Nvidia has just launched the DGX Spark, which they claim is the world’s smallest AI supercomputer currently available. Equipped with an NVIDIA AI software stack preinstalled and 128 GB of memory, it allows developers to prototype, fine-tune, and run inference on the latest generation of reasoning AI models from DeepSeek, Meta, NVIDIA, Google, Qwen, and others with up to 200 billion parameters locally.
NVIDIA CEO Jensen Huang also enthusiastically conducted the handover and introduced the DGX Spark to popular technology figures such as Elon Musk and Sam Altman. Here is the video of the DGX Spark handover that we obtained from NVIDIA’s YouTube channel.
At OpenAI’s Mission Bay headquarters. Nearly a decade after welcoming the first DGX-1 supercomputer, OpenAI once again partnered with NVIDIA, now with DGX Spark, the smallest AI supercomputer ever made, claimed by Nvidia..
Claimed by NVidia to be the smallest AI supercomputer, what are the specifications of DGX Spark itself? Here are the details:
Size
– Chassis Type: Small form factor (SFF)
– Dimensions: 150 mm (L) x 150 mm (W) x 50.5 mm (H)
– Weight: 1.2 kg (2.6 lbs)
Hardware Details
– NVIDIA Grace Blackwell architecture with integrated GPU and CPU
– 20-core Arm processor with high-performance cores
– 128 GB unified system memory
– Compact desktop form factor (150mm x 150mm x 50.5mm)
– Advanced connectivity including Wi-Fi 7, 10 GbE, and ConnectX-7
– Support for AI models up to 200 billion parameters (or 405B for dual-Spark configuration)
| Component | Specification |
|---|---|
| GPU | NVIDIA Blackwell Architecture with 5th Generation Tensor Cores, 4th Generation RT Cores |
| CPU | 20-core Arm processor (10 Cortex-X925 + 10 Cortex-A725) |
| Memory | 128 GB LPDDR5x unified system memory, 256-bit interface, 4266 MHz, 273 GB/s bandwidth |
| Storage | 1 TB or 4 TB NVMe M.2 with self-encryption |
| Network | 1x RJ-45 (10 GbE), ConnectX-7 Smart NIC, Wi-Fi 7, Bluetooth 5.4 |
| Connectivity | 4x USB Type-C, 1x HDMI 2.1a, HDMI multichannel audio |
| Video Processing | 1x NVENC, 1x NVDEC |
Rear Panel
-Power button
-4x USB Type-C (one for power delivery)
-1x HDMI 2.1a display connector
-1x RJ-45 Ethernet connector (10 GbE)
-2x QSFP Network connectors (ConnectX-7)
Compute Performance
-AI Compute: Up to 1,000 TOPS (trillion operations per second) inference and up to 1 PFLOP (petaFLOP) at -FP4 precision with sparsity
-CUDA Cores: 6,144
-Copy Engines: 2 (enables simultaneous data transfers to and from GPU memory, improving throughput for -AI workloads)
-CPU Performance: 20 cores (10 Cortex-X925 + 10 Cortex-A725)
-Memory Bandwidth: 273 GB/s
-Memory Channels: 16 channels (256 bit) LPDDR5X 8533
AI/ML Capabilities
-Model Support: AI models up to 200 billion parameters
-Tensor Performance: 5th Generation Tensor Cores with FP4 support
-Framework Support: TensorFlow, PyTorch, TRT-LLM, and other AI frameworks
-Use Cases: Inference, deployment, and fine-tuning of large language models
Where can I buy DGX Spark?
You can purchase the DGX Spark through the official Nvidia website by clicking here, or you can also buy it through brands partnered with Nvidia, such as Dell, HP, Lenovo, Asus, MSI, GIGABYTE, Acer, and also Micro Center locations in the United States.
Maybe you would like other interesting articles?