The new system, based on Nvidia's GB10 platform, will be available exclusively through hardware partners including Acer, Asus, Dell, Gigabyte, HP, and MSI. Despite the reduced memory capacity, Nvidia says the system is well suited for local AI inference tasks such as running models in the 26–35 billion parameter range, like Qwen 3.8 27B, rather than fine-tuning workloads that require larger memory.
The 64 GB model retains the same memory bandwidth of 273 GB/s, indicating the use of lower-capacity LPDDR5x memory modules rather than fewer modules. The system also keeps the same 20-core Arm processor from MediaTek and onboard ConnectX-7 networking, which allows clustering up to four GB10 devices at 200 Gbps each. Nvidia plans to roll out new software tools for DGX OS later this month to simplify cluster configurations.
The price creep of the DGX Spark has been notable since the Project Digits concept was unveiled at CES. Originally expected to retail for around $3,000, the appliance eventually launched at $4,000. The new 64 GB model, at $4,999, costs 25 percent more than the 128 GB version did at launch. The memory shortage that has driven up prices also casts uncertainty on the upcoming RTX Spark notebooks and mini PCs, which could face higher-than-expected pricing.
Meanwhile, competition is emerging from AMD's newly launched Gorgon Halo SoCs, which offer memory capacities ranging from 32 GB to 192 GB at lower prices. However, benchmark testing has repeatedly shown that Nvidia's GB10 GPU delivers substantially higher performance for AI tasks.