草榴社区

The growth of cloud data is driving an increase in compute density within both centrally located hyperscale data centers and remote facilities at the network edge. The increase in compute density is leading to demand for more energy-efficient CPUs to enable increased compute capability within the power and thermal budget of existing data center facilities. The demand for more energy-efficient CPUs has led to a new generation of server CPUs optimized for performance/watt.

This same increase in data volume is also driving demand for faster server interfaces to move data within and between servers. Movement of data within the server can be a major bottleneck and source of latency. Minimizing data movement as much as possible and providing high-bandwidth, low-latency interfaces for moving data when required are key to maximizing performance and minimizing both latency and power consumption for cloud and HPC applications. To improve performance, all internal server interfaces are getting upgrades:

? DDR5 interfaces are moving to 6400 Mbps

? Doubling the bandwidth of PCIe interfaces as they move from PCIe 4.0 at 16GT/s to PCIe 6.0 at 64GT/s

? Compute Express Link (CXL) provides a cache coherent interface that runs over the PCIe electrical interface and reduces the amount of data movement required in a system by allowing multiple processors/accelerators to share data and memory efficiently

? New high-speed SerDes technology at 56Gbps and 112Gbps using NRZ and PAM4 encoding and supporting protocols enables faster interfaces between devices including die, chips, accelerators, and backplanes