← /pulse/nvidia-dgx-spark-64gb-oem-oct-23-starting-4999
INFOHARDWARE LAUNCH·2026-10-20

NVIDIA DGX Spark 64GB: OEM-only, on sale Oct 23, starting at $4,999

▼ WHAT HAPPENED

NVIDIA announced a 64GB configuration of DGX Spark on October 2, 2026. Its blog says the 64GB model is available from Acer, ASUS, Dell, Gigabyte, HP and MSI on Friday, October 23, starting at $4,999. NVIDIA says it is sold only through manufacturer (OEM) partners. NVIDIA's developer-forum post from the same day lists the partners without ASUS. The 64GB model keeps the GB10 Grace Blackwell Superchip, DGX OS and the NVIDIA AI software stack of the 128GB model. NVIDIA says it supports models up to 100 billion parameters. NVIDIA's spec page lists 64 GB or 128 GB of LPDDR5X on a 256-bit interface and a single memory bandwidth figure of 273 GB/s, not split by memory size. It also lists up to 1 PFLOP FP4, a 20-core Arm CPU (10 Cortex-X925 + 10 Cortex-A725) and a 140 W GB10 TDP. Two units can connect over their ConnectX-7 ports with a QSFP cable. NVIDIA Sync Cluster Assistant then configures the network, and two 64GB units pool their memory to 128GB. NVIDIA says this expands support to models up to 200 billion parameters with twice the memory bandwidth. It also says that in its own Qwen 3.8 27B test, two clustered 64GB systems delivered up to 1.7x the performance of one. The spec page's sizing table also puts a single 128 GB unit at up to 200B parameters. The blog names llama.cpp, Ollama, vLLM and LM Studio as supported inference frameworks. NVIDIA's product page lists no price. In a forum post dated February 25, 2026, NVIDIA said the DGX Spark Founders Edition MSRP had moved from $3,999 to $4,699 because of memory supply constraints.

▼ OPERATOR ANGLE

If you already run a 128GB DGX Spark, there is nothing to change: NVIDIA says the 64GB model uses the same software stack. If you are buying, decide on memory before price. NVIDIA's 100B figure is a parameter count with no stated precision or context length, so size your model, quantization and KV cache with a VRAM calculator first. If you might outgrow 64GB, compare one 128GB unit with two 64GB units before you buy; clustering needs a second unit, a QSFP cable and NVIDIA Sync Cluster Assistant. Compare partner listings for storage and price after October 23 rather than relying on the starting figure. Start with llama.cpp, Ollama, vLLM or LM Studio, the runtimes NVIDIA names. We have not benchmarked the 64GB model or a two-unit cluster.
[pulse item] · runlocalai.co/pulse/nvidia-dgx-spark-64gb-oem-oct-23-starting-4999