Initial performance benchmarking in VCF 9.1 environments indicates throughput and efficiency variations on Granite Rapids-AP and Granite Rapids-SP 4-socket platforms, including a decrease in packet processing (PPS) and an increase in CPU overhead in high-density synthetic IO tests vs previous generation Intel Xeon processors. Performance differences were most evident under high-intensity benchmarks and are expected to be minimal under standard workloads.
The latest Intel® Xeon® Processors employ scalable architectures to enable higher core counts. Depending on the series, these architectures introduce different topological considerations:
Within VCF networking environments, this behavior can affect data and interrupt delivery paths between PCIe devices and processor cores, potentially reducing packet processing efficiency and increasing CPU utilization in latency-sensitive scenarios.
Broadcom and Intel are jointly evaluating potential software optimizations to mitigate these networking performance artifacts.
Recommendation: Customers can benchmark the specific workloads that are candidates for the Intel® Xeon® 6900-Series (Granite Rapids-AP) or Intel® Xeon® 6700-Series (Granite Rapids-SP) processors prior to production deployment to ensure acceptable performance with VCF.