{"product_id":"nvidia-dgx-spark-8211-personal-ai-desktop-supercomputer-desktop-gb10-grace-blackwell-chip","title":"NVIDIA DGX Spark™ – Personal AI Desktop Supercomputer – Desktop GB10 Grace Blackwell Chip","description":"\u003cp\u003eThe NVIDIA DGX Spark puts data-center-class AI compute in a chassis the size of a hardback book. It is built for machine learning engineers, researchers and small AI teams who need to prototype, fine-tune and run large models locally instead of queuing for cluster time or paying cloud GPU bills. If you spin up an A100 or H100 instance every time you test a fine-tune, this replaces that loop with hardware on your desk, with no idle-GPU billing and no dataset leaving the building.\u003c\/p\u003e\u003ch2\u003eWho it's for\u003c\/h2\u003e\u003cul\u003e\n\u003cli\u003eFine-tuning an open-weight 70B model with LoRA\/QLoRA\u003c\/li\u003e\n\u003cli\u003eIterating on a RAG pipeline against a private document set, offline and air-gapped if required\u003c\/li\u003e\n\u003cli\u003eBenchmarking quantization and serving inference for models in the 70B–200B class\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eKey features\u003c\/h2\u003e\u003cul\u003e\n\u003cli\u003e\n\u003cstrong\u003eNVIDIA GB10 Grace Blackwell Superchip:\u003c\/strong\u003e a 20-core Arm CPU (10x Cortex-X925 performance cores + 10x Cortex-A725 efficiency cores) fused to a Blackwell-architecture GPU with 5th-generation Tensor Cores and 4th-generation RT Cores over NVLink-C2C, rated for up to 1 petaFLOP of FP4 AI compute with sparsity.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e128GB unified LPDDR5x memory at 273GB\/s:\u003c\/strong\u003e CPU and GPU share one coherent 128GB pool over a 256-bit memory interface, with no PCIe round-trip copying between system RAM and VRAM. It holds models that would run out of memory on a 24GB or even 48GB discrete workstation GPU.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eRun and fine-tune models up to ~200B parameters:\u003c\/strong\u003e at FP4 precision, a single unit hosts models in the 70B–200B class for local inference, LoRA\/QLoRA fine-tuning and RAG pipeline development.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003e4TB self-encrypting NVMe SSD:\u003c\/strong\u003e local headroom for multiple model checkpoints, quantized variants, container images and full training datasets.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eConnectX-7 Smart NIC (200Gbps):\u003c\/strong\u003e pair two DGX Spark units with a certified QSFP112 interconnect cable to pool 256GB of unified memory and run inference on models up to roughly 405B parameters.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eFull NVIDIA DGX OS + AI stack, pre-installed:\u003c\/strong\u003e CUDA, cuDNN, NCCL, PyTorch and TensorFlow containers, NVIDIA AI Workbench, and accelerated frameworks such as Isaac, Metropolis and Holoscan. Workloads are container-portable to other Blackwell systems or DGX Cloud without code changes.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eDesktop footprint:\u003c\/strong\u003e 150 x 150 x 50.5mm and 1.2kg, powered from a single 240W supply, with 1x HDMI 2.1a and up to 3x DisplayPort 2.1 over USB-C.\u003c\/li\u003e\n\u003cli\u003e\n\u003cstrong\u003eConnectivity:\u003c\/strong\u003e 4x USB Type-C, 10 Gigabit Ethernet, Wi-Fi 7, Bluetooth 5.4.\u003c\/li\u003e\n\u003c\/ul\u003e\u003ch2\u003eSpecifications\u003c\/h2\u003e\u003ctable\u003e\n\u003cthead\u003e\u003ctr\u003e\n\u003cth\u003eSpec\u003c\/th\u003e\n\u003cth\u003eDetail\u003c\/th\u003e\n\u003c\/tr\u003e\u003c\/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd\u003eSuperchip\u003c\/td\u003e\n\u003ctd\u003eNVIDIA GB10 Grace Blackwell\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCPU\u003c\/td\u003e\n\u003ctd\u003e20-core Arm (10x Cortex-X925 + 10x Cortex-A725)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eGPU\u003c\/td\u003e\n\u003ctd\u003eNVIDIA Blackwell architecture, 5th Gen Tensor Cores, 4th Gen RT Cores\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eCPU-GPU Interconnect\u003c\/td\u003e\n\u003ctd\u003eNVLink-C2C\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eAI Performance\u003c\/td\u003e\n\u003ctd\u003eUp to 1 petaFLOP (FP4, with sparsity)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eSystem Memory\u003c\/td\u003e\n\u003ctd\u003e128GB LPDDR5x, coherent unified memory\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMemory Interface \/ Bandwidth\u003c\/td\u003e\n\u003ctd\u003e256-bit \/ up to 273GB\/s\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eStorage\u003c\/td\u003e\n\u003ctd\u003e4TB NVMe M.2, self-encrypting\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eNetworking\u003c\/td\u003e\n\u003ctd\u003eNVIDIA ConnectX-7 Smart NIC (200Gbps), 10 Gigabit Ethernet (RJ-45), Wi-Fi 7, Bluetooth 5.4\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eUSB\u003c\/td\u003e\n\u003ctd\u003e4x USB Type-C\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDisplay Outputs\u003c\/td\u003e\n\u003ctd\u003e1x HDMI 2.1a, up to 3x DisplayPort 2.1 (USB-C DP Alt Mode)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eVideo Encode\/Decode\u003c\/td\u003e\n\u003ctd\u003e1x NVENC \/ 1x NVDEC\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eOperating System\u003c\/td\u003e\n\u003ctd\u003eNVIDIA DGX OS with full NVIDIA AI software stack (CUDA, cuDNN, NCCL, container runtime)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eMax Model Size\u003c\/td\u003e\n\u003ctd\u003e~200B parameters (single unit); ~405B parameters (dual-unit clustered, 256GB pooled memory)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003ePower Supply\u003c\/td\u003e\n\u003ctd\u003e240W (GB10 TDP 140W)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eDimensions\u003c\/td\u003e\n\u003ctd\u003e150 x 150 x 50.5mm\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eWeight\u003c\/td\u003e\n\u003ctd\u003e1.2kg\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003ctr\u003e\n\u003ctd\u003eConfigurations\u003c\/td\u003e\n\u003ctd\u003eSingle Unit; Dual Unit + Interconnect Cable (NVIDIA-certified ConnectX-7 QSFP112)\u003c\/td\u003e\n\u003c\/tr\u003e\n\u003c\/tbody\u003e\n\u003c\/table\u003e\u003ch2\u003eGood to know\u003c\/h2\u003e\u003cul\u003e\n\u003cli\u003eThe 1 petaFLOP figure is a peak FP4-with-sparsity number. Real dense-precision throughput will be lower.\u003c\/li\u003e\n\u003cli\u003eIt is not a from-scratch training rig: with 273GB\/s memory bandwidth and single-GPU-equivalent compute, pre-training multi-billion-parameter models from random initialization is cluster or cloud territory.\u003c\/li\u003e\n\u003cli\u003eA 70B model in FP4\/INT4 quantization, or a 30B model at higher precision with generous context length, loads without tensor-parallel sharding.\u003c\/li\u003e\n\u003cli\u003eTwo configurations: a Single Unit, or a Dual Unit + Interconnect Cable bundle (two units plus an NVIDIA-certified ConnectX-7 QSFP112 cable) for 256GB of pooled memory.\u003c\/li\u003e\n\u003c\/ul\u003e\u003cp\u003eChoose the single unit if your largest model fits in 128GB, or the dual-unit bundle for models up to roughly 405B parameters.\u003c\/p\u003e","brand":"NVIDIA","offers":[{"title":"Single Unit","offer_id":58232537219454,"sku":"NVIDIA-DGXSPARK-SINGLE","price":4699.0,"currency_code":"USD","in_stock":true},{"title":"Dual Unit + Interconnect Cable","offer_id":58232537252222,"sku":"NVIDIA-DGXSPARK-DUAL","price":9617.99,"currency_code":"USD","in_stock":false}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/1022\/4468\/0062\/files\/516QauLL31L._SL1500_faf69533-16fc-49ac-8eb7-e3869a4085db.jpg?v=1792094024","url":"https:\/\/computahardware.net\/products\/nvidia-dgx-spark-8211-personal-ai-desktop-supercomputer-desktop-gb10-grace-blackwell-chip","provider":"ComputaHardware","version":"1.0","type":"link"}