Run big models locally
Inference on models up to 200 billion parameters — Llama, DeepSeek, Qwen, Gemma and others — straight from the 128 GB unified memory, with no copying between CPU and GPU memory.
NVIDIA DGX Spark, now in India
A petaflop of AI
on your desk.
The GB10 Grace Blackwell Superchip with 128 GB of unified memory, in a box 15 cm wide. Run, test and fine-tune large AI models locally — no cloud GPU bills, and your data never leaves the room.
- AI performance
- 1 petaFLOP FP4
- Unified memory
- 128 GB
- Models on one unit
- 200B parameters
- Size
- 15 × 15 cm, 1.2 kg
What you can do with it
Fine-tune up to 70B
Adapt open models to your own data for chatbots, code assistants, document search or trading research, and keep that data private on your own hardware.
Scale to 405B with two units
Link two DGX Spark systems through the built-in ConnectX-7 200 Gb/s ports to work with models up to 405 billion parameters.
Build AI agents
Ships with NVIDIA DGX OS and the NVIDIA AI software stack — CUDA, PyTorch, NIM microservices and pre-trained models — ready to develop and test agents.
Data science at speed
RAPIDS and GPU-accelerated analytics on large datasets, with 1 petaFLOP of AI compute beside your laptop.
Prototype, then deploy
The same NVIDIA software runs in the data center and cloud, so what you build on your desk moves to production without rewriting.
Available DGX Spark systems
Partner systems built on the NVIDIA DGX Spark GB10 platform. Same superchip and memory; they differ in chassis, cooling, storage and warranty.
Nvidia
NVIDIA DGX Spark Personal AI Supercomputer | 128GB RAM Unified Memory, 4TB SSD | Grace Blackwell GB10
₹529,989 ₹590,000incl. GST
Link two units
Connect two DGX Spark systems through their ConnectX-7 ports to work with models up to 405 billion parameters.
Specifications
| Superchip | NVIDIA GB10 Grace Blackwell |
|---|---|
| GPU | NVIDIA Blackwell, 5th-gen Tensor Cores, 4th-gen RT Cores |
| CPU | 20-core Arm (10× Cortex-X925 + 10× Cortex-A725) |
| AI performance | Up to 1 petaFLOP (FP4, with sparsity) |
| Memory | 128 GB LPDDR5x unified system memory, 273 GB/s |
| Storage | 1 TB or 4 TB NVMe M.2, self-encrypting (varies by model) |
| Networking | 10 GbE RJ-45, ConnectX-7 (2× QSFP), Wi-Fi 7, Bluetooth 5.4 |
| Ports | 4× USB Type-C, 1× HDMI 2.1a |
| Software | NVIDIA DGX OS (Ubuntu-based Linux) with the NVIDIA AI software stack |
| Size and weight | 150 × 150 × 50.5 mm, 1.2 kg |
Specifications from NVIDIA for the DGX Spark platform. Partner systems may differ in storage, cooling and accessories — see each product page.
DGX Spark or DGX Station?
DGX Spark
For one developer or a small team
- GB10 Grace Blackwell
- 128 GB unified memory
- Up to 1 petaFLOP FP4
- Models up to 200B (405B with two units)
- Desktop, quiet, 1.2 kg
DGX Station
For labs and enterprise AI teams
- GB300 Grace Blackwell Ultra
- Several hundred GB of coherent memory
- Data-center-class AI performance at a desk
- Trillion-parameter-scale work
- Built to order
Why buy from TechCure India
GST invoice
Full tax invoice with input credit for companies, colleges and research labs.
Setup and onboarding
We unbox, update DGX OS and help you run your first model — on-site in Surat and Gujarat, remote anywhere in India.
Insured delivery
Shipped insured to any pin code in India, with warranty claims handled through us.
Questions
Can DGX Spark run Windows or games?
No. It runs NVIDIA DGX OS, an Ubuntu-based Linux built for AI development on the Arm-based GB10 chip. For gaming or Windows software, a custom PC is the better choice.
How large a model can I run?
Inference on models up to about 200 billion parameters and fine-tuning up to about 70 billion on one unit. Two linked units handle up to about 405 billion parameters.
Why is DGX Spark better than a gaming GPU for AI?
Its 128 GB of memory is shared by the CPU and GPU, so large models that do not fit in a graphics card's 16–32 GB of VRAM can load and run. For small models, a high-end graphics card can be faster; for large models, memory size decides what you can run at all.
Which DGX Spark model should I choose?
All use the same GB10 superchip and 128 GB memory. Choose on storage (1 TB or 4 TB), cooling and noise, warranty and price — tell us your use case on WhatsApp and we will recommend one.
Do you deliver outside Surat?
Yes, across India with insurance. On-site setup is available in Surat and Gujarat.
Tell us what you want to build.
Share your model size, team size and budget — we reply with the right system and a GST quotation.
Chat with us on WhatsApp