AMD EPYC Genoa 96-Core CPUs Benchmarked, Up To 17% Faster Single & 28% Faster Multi-Threaded Performance
The AMD EPYC Genoa CPU is the same sample that’s been doing the rounds over the internet for the past couple of months. The EPYC 9000 CPU bearing the OPN ID “100-000000997-01” features 96 cores based on the Zen 4 core architecture and 192 threads. The CPU was apparently tested on a motherboard known as the Suma 65GA24 which features 768 GB of DDR5 memory. Once again, we see that the CPU has been listed with an incorrect cache configuration. While the L2 cache is correctly listed at 1 MB per core or 96 MB in total, the L3 cache should be 32 MB per CCD for a total of 384 MB and not 192 MB as listed by the software. There are 12 CCDs so that’s detected correctly. In terms of clock speeds, the chip was running at a frequency of around 3.51 GHz and the detailed JSON shows a 3.49-3.50 GHz clock speed averaging across the cores. We don’t know what the max or base frequencies for the chip are but an EPYC Milan 7763 which is used for comparison to EPYC Genoa, has a clock speed rated at 2.45 GHz base and 3.5 GHz boost. We used a dual-socket result since the AMD EPYC Genoa was also running in a 2P configuration. That’s a total of 192 cores and 384 threads. In terms of performance, the 96-Core AMD EPYC Genoa CPU scored 1460 points in single-core and 96,535 points in the multi-core tests. The AMD EPYC 7763 “Milan” CPU scored 1249 points in single and 75,539 points in the multi-core tests. That’s an improvement of 17% in single and 28% in multi-threaded workloads. Now both figures are very close to the >15% single-threaded and >35% multi-threaded performance gains that AMD has touted for its Zen 4 cores vs Zen 3. We can’t say if Geekbench 5 is optimized enough to support up to 384 threads that the AMD EPYC Genoa chip has to offer but looking at the clock speeds, it looks like we may have a misreported number here too because the >15% single-threaded gain that AMD talked about came from a mix of IPC and frequency (8-9% IPC gain alone). Here, both chips are shown to be running at almost the same identical clocks of around 3.5 GHz. So as you can tell, there are some issues with the way the benchmark is reporting clocks or it could be possible that the AVX-512 instructions on Genoa are being put to good use here but despite that, the performance gain over Milan is really good and we can expect even better results once the final product is out.