NVIDIA expands DGX Spark family The new version comes with 64GB of unified memory. This configuration comes as the cost of memory components continues to put pressure on devices dedicated to artificial intelligence, and at least in the U.S. market, it will offer Same $4,999 price as previous 128GB model.
The DGX Spark sells for $3,999, but over the months the price has increased to $4,699 and $4,999. The 128GB version should now be placed The price range is even higherwhile the new model has reduced memory allocation to maintain the reference price.
DGX Spark 64 GB: Features and Usage in the Cluster
In addition to memory, the new configuration Maintains the features of DGX Spark. The system integrates DGX OS and NVIDIA’s AI software stack, and connectivity includes a ConnectX-7 interface that allows you to directly connect multiple units via QSFP cables. Clustering allows you to combine unified memory from different systems, increasing the space available for running large language models. NVIDIA reports that two connected DGX Sparks provide up to 1.7x performance improvements and double the memory bandwidth.
System configuration can be performed through guides published by NVIDIA and the Sync Cluster Assistant, which is designed to simplify the operations required to connect multiple machines. The ability to use multiple DGX Spark build configurations represents One of the unique elements of the platformespecially for those using memory-intensive models.
RTX Spark: We tried the game on the new Ultrabook with NVIDIA SoC
New 64 GB version Will be available on October 23, 2026 Through partners such as Acer, Asus, Dell, Gigabyte, HP and MSI. NVIDIA is selling this configuration in the US for $4,999.
In terms of competition, the DGX Spark was compared to AMD Ryzen AI Max and Ryzen AI Halo-based systems, which can be configured with up to 128 or 192 GB of unified memory. However, the prices reported by these platforms vary based on manufacturer and configuration, making a direct comparison less direct.
Memory issues become particularly important in AI loading: larger models require more capacity to load and execute locally. However, NVIDIA also focuses on Possibility to combine multiple DGX Sparks and CUDA ecosystem, while the AMD platform relies on the ROCm stack and supports Windows and Linux configurations.
Therefore, the new version does not change the architecture of the DGX Spark, but it modifies the trade-off between memory and price. The 64 GB model extends the range to those who don’t need 128 GB, while the larger capacity version is still suitable for workloads that benefit from a larger memory allocation.