: This is the magical glue of the GB2 work cycle. Instead of forcing the CPU and GPU to talk across a restrictive PCIe Gen 5 lane (which tops out around 64 GB/s bi-directional), they communicate over a custom NVLink-C2C (Chip-to-Chip) connection. It delivers an astronomical 900 GB/s of bidirectional bandwidth , reducing latency by up to 7x compared to conventional x86 architectures. Core Mechanics: How the CPU and GPU Work in Unison
Geekbench 2 provides two scores. Look closely at your workload:
The GB2 work benchmark is based on a series of tests that evaluate a CPU's performance in handling multiple threads and processes. The test suite consists of various workloads, including: