Intel today detailed some very interesting “benchmarks” of its upcoming 10nm Ice Lake-SP, despite the fact that it has repeatedly asked reviewers to not base their verdict based on “benchmarks”. Regardless, as per Intel’s internal tests, a 32-core Intel Ice Lake CPU manages to beat its Epyc Rome rival which features twice as many cores.
However, as with every other marketing slide ever, there’s a big catch. All these benchmarks leverage the AVX512 library which are much slower on non-Intel chips. Furthermore, there are only three benchmarks that Intel decided to show, with no general server workloads.
While the performance gains over Cascade Lake are quite decent, this is a product Intel should have launched a year ago when AMD’s Rome debuted. Now when the latter is about to launch its second 7nm based platform in the form of Milan, Intel is finally getting to launching its first 10nm lineup.
Not much else to say here. Ice Lake-SP will be either on par or slower than the Rome refresh and a fair deal slower than Milan in almost every benchmark. Not really a great start for Intel’s 1st 10nm volume product.