Home / Enterprise Solutions / NVIDIA DGX Spark
NVIDIA DGX Spark · 128GB unified memory · Egypt & UAE
NVIDIA DGX Spark: an AI supercomputer on your desk.
NVIDIA’s GB10 Grace Blackwell Superchip and 128GB of unified memory in a 1.2kg box. Develop, fine-tune and run AI models of up to 200 billion parameters locally, then scale them up to the data centre. Available in Egypt and the UAE from Mega Build Up.
Up to 1 petaFLOP of AI
FP4 performance with sparsity, from the GB10 Grace Blackwell Superchip.
128GB of unified memory
Shared by CPU and GPU: enough for models of up to 200B parameters.
240W and desk-sized
150 × 150 × 50.5 mm. No server room needed.
What it’s for
Built for the people who build AI.
Give developers and researchers their own AI system, without waiting for time on a shared server or sending data to the cloud.
AI developers and data scientists
A dedicated machine to build, test and debug models and AI applications.
Prototyping and fine-tuning
Fine-tune small and mid-size models and test ideas before you scale up.
Private local LLMs
Run large language models on your own desk, with sensitive data kept off the cloud.
AI agents and RAG
Build agents and retrieval systems on your own documents and data.
Universities and research labs
Hands-on AI for teaching and research groups, one unit per team.
Image and video generation
Run creative AI tools such as ComfyUI locally.
Why DGX Spark
Data-centre AI technology, desk-sized.
GB10 Grace Blackwell Superchip
A 20-core Arm CPU (10 Cortex-X925 and 10 Cortex-A725) with a Blackwell GPU and 6,144 CUDA cores.
128GB of unified memory
LPDDR5x shared by CPU and GPU, with 273 GB/s of bandwidth.
NVIDIA AI software stack
DGX OS with CUDA, cuDNN, TensorRT, NVIDIA Container Runtime and AI Workbench.
Link two for 405B models
A ConnectX-7 SmartNIC with two QSFP ports connects two units for models of up to 405B parameters.
Self-encrypting storage
1TB or 4TB of NVMe M.2 storage with self-encryption.
Connected
10GbE, Wi-Fi 7, Bluetooth 5.4, four USB-C ports and HDMI 2.1a.
Specifications
NVIDIA DGX Spark specifications.
NVIDIA’s published hardware specifications for DGX Spark. Storage options and availability are confirmed on your quote.
NVIDIA DGX Spark
| Superchip | NVIDIA GB10 Grace Blackwell |
|---|---|
| CPU | 20-core Arm: 10 Cortex-X925 + 10 Cortex-A725 |
| GPU | NVIDIA Blackwell with 5th-gen Tensor Cores, 4th-gen RT Cores and 6,144 CUDA cores |
| AI performance | Up to 1 PFLOP FP4 (with sparsity) |
| Memory | 128GB unified LPDDR5x, 256-bit |
| Memory bandwidth | 273 GB/s |
| Storage | 1TB or 4TB NVMe M.2, self-encrypting |
| Networking | ConnectX-7 SmartNIC with 2 × QSFP; 1 × 10GbE RJ-45 |
| Wireless | Wi-Fi 7, Bluetooth 5.4 |
| Ports | 4 × USB-C, 1 × HDMI 2.1a |
| Power | 240W external power supply |
| Size and weight | 150 × 150 × 50.5 mm, 1.2kg |
| Model size | Up to 200B parameters; 405B with two linked units |
Included software
| Operating system | NVIDIA DGX OS (based on Ubuntu) |
|---|---|
| AI libraries | CUDA, cuDNN, TensorRT |
| Containers | NVIDIA Container Runtime |
| Development | NVIDIA AI Workbench |
| Remote access | DGX Dashboard and NVIDIA Sync |
Software details are from NVIDIA’s published documentation.
How we deliver
From order to your first model.
01
Consult & size
Tell us what you will build and we recommend the right setup: one unit, two linked units or a GPU server.
02
Source factory-new
New, sealed units with full warranty.
03
Deliver & set up
Delivered and set up on your network, ready for your team.
04
Support for life
Warranty and support from our engineers, plus a clear path to GPU servers when you scale.
Compare our AI systems
From your desk to a full AI cluster.
For development
NVIDIA DGX Spark
128GB unified memory
On your desk · models of up to 200B parameters
You are here
For inference
RTX PRO 6000 Blackwell 8-GPU Server
768GB GPU memory
8 × 96GB GDDR7 · AI video, rendering and simulation
For training
NVIDIA H200 8-GPU Server
1,128GB GPU memory
8 × 141GB HBM3e · NVLink
What can NVIDIA DGX Spark run?
With 128GB of unified memory, one DGX Spark can work with AI models of up to 200 billion parameters, and two linked units with models of up to 405 billion parameters. It is designed for developing, fine-tuning and running models locally.
Is DGX Spark a replacement for a GPU server?
No. DGX Spark is a development system for one person or a small team, and its 273 GB/s of memory bandwidth is far below a data-centre GPU. To serve many users, choose an RTX PRO 6000 8-GPU server; for large-scale training, an H200 8-GPU server.
What software comes with it?
DGX Spark runs NVIDIA DGX OS, based on Ubuntu, with CUDA, cuDNN, TensorRT, NVIDIA Container Runtime and NVIDIA AI Workbench, plus DGX Dashboard and NVIDIA Sync for remote access.
Can I connect two DGX Spark units?
Yes. The built-in ConnectX-7 SmartNIC with two QSFP ports links two units directly, so they can work together on models of up to 405 billion parameters.
Does it need a server room?
No. It runs from a 240W external power supply, weighs 1.2kg and sits on a desk.
Do you supply DGX Spark in Egypt and the UAE?
Yes. We supply DGX Spark to businesses, universities and research teams in Egypt and the UAE, and can set it up on your network.
How do I get a price?
Answer three quick questions in the form on this page and we will reply within 24 working hours with availability and a price.
Get a quote
Order NVIDIA DGX Spark.
Answer three quick questions and an engineer replies within 24 working hours with a recommended configuration and price.
- No obligation, no sales pressure
- Sized by an engineer, not a script
- Factory-new hardware with full warranty
Step 1 of 3
Talk to our AI team about DGX Spark.
Call, WhatsApp or email our team and we’ll reply within 24 working hours.
