AMD has released its Epyc Bergamo processors for the cloud market. Optimized for dense compute workloads, it uses the Zen 4C core architecture with a stripped-down L3 cache to cram twice as many cores. Unlike Intel’s E-cores, Zen 4C and Zen 4 support the same instructions (including AVX512) thanks to a common ISA (Instruction Set Architecture). And since everything other than the L3 is identical, the performance is mostly the same.
A V-Ray 5 benchmark of an Epyc 9754 processor duo highlights this. The two 128-core Epyc 9754 CPUs score 221,018 points in a 2S configuration, making it the fastest chip combo in 3D rendering. The 96-core Epyc 9654 (Genoa) duo nets 147,975 points, making it the second-fastest configuration. In this case, a single Epyc 9754 is nearly as fast as two Epyc 9754 parts. This is because rendering workloads are heavily compute-oriented, unaffected by the higher-level cache.
Intel’s Xeon Platinum 8490H (Sapphire Rapids-SP) combo, with 60 cores each, scores a paltry 93,059 points in V-Ray 5. The Epyc 9754 duo is nearly 2.4x faster in the same workload, courtesy of the intense compute density. You can expect similar speedups in heavily threaded cloud applications with Bergamo. Arm offerings from Amazon, Ampere, and Google can expect a similar thrashing in these workloads.
Source: HXL.
