Last August SemiAccurate compared the performance of Intel’s upcoming Ice Lake-SP server to AMD’s Milan with some leaked benchmarks. Our conclusion then was that, “Intel should not launch Ice Lake-SP“.
Our previous analysis used pre-release Ice Lake samples and Milan samples near release. Since then a lot has changed on both sides. The numbers, on paper anyway, for Ice-SP have gotten a bit better as you will see in the SKU list below, and Milan uses a bit more energy than that test, but only by about 5W.
Given where that initial data came from, we strongly suspect those numbers listed earlier will be a bit closer to real world energy use than the official numbers. A good example of this is Intel’s recent Rocket Lake consumer CPU that has a 125W TDP but independent testing by Anandtech saw nearly 300W.
The core counts have changed, again on paper, SemiAccurate isn’t confident that at least two of the Ice-SP SKUs are manufacturable in anything more than PR quantities. That said the long and rather tortured gestation of Ice Lake-SP is a tale for another day, today we focus on how well the final SKUs perform using SpecIntRate 2017 numbers from a biased source, Intel. Even with that in mind the performance backs up our earlier claim and then some, it is bad.
Note: The following is for professional and student level subscribers.
Disclosures: Charlie Demerjian and Stone Arch Networking Services, Inc. have no consulting relationships, investment relationships, or hold any investment positions with any of the companies mentioned in this report.