ailocalinference

← Glossary

NVLink-C2C

The ultra-fast coherent interconnect inside Grace Blackwell superchips

Hardware

In plain English

A teleporter between the CPU and GPU so they share one memory space without slow PCIe cable trips.

Technically

NVIDIA's chip-to-chip coherent interconnect delivering 900 GB/s bidirectional bandwidth between Grace CPU and Blackwell GPU. Provides hardware cache coherence across 128GB-748GB unified memory pools on DGX Spark and DGX Station.

Related terms

DGX SparkGB10 SuperchipUnified Memory Architecture (UMA)

Lessons that use it

More in Hardware

DGX SparkDGX StationGB10 SuperchipMikroTik CRS504-4XQ-INMikroTik CRS804-4C-INMikroTik CRS812-8C-4XQ-INRoCEv2 (RDMA over Converged Ethernet)VRAM