Loading...
Tap to expand1/1Inclusive of all taxes
1 Year Manufacturer's Warranty
Enter your pincode to check delivery options.
₹6,89,990 including GST
Buy 10 or more of this system: ₹6,55,490.50 each including GST (extra 5% off the website price). Applies to units of this same product in one order, subject to availability.
Add your GSTIN at checkout to get a GST invoice. Input tax credit may be available to registered businesses, subject to eligibility.
The NVIDIA DGX Spark is a compact desktop AI computer built on the NVIDIA GB10 Grace Blackwell Superchip, pairing a 20-core Arm CPU with a Blackwell GPU and 128 GB of coherent unified LPDDR5x memory. NVIDIA positions it for developers who test, validate and run inference on AI models locally, up to 200 billion parameters on the 128 GB configuration. Key specifications Specification Value Architecture NVIDIA Grace Blackwell (GB10 Grace Blackwell Superchip) CPU 20-core Arm: 10 Cortex-X925 + 10 Cortex-A725 GPU Blackwell architecture with 5th generation Tensor Cores and 4th generation RT Cores CUDA cores 6,144 Tensor performance Up to 1 PFLOP at FP4 precision System memory 128 GB LPDDR5x, coherent unified system memory (NVIDIA also lists a 64 GB option) Memory interface and bandwidth 256-bit, 273 GB/s Storage NVMe M.2 with self-encryption, up to 4 TB USB 4x USB Type-C Ethernet 1x RJ-45, 10 GbE Network adapter ConnectX-7 NIC at 200 Gbps Wi-Fi Wi-Fi 7 Bluetooth Bluetooth 5.4 Display connectors 1x HDMI 2.1a; up to 3x DisplayPort over USB-C (DP Alt Mode) Audio output HDMI multichannel audio Operating system NVIDIA DGX OS Power supply 240 W GB10 TDP 140 W Dimensions (L x W x H) 150 x 150 x 50.5 mm Weight 1.2 kg Manufacturer part number 940-54242-0000 Compatibility The system runs NVIDIA DGX OS with the NVIDIA AI software stack preinstalled. Display output is one HDMI 2.1a port plus DisplayPort over the USB Type-C ports (DP Alt Mode); there is a single HDMI port, not four. Networking is one 10 GbE RJ-45 port, a ConnectX-7 NIC rated at 200 Gbps, Wi-Fi 7 and Bluetooth 5.4. Good to know NVIDIA states that inference on models of up to 200 billion parameters is supported on the 128 GB configuration. The 128 GB is unified system memory shared by the CPU and GPU, not a separate graphics-card memory pool. The 1 PFLOP figure is an FP4 rate (NVIDIA says 'up to'), not a general floating-point or benchmark figure.
Genuine Product
Brand-sealed, 100% authentic
GST Invoice
Tax invoice included with every order
7-Day Returns
Hassle-free return & replacement
Secure Payment
256-bit SSL encrypted checkout
No, it is not a separate VRAM pool. NVIDIA describes the DGX Spark's memory as 128 GB of LPDDR5x coherent unified system memory with 273 GB/s of bandwidth, shared as one memory space rather than a dedicated graphics-card memory bank. A discrete workstation GPU such as the RTX PRO 6000 instead carries its own 96 GB of GDDR7. Check that your model fits within 128 GB.
NVIDIA says the DGX Spark can test, validate and run inference on AI models of up to 200 billion parameters using its 128 GB of unified system memory. Actual limits depend on the model's precision and format, since the memory is shared with the operating system and the CPU. This is a capacity statement from NVIDIA, not a speed benchmark.
The DGX Spark has four USB Type-C ports, one HDMI 2.1a port, one RJ-45 10 GbE port and a ConnectX-7 network adapter rated at 200 Gbps. Wireless is Wi-Fi 7 and Bluetooth 5.4. NVIDIA also lists up to three DisplayPort outputs over USB-C using DP Alt Mode.
It uses the NVIDIA GB10 Grace Blackwell Superchip: a 20-core Arm CPU (10 Cortex-X925 plus 10 Cortex-A725 cores) and a Blackwell GPU with 5th generation Tensor Cores and 4th generation RT Cores. NVIDIA rates it at up to 1 PFLOP at FP4 precision, and the GB10 has a 140 W TDP.