NVIDIA has announced a 64GB version of its DGX Spark personal AI supercomputer, priced from $4,999 and available from hardware partners starting October 23. The new SKU retains the GB10 Grace Blackwell Superchip, DGX OS, and the full NVIDIA AI software stack, and supports up to 100-billion-parameter models on device. Two 64GB units can be clustered via NVIDIA Sync Cluster Assistant, pooling memory to 128GB and delivering up to 1.7x performance in NVIDIA's Qwen 3.8 27B test, according to the company's blog post.
All three sources agree on the core specifications and clustering capabilities, but they frame the launch differently. NVIDIA's own announcement emphasizes accessibility and a "more affordable" configuration without mentioning any trade-offs. The Register, however, notes that the 64GB model also has half the storage of the 128GB version and is less suited for memory-hungry workloads like fine-tuning. It also points out that the 128GB DGX Spark's price has jumped to $6,950 — nearly 75% above its original launch price — due to the ongoing memory shortage. Tom's Hardware similarly cites the "RAMpocalypse," reporting that 128GB systems now sell for roughly $7,000 to $9,000 on the street, and argues that a 64GB option makes sense for local inference with dense models like Qwen 3.8 27B that fit in 32GB.
All sources agree that the 64GB Spark retains the same 273 GB/s memory bandwidth and ConnectX-7 networking as its larger sibling, and that NVIDIA's new Sync Cluster Assistant simplifies multi-node setup. Tom's Hardware adds that a Model Launcher feature will automatically download and run models like Qwen 3.8 27B across one or more systems, while The Register notes that clustering previously required manual CLI configuration. The new 64GB configuration gives local AI developers a lower-cost entry point, though whether it's a genuine bargain or a necessary adjustment to inflated memory prices remains a matter of perspective.