PRESSZONE
NVIDIA DGX Spark Now Available to Pre-Order at Scan
If you missed it the NVIDIA DGX Spark was one of the biggest announcements at the spring GTC conference in San Jose. In contrast to the existing range of DGX appliances, which are deployed in datacentres to run the most demanding training and inferencing workloads, the diminutive DGX Spark is a desktop device for AI developers.
Small but mighty
Although the DGX Spark is extremely compact, for reference it’s about half the size of a typical Mini-ITX PC, it still packs a hell of a punch thanks to doing away with a traditional (and wasteful) x86 CPU and discrete GPU. Instead, it’s powered by a highly-efficient SoC, the GB10 Grace Blackwell Superchip comprising a Blackwell GPU and 20 Arm CPU cores, sharing a unified 128GB of memory.
Wi-Fi, Bluetooth, USB
NCCL, RDMA, GPUDirect
1 PetaFLOP FP4 AI Compute
20 Arm Cores
128 GB Low Power DDR5X
With 4TB of NVMe SSD storage and ConnectX networking the DGX Spark delivers up to 1,000 TOPS of AI performance at FP4, roughly a third of a GeForce RTX 5090 or quarter that of an RTX PRO 6000 Blackwell. However, the DGX Spark has four times more memory than the 5090, and 33% more than the 6000, enabling you to run much larger AI models with up to 200 billion parameters.
This makes the DGX Spark ideally suited for developing generative AI and reasoning AI models that are too large to run on traditional discrete GPUs. There’s also the ability to link two DGX Spark’s together, effectively doubling the RAM, to run even larger AI models.
Follow this link to pre-order your NVIDIA DGX Spark now or talk to our AI experts about your project requirements.