NVIDIA DGX Spark
LENOVO ThinkStation PGX
Simple & Accessible AI Development
NVIDIA® GB10 Grace Blackwell Superchip Integrates a 20-core Arm CPU and Blackwell GPU architecture delivering up to 1 petaFLOP (1000 TOPS) of FP4 AI compute.
128 GB Unified System Memory Ultra-fast LPDDR5X-8533 memory running on a 256-bit bus with 273 GB/s bandwidth, allowing local execution, prototyping, and fine-tuning of AI models up to 200 billion parameters right at your desk.
Preloaded Full-Stack AI Platform Ships with NVIDIA DGX™ OS along with pre-installed CUDA 13, TensorRT, PyTorch, Jupyter Notebooks, and AI Workbench for frictionless, sandbox AI development.
Ultra-Compact & Efficient Weighs just 1.2 kg with a tiny 150mm x 150mm x 50.5mm footprint and a modest 240W power consumption.
High-Speed Multi-Unit Scaling Features 10GbE networking alongside dual 200Gbps QSFP ports (ConnectX-7 NIC) to directly link two ThinkStation PGX systems and handle models up to 405 billion parameters.
