Nvidia debuts 64GB DGX Spark at $4,999 amid memory crunch

Nvidia has introduced a cut-down 64GB configuration of its DGX Spark personal AI computer: it halves the unified memory and storage of the 128GB model. It goes on sale through Acer, ASUS, Dell, Gigabyte, HP and MSI on October 23, starting around $4,999. It arrives amid the memory crunch: Nvidia has raised the price of the 128GB model to $6,950, nearly 75 percent above its launch price.

What stays the same

Otherwise the system is unchanged, Nvidia says: it keeps the GB10 Grace Blackwell Superchip, the 20-core Arm CPU and 273 GB/s of bandwidth, plus onboard ConnectX-7 networking for clustering. The Register notes the unchanged bandwidth points to lower-capacity LPDDR5X modules. One unit supports up to 100 billion parameters; two linked by a QSFP cable pool their memory to 128GB and handle up to 200 billion parameters.

Easier clustering

Nvidia is also adding the NVIDIA Sync Cluster Assistant, a Sync app feature that detects units, checks their configuration and sets up the ConnectX-7 network automatically. Previously, multi-node setups required command-line work and vLLM or TensorRT-LLM tweaks. Tom's Hardware notes Sync also gains a Model Launcher that downloads and runs models such as Qwen 3.8 27B across systems and links them to the OpenCode coding agent.

The 64GB Spark ships with DGX OS and Nvidia's AI stack, including the Agent Toolkit, CUDA-X libraries, Nemotron models and runtimes such as Ollama, vLLM and PyTorch.

Why it matters

The $4,999 price is still about 25 percent above the 128GB launch price, while in-stock 128GB GB10 systems sell for roughly $7,000 to $9,000, Tom's Hardware notes. The 64GB model is less suited to memory-hungry workloads such as fine-tuning, but for local inference and private agents, models in the 26-35 billion parameter range, like Qwen 3.8 27B, are good enough.

The pricing pressure could also shape Nvidia's RTX Spark notebooks and mini PCs, built on the same GB10 silicon and due this fall. AMD's Ryzen AI Max+ 495 systems cost less and offer more memory capacity, but The Register's testing shows the GB10 GPU delivers substantially higher AI performance.

Sources: The Register · NVIDIA Newsroom · Tom's Hardware