The Incredible Performance & Power Efficiency Of AMD Zen 1 vs. Zen 4C

While we are beginning to see AMD Zen 4C cores in client systems, these smaller cores have already proven themselves very interesting and capable with the AMD EPYC Bergamo high core count server processors and the extremely power efficient EPYC 8004 "Siena" processors. For showing how far Zen has come in power efficiency, I thought it would be fun to show how the original flagship EPYC 7601 "Zen 1" processor with 32-cores / 64-threads compared to Zen 4C with the EPYC 8324P(N) 32-core processors. But as that isn't even the top-end Siena part, I also tossed in the 64-core EPYC 8534PN too for a top of stack look for the current EPYC 8004 line-up.

The AMD EPYC 7601 launched back in June 2017 as AMD's original flagship EPYC processor based on a 14nm process and featuring 32 cores / 64 threads with eight channel DDR4-2666 memory support, 64MB L3 cache, a base clock of 2.2GHz, and a boost frequency of 3.2GHz while having a 180 Watt TDP.

Fast forward six years, the EPYC 8324P Siena processor has 32 cores / 64 threads while having a 2.65GHz base clock and 3.0GHz boost clock while having a 180 Watt default TDP (configurable TDP 155 to 225 Watts), 128MB L3 cache, and DDR5-4800 memory support. The EPYC 8324PN for the NEBS-compliant model has a 2.05GHz base clock and 3.0GHz all-core boost clock with a 130 Watt default TDP. Since looking at the power efficiency too, I also carried out a run of the EPYC 8324P when running at its 155 Watt minimum cTDP value.

The 32-core Siena was the primary focus to match the 32 Zen 1 cores with the EPYC 7601. But for a top-of-stack look the EPYC 8534PN was also re-tested for that part with 64 cores / 128 threads with a 2.0GHz base clock and 3.05GHz all-core boost speed (3.1GHz maximum boost clock), 128MB L3 cache, and 175 Watt default TDP.

The EPYC 7601 and EPYC 8324P / 8324PN / 8535PN processors were all freshly re-tested on Ubuntu 23.10 with the Linux 6.6 kernel for a leading-edge look at the Linux performance. All processors were tested at their maximum number of memory channels / speed and other settings at their defaults unless otherwise specified (e.g. the 155W cTDP run). The CPU power consumption was also monitored and performance-per-Watt data collected using the PowerCap/RAPL interface.