NVIDIA announced a new DGX Spark configuration with 64GB of unified memory on October 2, 2026. Availability is scheduled to begin on October 23 through Acer, ASUS, Dell, Gigabyte, HP and MSI. In other words, this is an announcement of a model arriving later in the month, not a statement that every reseller already has it in stock.
Less memory, the same system family
According to NVIDIA, the configuration retains the GB10 Grace Blackwell Superchip, DGX OS and the company's AI software environment, as in the 128GB model. The company states support for models of up to 100 billion parameters on a single unit. This is a stated capability limit, not a promise of consistent speed for every model and runtime setting.
For anyone running AI locally, model size is only the start of the choice. It is also necessary to check how much information will be fed into it, how many users will work simultaneously and what response time the real task achieves. A system that can load a model is not necessarily the best fit for the planned workload.
NVIDIA reports that in its test with Qwen 3.8 27B, two connected 64GB units delivered up to 1.7 times the performance of one unit. The figure belongs to that specific company test. It should not be turned into a promise that two units will shorten every job by the same ratio.
A useful suitability test will use the same model and task intended for work after the purchase. A demonstration of response time and memory consumption under the planned load can be requested. That way, the comparison between configurations rests on a defined need, rather than only the maximum parameter count in the announcement.
The connection to a personal computer is part of the experience
NVIDIA Sync documentation explains that the application allows users to connect to remote Linux systems and run applications and development tools on them. It is available for Windows, Mac and Ubuntu. Users can work from their regular computer while computation takes place on another system.
Cluster Assistant helps configure ConnectX-7 connections between DGX Spark units. According to the documentation, network configuration is a step that allows compatible libraries and runtime environments to use the high-speed link. Connecting two devices should not be assumed to distribute every program between them automatically.
Alongside this, NVIDIA announced that Model Launcher is expected to arrive at the end of the month. What is documented today needs to be separated from a future capability included in the same announcement. A software addition's launch date is not the same as the hardware availability date.
For a buyer in Israel, the announcement still lacks a local price, stock at a specific reseller and local warranty terms. Before ordering, it is advisable to request the exact model rather than settling for the DGX Spark name, because memory configurations differ. When comparing offers, the required connection components should also be included if work with two units is planned. The announcement adds a choice; whether the purchase makes sense depends on the task and the actual price offered.