×

HOW TO SHOP

1 Login or create new account.
2 Review your order.
3 Payment & FREE shipment

If you still have problems, please let us know, by sending an email to support@website.com . Thank you!

SHOWROOM HOURS

Mon-Fri 9:00AM - 6:00AM
Sat - 9:00AM-5:00PM
Sundays by appointment only!

FORGOT YOUR DETAILS?

CREATE ACCOUNT

← All news
PC

NVIDIA’s RTX Spark PCs target local AI as DGX kit links four systems

Original article image from Guru3D

NVIDIA is extending its local AI hardware lineup in two directions: compact Windows desktops for individual users and a four-system developer kit for researchers. According to Guru3D’s November 18 report, NVIDIA says RTX Spark PCs from ASUS, Dell and MSI are available, with additional manufacturers expected to follow.

Windows desktops built around local AI

RTX Spark combines a Grace CPU with a Blackwell GPU, offering up to 128GB of unified memory. NVIDIA claims up to one petaflop of AI compute and support for models containing up to 120 billion parameters. Its stated uses include running AI agents, generating images and video, and fine-tuning models locally.

Acer, GIGABYTE, HP and Microsoft are expected to introduce systems too. Pricing and detailed configurations for individual partner models were not provided in the report.

NVIDIA also lists native Windows on Arm support, RTX acceleration in Adobe Creative Cloud applications and DLSS 4.5 Ray Reconstruction in Blender 5.3. Availability and functionality depend on the application and its software release. For PC users, those dependencies matter when assessing which advertised workflows a particular configuration can support.

A research cluster with a storage discrepancy

Introduced at Supercomputing 26 in Chicago, the DGX Spark HPC Developer Kit connects four systems directly through ConnectX-7 networking in a ring, without a separate network switch. NVIDIA lists 512GB of combined memory and claims support for models up to 700 billion parameters using NVFP4.

The kit targets development and debugging of multi-node workloads before researchers scale them to larger supercomputers. NVIDIA cites LAMMPS with Kokkos and NAMD demonstrations; an HPC SDK playbook remains forthcoming.

One configuration detail needs clarification: NVIDIA’s SC26 blog lists 16TB of NVMe storage, while its press information gives 20TB. Buyers should confirm the final capacity with the seller. The report says the kit is on display and available for sale at SC26 through NVIDIA and listed hardware partners.

Leave a Reply

Your email address will not be published. Required fields are marked *

TOP