Intel’s next-gen Xeon Scalable processors are all set to launch in the coming months. Codenamed Sapphire Rapids-SP, these MCM CPUs will be the first to feature on-die HBM2e memory and multiple memory modes (Flat, Cache, HBM-Only). Team Blue is promising massive performance uplifts with its 4th Gen Xeon SP chips. However, the repeated delays are likely going to cost Intel dearly. AMD is prepping its 96 core Genoa parts (Zen 4) for late 2022 and the 128-core Bergamo (Zen 4c) offerings for an early 2023 release. As such, Sapphire Rapids will face tough competition from the get-go.
Now, coming to the leaked benchmarks, the numbers are all over the place. The standard synthetic benchmarks such as CPU-Z, Cinebench, y-cruncher, and V-Ray show the 64-core Epyc Milan/Milan-X processors in the lead but purely vector workloads (AVX256/512) show healthy gains for the Sapphire Rapids flagship.
In the end, unlike the consumer market, the prices and target performance in the server market vary from customer to customer with different use cases. A metric that is useful to one client is going to be redundant to the next. Workloads vary from segment to segment as well as company to company. The profit margins here are fat, and large conglomerates are willing to pay large sums to get the best performance for their specific applications. Therefore, these benchmarks are more or less useless at the moment.