EdgeAI.computer

NVIDIA

NVIDIA DGX Spark

Desktop AI system built on the GB10 Grace Blackwell Superchip with 128 GB of unified memory in a 150 mm chassis.

Manufacturer pageAdd to comparison
AI Workstations
NVIDIA
Compact desktop AI supercomputer beside a monitor in a dim studio

Overview

DGX Spark places a Grace Blackwell Superchip on a desk: a twenty-core Arm CPU with an integrated Blackwell GPU of 6,144 CUDA cores, 128 GB of unified LPDDR5X memory at 273 GB/s, and ConnectX-7 networking for pairing two units. It is aimed at local development and inference of large models.

Typical use cases

  • Local large model inference
  • Model fine-tuning
  • Robotics and vision development

Deployment environment

Desk-side development environment or small on-premise lab

Key specifications

AI performance
Up to 1,000 TOPS inference, up to 1 PFLOP FP4 (sparse)
Processor
NVIDIA GB10 Grace Blackwell Superchip, 20-core Arm CPU (10× Cortex-X925, 10× Cortex-A725)
GPU
Blackwell GPU, 6,144 CUDA cores, 5th-gen Tensor cores
Memory
128 GB LPDDR5X unified, 256-bit, 273 GB/s
Storage and networking
1 TB or 4 TB NVMe M.2; 10GbE RJ-45, ConnectX-7 (2× QSFP), Wi-Fi 7
Power and size
240 W external PSU (140 W SoC TDP); 150 × 150 × 50.5 mm, 1.2 kg

Specifications taken from current manufacturer documentation and last checked on 2026-09-08.

Source: manufacturer documentation

Specifications are summarised from publicly published manufacturer material and are provided for orientation only. Always confirm current figures directly with the manufacturer before purchasing or designing in.

Related solutions

Compact desktop AI machine on a dark workbench
AMD

AMD Ryzen AI Max+ 395

Sixteen-core Zen 5 processor with an XDNA 2 NPU rated above 50 TOPS and up to 128 GB of unified memory.

NPU
XDNA 2, more than 50 peak AI TOPS
CPU
16-core Zen 5, 32 threads, up to 5.1 GHz boost
GPU
RDNA 3.5, up to 40 compute units
Deployment
Mobile workstation, mini PC or compact desktop
Local large model inference
Creative and engineering workstations
On-premise prototyping
View DetailsCompareWebsite
Professional AI workstation tower beside two monitors in a dim studio
Lenovo

Lenovo ThinkStation PX

Dual-socket workstation supporting up to 128 CPU cores and four NVIDIA RTX 6000 Ada GPUs for local model work.

CPU
Dual 5th Gen Intel Xeon Scalable, up to 128 cores in total
GPU
Up to 4× NVIDIA RTX 6000 Ada Generation
Expansion
Up to 9 PCIe slots
Deployment
Office, lab or on-premise engineering environment
Model fine-tuning
Local LLM inference
Dataset preparation and simulation
View DetailsCompareWebsite