NVIDIA DGX Spark 4TB Personal AI Supercomputer

📞 Call for best price 📞 Call 💬 WhatsApp

  • AI Processor: NVIDIA® GB10 Grace Blackwell Superchip — 1 petaFLOP AI Performance (FP4, Sparse); 5th-Gen Tensor Cores; 4th-Gen RT Cores
  • CPU: NVIDIA® Grace CPU — Arm 20-core (10× Cortex-X925 + 10× Cortex-A725)
  • Memory: 128 GB LPDDR5x Unified System Memory — 256-bit interface, 273 GB/s bandwidth; run models up to 200B parameters
  • Storage: 4 TB NVMe M.2 SSD with self-encryption
  • Connectivity: Wi-Fi 7 | Bluetooth® 5.3 | 10 GbE (ConnectX-7 Smart NIC @ 200 Gbps) | 4× USB Type-C
  • Display Output: 1× HDMI 2.1a + up to 3× DisplayPort via USB-C
  • OS: NVIDIA® DGX™ OS — AI software stack pre-installed
  • Form Factor: Compact Mini PC — 150 × 150 × 50.5 mm, ~1.14L, 1.2 kg | Cluster 2 units for 405B-parameter models | 1-Year Warranty

Out of stock

Email when stock available

Ask about this on WhatsApp
Pre-composed with SKU + price. Reply usually within minutes during business hours.
SKU: 940-54242-0008-000 Categories: ,
Description

The NVIDIA DGX Spark (4TB) brings petaFLOP-class AI computing to your desk. Powered by the NVIDIA® GB10 Grace Blackwell Superchip with 128 GB LPDDR5x unified system memory and a 4 TB self-encrypting NVMe M.2 SSD, it delivers 1 petaFLOP AI performance (FP4) in a compact ~1.14-litre form factor. Run large language models up to 200 billion parameters, fine-tune models up to 70 billion parameters, or cluster two DGX Spark units to tackle models up to 405 billion parameters. Built on the NVIDIA® DGX™ Spark platform, the DGX Spark ships with the full NVIDIA DGX software stack — ready for AI development, deep learning, data science, fine-tuning, inference, and computer vision workflows straight out of the box.

AI Performance
PlatformNVIDIA DGX™ Spark
SuperchipNVIDIA® GB10 Grace Blackwell Superchip
AI Performance (FP4)1 PFLOP (petaFLOP) — Sparse, FP4
Tensor Cores5th Generation
RT Cores4th Generation
Supported PrecisionsTF32, FP16, BF16, INT8, FP8, FP6, FP4
NVENC / NVDEC1× NVENC / 1× NVDEC
CPU & Memory
CPUNVIDIA® Grace CPU — Arm 20-core (10× Cortex-X925 + 10× Cortex-A725)
System Memory128 GB LPDDR5x Coherent Unified System Memory
Memory Interface256-bit
Memory Bandwidth273 GB/s
Max Model Size (Inference)Up to 200 billion parameters
Max Model Size (Fine-Tuning)Up to 70 billion parameters
Storage
Primary Storage4 TB NVMe M.2 SSD (Self-Encrypting)
Connectivity
Wi-FiWi-Fi 7
BluetoothBluetooth® 5.3
Ethernet1× RJ-45 10 GbE
Smart NICNVIDIA ConnectX-7 Smart NIC (200 Gbps)
USB4× USB Type-C
Display Outputs
HDMI1× HDMI 2.1a
DisplayPort (via USB-C)Up to 3× DisplayPort 1.4a (DP Alt Mode over USB-C)
Audio OutputHDMI Multichannel Audio
Max Simultaneous DisplaysUp to 4 (1× HDMI + up to 3× via USB-C)
Operating System
OSNVIDIA® DGX™ OS (pre-installed)
AI Software StackNVIDIA AI software stack pre-installed — optimised tools, libraries, frameworks, and pre-trained model support
Dimensions, Weight & Power
Dimensions (L × W × H)150 × 150 × 50.5 mm (~1.14L)
Weight1.2 kg
Power Supply240 W
GB10 TDP140 W
Multi-Unit AI Clustering
Cluster InterfaceConnectX-7 (QSFP — 200 Gbps)
Max Cluster Size2 units
Combined AI Performance2 PFLOP FP4 (2× DGX Spark)
Max Clustered Model SizeUp to 405 billion parameters (2× 128 GB unified memory)
QSFP CableSold separately
Additional information
AI SuperchipNVIDIA GB10 Grace Blackwell
AI Performance1 PetaFLOP (FP4)
Unified Memory128 GB LPDDR5x
Model CapacityUp to 200B params
AI Storage4 TB NVMe
AI Networking10 GbE, ConnectX-7 200GbE, Wi-Fi 7
AI CPU20-core Arm (Grace)
AI Operating SystemNVIDIA DGX OS
AI Form FactorMini