Insiders YuuKi_AnS divided Learn more about system performance based on two technical examples of the 4th Gen 48-core, 96-thread Intel Xeon Scalable server processor (Sapphire Rapids) in various synthetic benchmarks. Previously, the same source reported on the performance of these processors in AIDA64 and Cinebench R15 tests.
An insider compared a system based on two Sapphire Rapids processors with systems based on two Xeon Platinum 8380 (Ice Lake-SP) and two AMD Epyc 7773X (Milan-X). Note that the tested 48-core chips, unlike the latter, are not flagships – 56-core chips will be such. Also note that samples from Sapphire Rapids were tested with AVX-512 instructions disabled. According to insiders, support for this guide is offered as an additional option for a fee in this model. It is also noted that the BIOS for the motherboard on which the Sapphire Rapids chips were tested is not yet optimized.
The AMD Epyc platform used DDR4-2800 RAM. Remember that they officially announced support for DDR4-3200. The informant did not provide any information about the storage standard used by the Sapphire Rapids and Ice Lake-SP platforms. However, we would like to point out that Sapphire Rapids officially supports the DDR5-4800 standard. The difference in the memory standard used should hardly be noticeable in Cinebench tests, but can affect the result in other cases, as can be seen from the CPU-Z and V-Ray benchmarks. System performance tested in benchmarks Cinebench R20Cinebench R23, V-Ray, CPU-Z and AIDA64.
As you can see in the image above, the results are mixed. The difference between Ice Lake-SP and Sapphire Rapids in Cinebench R23 single-core and multi-core tests was minimal. Under ideal conditions, a system based on two Sapphire Rapids processors could score closer to 80,000, which is much closer to the AMD Epyc 7773X. However, this could be prevented by an unoptimized Intel platform BIOS firmware. As expected, the system based on two Milan-X performs better in tests of L2 and L3 cache performance since each CPU is equipped with 768 MB of 3D V-cache buffer memory. At the same time, AMD platform main memory stack latencies were higher than the competition – 126 ns versus 106.4 ns for Sapphire Rapids and 84 ns for Ice Lake-SP.