Overview
PNY DGX Spark Cluster Bundle pairs two NVIDIA DGX Spark GB10 Grace Blackwell systems with a 200Gb/s QSFP56 interconnect cable, enabling NVLink-C2C scale-out. Each GB10 delivers 1 petaFLOP AI with 128GB LPDDR5x unified memory on a 20-core ARM CPU. Together, this two-node cluster provides 2 petaFLOPS and 256GB combined for 200B+ model training.
Key Features
- 2x NVIDIA DGX Spark GB10 Grace Blackwell systems
- 1 petaFLOP per unit — 2 petaFLOPS combined
- 256GB LPDDR5x unified memory across both nodes
- 200Gb/s QSFP56 interconnect (IB HDR) included
- DGX OS pre-installed on both units
Ideal For
Research teams and enterprises who want desktop-scale 200B+ parameter training and inference without datacenter infrastructure, using NVIDIA's Grace Blackwell architecture in a two-node desktop cluster configuration.
DGX Spark/GB10 System
2 x PNY Nvidia DGX Spark Grace Blackwell Architecture 20 core Arm10 Cortex-X925 + 10 Cortex-A725 128GB LPDDR5x 1000 AI TOPS DGX OS 4TB NVME.M2 10 GbE Wi-Fi 7 BT 5.3
QSFP 200GBe Cables
1 x NVIDIA MCP1650-H001E30 DAC Cable IB HDR up to 200Gb/s QSFP56 1m - 3.28 ft QSFP56 Network Cable for Network Device First End: 1x SFF-8665 QSFP56 Network - Second End: 1x SFF-8665 QSFP56 Network
$8,049.93
Prices may vary. Verify on vendor site.
Quick Specs
- DGX Spark/GB10 System
- 2 x PNY Nvidia DGX Spark Grace Blackwell Architecture 20 core Arm10 Cortex-X925 + 10 Cortex-A725 128GB LPDDR5x 1000 AI TOPS DGX OS 4TB NVME.M2 10 GbE Wi-Fi 7 BT 5.3
- QSFP 200GBe Cables
- 1 x NVIDIA MCP1650-H001E30 DAC Cable IB HDR up to 200Gb/s QSFP56 1m - 3.28 ft QSFP56 Network Cable for Network Device First End: 1x SFF-8665 QSFP56 Network - Second End: 1x SFF-8665 QSFP56 Network
Tags
llm-inferencemodel-traininghpc