NVIDIA · AI Systems
NVIDIA DGX Spark
A petaflop of AI compute on your desk, in a 1.2 kg box the size of a hardback. The NVIDIA DGX Spark pairs the GB10 Grace Blackwell Superchip with 128 GB of coherent unified memory, so the model and the data sit in one address space — prototype, fine-tune and run inference on models up to 200 billion parameters locally, with no cloud queue, no per-hour meter and no data leaving the room. Ships with NVIDIA DGX OS and the full NVIDIA AI stack, and links to a second unit over ConnectX-7 for models up to 405 billion parameters.
- AI performance
- Up to 1 PFLOP (1000 TFLOPS) FP4
- Unified memory
- 128 GB LPDDR5x coherent unified system memory
- Inference capacity
- Models up to 200 billion parameters
- Storage
- 4 TB NVMe M.2 with self-encryption
Price breakup · GST inclusive
- Taxable value
- ₹5,08,389.83
- GST @ 18%HSN 8471
- ₹91,510.17
- You pay
- ₹5,99,900
Shown as CGST 9% + SGST 9% or IGST 18% on your invoice, decided by your delivery state. Add a GSTIN at checkout for a B2B tax invoice you can claim input credit against.
Launch allocation
50 of 50 remaining
0 of 50 units in this batch claimed.
- Reserved allocation · dispatch within 45 days
- 12-month NVIDIA warranty · RMA handled by RIGGED
- GST invoice · add your GSTIN at checkout for input credit
- Insured, tracked delivery across India
Free insured shipping · 48h burn-in · 3-year warranty
Before you pay
Each order commits one unit from a limited launch allocation, so orders can't be cancelled for a change of mind once payment is confirmed. If we miss the 45-day dispatch window, or the unit arrives damaged or not as described, you get a replacement or a full refund — see returns.
Full specification
Manufacturer specification as published by NVIDIA.
- Os
- NVIDIA DGX OS
- Cpu
- 20-core Arm — 10× Cortex-X925 + 10× Cortex-A725
- Gpu
- NVIDIA Blackwell · 5th-gen Tensor Cores · 4th-gen RT Cores
- Ports
- 4× USB Type-C · 1× HDMI 2.1a · up to 3× DisplayPort over USB-C
- Power
- 240 W power supply · GB10 TDP 140 W
- Memory
- 128 GB LPDDR5x coherent unified system memory
- Weight
- 1.2 kg
- Storage
- 4 TB NVMe M.2 with self-encryption
- Best for
- LLM fine-tuning · local inference · agentic AI development · robotics · data science
- Software
- NVIDIA AI software stack with NIM microservices
- Dimensions
- 150 × 150 × 50.5 mm
- In the box
- NVIDIA DGX Spark · 240 W power adapter · regional power cable · quick-start guide
- Networking
- ConnectX-7 @ 200 Gbps · 10 GbE RJ-45 · Wi-Fi 7 · Bluetooth 5.4
- Architecture
- NVIDIA Grace Blackwell — GB10 Superchip
- Model scaled
- Up to 405 billion parameters across two linked units
- Ai performance
- Up to 1 PFLOP (1000 TFLOPS) FP4
- Model finetune
- Fine-tune models up to 70 billion parameters
- Model inference
- Models up to 200 billion parameters
- Memory bandwidth
- 273 GB/s
- Memory interface
- 256-bit
Owner reviews
No reviews yet — this unit has only just launched in India. Reviews here come from verified RIGGED purchases only, so this section stays empty until real owners have theirs on a desk.
