Nvidia’s DGX Spark 64GB goes on sale Friday, Oct. 23, through Acer, Asus, Dell, Gigabyte, HP and MSI, with prices starting at $4,999. The new model keeps the same GB10 Grace Blackwell chip and software stack as the 128GB version but halves the unified memory, making it a cheaper entry point for developers who want to run AI models on their own desk instead of in a cloud account.
Cheaper needs context. The 128GB DGX Spark launched at about $3,999, and Nvidia’s listing for that model has since risen to $6,950 as memory prices climb, The Register reported. The 64GB unit is therefore less expensive than the current 128GB system, but roughly 25% more than the original 128GB model cost at launch.
What the DGX Spark 64GB keeps
The smaller system is not a stripped-down processor. It uses the same 20-core Arm CPU from MediaTek, the same ConnectX-7 networking and the same DGX OS with Nvidia’s CUDA-X libraries, Nemotron models and support for Ollama, vLLM and PyTorch. Memory bandwidth remains 273 GB/s, according to The Register, which suggests Nvidia switched to lower-capacity LPDDR5X modules rather than removing memory channels.
Nvidia says a single unit can run models with up to 100 billion parameters entirely on device. Two 64GB systems can be linked with a QSFP cable to pool 128GB and support models up to 200 billion parameters. In Nvidia’s test with Qwen 3.8 27B, a pair delivered up to 1.7 times the performance of one system. The company’s Sync Cluster Assistant detects the connected units, checks their configuration and sets up the network.
The sales pitch is a private AI workstation. A coding or research agent can run around the clock on local data without sending prompts to a remote API. Nvidia describes the 64GB model as running capable local agents “privately, without cloud dependency.” That fits the broader push toward agentic AI, where models plan steps and call tools rather than answer one question at a time.
The price is lower, the value is not
The 64GB model looks like a response to a memory shortage rather than a breakthrough in cost. Nvidia cut the memory and storage while raising the price compared with the original 128GB launch, because the components inside have become more expensive. The Register reported that the new configuration also halves storage, though Nvidia’s own announcement focuses on memory and does not publish one storage, port and warranty specification across all six partners.
Less memory also narrows the workload. A 64GB system is less suited to fine-tuning large models, where weights, optimizer states and training data compete for the same pool. Nvidia argues that models in the 26-35 billion parameter range, such as Qwen 3.8 27B, are now good enough for local inference and private agents. That may be true for many coding and document workflows, but it is a different job from training or serving a frontier model.
Buyers should also check what ships on day one. TechRepublic reported that the Sync Cluster Assistant is described as arriving later this month rather than being available to every buyer immediately. Partner systems may differ in storage, ports and support terms. A second unit also doubles the cost to nearly $10,000 before networking and power, which weakens the “cheap” case for anyone who needs the full 128GB pool.
Who the 64GB Spark is for
The best fit is a developer or small team running inference, private agents, coding assistants or models that comfortably fit in 64GB. It is a reasonable way to keep sensitive data on site and avoid per-token cloud bills for steady workloads. It is a poor fit for teams that need to fine-tune large models, serve many concurrent users or keep a 200-billion-parameter model in one box.
The comparison is not only against Nvidia’s own 128GB system. AMD’s Gorgon Halo chips offer memory configurations from roughly 32GB to 192GB, and systems using them currently sell for less than Nvidia’s 128GB DGX Spark, The Register reported. Nvidia’s counter is that the GB10’s GPU is substantially faster for AI tasks, based on the publication’s testing. Buyers are left weighing capacity against throughput and software maturity.
The DGX Spark 64GB makes local AI more accessible relative to today’s inflated 128GB price. It does not make it cheap. Nvidia has produced a carefully cut workstation for a market where memory is the scarce resource, and the sticker price reflects that scarcity as much as the silicon.
I am a person who is positive about every aspect of life.I have always been an achiever be it academics or professional life. I believe in success through hard work & dedication.
Technology Blogger at TechnoSecrets.com
