Nvidia’s DGX Spark is turning heads in the compact workstation world, largely because it promises serious local AI performance in a tiny mini PC-sized package. Built around Nvidia’s powerful GB10 “Grace Blackwell” chip, DGX Spark is designed for running large language models (LLMs) efficiently without the extreme power draw you might expect from modern high-end AI hardware.
At the heart of DGX Spark is the GB10 “Superchip,” combining a 20-core ARM CPU and a Blackwell-based GPU into one purpose-built platform. The CPU layout includes 10 Cortex-X925 cores for heavy lifting alongside 10 Cortex-A725 cores for efficiency-focused workloads. Memory is a major selling point here: DGX Spark comes with 128GB of LPDDR5X on a 256-bit bus, a configuration aimed at helping local AI tasks that are often bottlenecked by memory capacity and bandwidth rather than raw compute alone.
On the graphics side, the integrated Blackwell GPU includes 4th-generation ray tracing cores and 5th-generation Tensor cores, with a total of 6,144 CUDA cores. Nvidia is positioning the system as a compact AI powerhouse, even claiming up to 1 petaflop of compute at FP4 precision using sparse operations, a metric that aligns with modern AI inference techniques optimized for lower precision.
That performance comes at a premium, and now it’s set to cost even more. DGX Spark originally launched with a suggested price of $3,999, but Nvidia has announced a significant price increase of $700. The new MSRP climbs to $4,699, with the company pointing to rising DRAM and storage costs as the main reason behind the jump. Given ongoing supply chain pressures across the tech industry, this kind of increase—especially tied to memory pricing—isn’t entirely unexpected.
The price news also arrives alongside continued concerns about hardware availability. Nvidia has indicated that GPU supply is expected to remain tight for the next few quarters, which adds even more uncertainty for buyers trying to plan AI workstation purchases or upgrades.
For anyone who wants to run LLMs locally but can’t justify a near-$5,000 mini workstation, there are more budget-friendly paths. Systems based on AMD’s Strix Halo platform are being discussed as a more affordable alternative that can handle local AI workloads while also offering solid gaming capability. One example mentioned is the GMKtec Evo-X2, which is positioned as a lower-cost option at around $2,700 and includes 128GB of memory and a 2TB SSD—key specs for users who care about loading larger models and datasets without immediately running into memory limits.
Bottom line: Nvidia DGX Spark is an exciting compact AI workstation built around the GB10 Grace Blackwell platform, offering high-end local LLM performance in a small form factor—but the latest price hike makes it even harder to justify unless you truly need Nvidia’s specific AI stack and the performance it’s targeting.
