Intel by the end of this year should present the first family of 10nm Ice Lake-SP server processors. The results of testing a system with two processors of this family were noticed by a well-known network source with the alias Tum_Apisak in the Geekbench 4 benchmark. This allows at least to some extent imagine what can be expected from Intel’s first non-mobile 10nm processors.
The system tested consisted of two 28-core Ice Lake-SP processors for a total of 56 physical cores and 112 compute threads. The cache size of the third level for each chip was 42 MB. Each processor ran at a base frequency of 1.5 GHz, with a maximum turbo frequency of 3.2 GHz. Such low frequencies indicate that rather early engineering samples passed the testing. In any case, I would like to believe that the final versions of Ice Lake-SP will offer much more impressive frequencies.
The Ice Lake-SP benchmark scored 3424 points for single core performance, while the system’s multi-core performance scored 38,079 points. For comparison, one 64-core AMD EPYC 7442 scores 4398 and 35 492 points in these tests, respectively. That is, the difference is 28.4 and 7.3%, respectively. Only in the first case the difference is in favor of AMD, and in the other in favor of Intel. Note that the AMD processor has higher frequencies of 2.25 / 3.4 GHz, and also has 256 MB of L3 cache.
It is worth adding here that Intel Xeon processors of the Ice Lake-SP family will be able to boast of support for AVX-512 instructions, which AMD chips are completely devoid of. However, AMD often has a much lower cost, and in this case, it is unlikely that everything will be different. Finally, AMD chips offer a higher density: where Intel can offer a pair of 28-core chips, AMD is able to provide two 64-core processors.
And of course, do not forget that it is not very fair to judge the performance of processors based on a single test, especially when it comes to early tests of engineering samples. However, the results described above only fueled interest in Ice Lake-SP.