fifth-generation Tensor Cores and FP4 precision to deliver 1,000 TOPS (trillion operations per second) of AI compute power. Key features include:

  • 128GB unified memory (LPDDR5X)
  • 4TB NVMe SSD storage
  • NVLink-C2C interconnect (5x bandwidth over PCIe Gen5)
  • Compatibility with advanced AI models (e.g., NVIDIA Cosmos Reason, GR00T N1)

Targeted at professionals, researchers, and developers, the DGX Spark aims to democratize access to supercomputer-level AI capabilities in a compact, affordable form factor.

NVIDIA DGX Spark Availability

Right now NVIDIA DGX Spark is available for reservation, we don’t know when it will be released yet.

If you are wondering if you can buy it in your country - NVIDIA FAQ answers it clearly

Currently reservations are available in the UK, France, Italy, Germany, Spain and European Union.

People expect it to come to customers in July 2025.

NVIDIA DGX Spark Performance

dgx-spark-specs

The NVIDIA DGX Spark has 1 PFLOPS AI performance which is lower then 1.5 PFLOPS of RTX 6000 Ada, But the price is also a bit lower.

NVIDIA DGX Spark Pricing and Competition

Pricing for the DGX Spark varies by configuration and partner:

  • Base model: Estimated at $3,000 - “$4,000 (pre-order available for (north hemisphere) summer 2025 delivery).
  • Partner variants:
    • Asus Ascent GX10: $3,000 (1TB storage, 128GB memory).
    • Dell Pro Max with GB10: Equivalent to DGX Spark.
    • HP ZGX Nano AI Station G1n: Likely equivalent to DGX Spark.
  • DGX Station (more powerful variant): Expected to be significantly higher (estimated at $50,000 for 784GB memory).

Note: Prices are region-specific and not directly listed on NVIDIA’s website. Users must contact partners (e.g., ASUS, MSI, GIGABYTE) for procurement.

Apple Studio has similar memory config, actualy it can have 512GB unified RAM, and it might have comparable performance. I didn’t check, but it might.

The price for this cool DGX Spark is expected to be around $4K USD which is somehow lower than Apple Studio

For professionals seeking affordable, high-performance AI tools, the DGX Spark is a compelling option. However, further analysis is needed to fully assess its competitive landscape and long-term market impact.

Compatibility

https://docs.nvidia.com/dgx/dgx-os-6-user-guide/introduction.html

The DGS Spark will have DGX OS 6 installed

The main info about it:

  • Based on Ubuntu 22.04 with the latest long-term Linux kernel version 5.15 for the recent hardware and security updates and updates to software packages, such as Python and GCC.
  • Includes the NVIDIA-optimized Linux kernel, which supports GPU Direct Storage (GDS) without additional patches.
  • Provides access to all NVIDIA GPU driver branches and CUDA toolkit versions.

Should be good.

Other useful hardware