Today, we are talking about the new Nvidia DGX Spark Mini AI Supercomputer because it would be a mistake to call it a workstation. In fact, Nvidia has made such a small yet powerful AI supercomputer. Nvidia had first made its first mini supercomputer in 2016, which it named Nvidia DGX1. Nvidia had spent billions of dollars to develop it, but no one was ready to buy it because the only thing that sells the most from Nvidia is its gaming GPU. But now is not the time; AI technology is at its peak, and everyone is getting into AI development, but to work on AI, you need a powerful special AI computer. Nvidia has fulfilled the needs of AI developers by launching DGX Spark. This small machine is a powerhouse.
Image Source: Nvidia
Table of Contents
GB10 Grace Blackwell Superchip:
powered by the GB10 Grace Blackwell Superchip, which combines a next-generation Blackwell GPU with fifth-generation Tensor Cores and FP4 (4-bit) precision with a 20-core Arm CPU (10x Cortex-X925 + 10x Cortex-A725).
The Nvidia GB10 Grace Blackwell superchip delivers up to 1 petaFLOP of AI performance (1,000 TOPS) using FP4/sparsity.
Image Source: Nvidia
You might not know it, but MediaTek has collaborated with Nvidia to develop the CPU cores of the NVIDIA GB10 Grace Blackwell Superchip, meaning the 20 Arm cores that you get, consisting of 10x Cortex-X925 + 10x Cortex A725, are courtesy of MediaTek, as Nvidia thought it appropriate to take MediaTek’s help in designing Arm-based CPUs.
Image Source: Nvidia
Nvidia DGX Spark Memory:
Nvidia DGX Spark is equipped with 128 GB of LPDDR5X coherent unified system memory, enabling local handling of large generative models—up to 200 billion parameters. It comes with 4 TB NVMe storage.
Nvidia DGX Spark OS & Software:
Nvidia DGX Spark comes with its own DGX OS and is also preinstalled with the full NVIDIA AI software stack, RAPIDS, frameworks like TensorFlow/PyTorch, NGC containers, pretrained models, Blueprints, NIM microservices, and AI Workbench. Seamless migration to DGX Cloud or larger infrastructure is supported.
Nvidia DGX Spark Availability:
You can buy Nvidia DGX Spark from Nvidia’s official site, and Asus, Dell, Gigabyte, HP, Lenovo, and MSI are launching their mini AI workstations with the Nvidia GB10 Grace Blackwell Superchip soon. Just Acer launched its powerful mini AI Workstation with this superchip—the Acer Veriton GN100
Image Source: Nvidia
What is the use of Nvidia DGX Spark?
The DGX Spark is not necessarily intended for high-end production performance, but rather as a platform for development and workflow emulation. Larger DGX systems are intended to be mirrored by it; anything that utilizes Spark ought to function in production on extensive infrastructure.
Workloads on models with up to 405 billion parameters can be accomplished by networking two DGX Spark units (through the ConnectX-7 Smart NIC) to scale memory and compute capacity.
However, in certain LLM workloads, memory bandwidth (roughly ~273 GB/s) continues to be a possible bottleneck.
Nvidia DGX Spark detailed specifications:
| Category | Specification |
|---|---|
| Architecture | NVIDIA Grace Blackwell |
| GPU | NVIDIA Blackwell Architecture |
| CPU | 20-core Arm (10 Cortex-X925 + 10 Cortex-A725) |
| CUDA Cores | NVIDIA Blackwell Generation |
| Tensor Cores | 5th Generation |
| RT Cores | 4th Generation |
| Tensor Performance | 1 PFLOP |
| System Memory | 128 GB LPDDR5x, unified system memory |
| Memory Interface | 256-bit |
| Memory Bandwidth | 273 GB/s |
| Storage | 1 or 4 TB NVMe M.2 with self-encryption |
| USB | 4 × USB Type-C |
| Ethernet | 1 × RJ-45 (10 GbE) |
| NIC | ConnectX-7 Smart NIC |
| Wi-Fi | Wi-Fi 7 |
| Bluetooth | BT 5.3 |
| Audio Output | HDMI multichannel audio output |
| Power Consumption | TBD |
| Display Connectors | 1 × HDMI 2.1b |
| NVENC / NVDEC | 1 × 1 × |
| OS | NVIDIA DGX™ OS |
| System Dimensions | 150 mm (L) × 150 mm (W) × 50.5 mm (H) |
| System Weight | 1.2 kg |
Also Read: NVIDIA DGX Station: A Powerful AI Supercomputer for Your Desk!
Nvidia DGX Spark Price:
If we talk about the price, then you will get Nvidia’s DGX Spark mini AI supercomputer for 3,399 USD to 3,999 USD; however, Acer’s VERITON GN 100 with the Nvidia GB10 Grace Blackwell superchip has been launched, and you can buy it. Its starting price is 3,999* USD.
