What Is NVLink-C2C?_
NVLink Chip-to-Chip (NVLink-C2C) is the high-bandwidth interconnect between NVIDIA Grace CPUs and GPUs within a single module. NVLink-C2C provides 900 GB/s bidirectional bandwidth, eliminating the PCIe bottleneck that exists in traditional CPU+GPU server architectures. This is an internal connection with no external cabling requirements.
Technical Details
NVLink-C2C is a chip-to-chip interconnect that operates at the package level, connecting the Grace CPU die to the GPU die within a single module (GH200 Superchip) or across closely coupled modules in compute trays (GB200/GB300). The 900 GB/s bandwidth is approximately 7x what PCIe Gen5 x16 provides, enabling the CPU and GPU to share a coherent memory space efficiently. For infrastructure teams, the key implication is that NVLink-C2C is entirely internal — there are no cables, connectors, or external components associated with this interconnect. It does, however, change performance characteristics: workloads that were previously CPU-GPU memory bandwidth limited can now run more efficiently, which may increase sustained GPU utilization and power draw.
How Leviathan Systems Works with NVLink-C2C
While NVLink-C2C is an internal interconnect requiring no physical cabling, Leviathan Systems validates its functionality during commissioning to ensure the CPU-GPU link is operating at full bandwidth.