In this excellent article Charlie Demerjian went over the Intel Cascade Lake Xeon launch from last week. He summed it up with:
“The problem for Intel is that in a vacuum, this is a clear win, with competition chomping at their heels it is not. 18+ months for a 6-8% performance increase is not a win in anyone’s book and is severely under the curve for expected increases from the company. The gap between Skylake-SP and Cascade Lake-SP is about the gap between Sky-SP and AMD’s current ‘Naples’ based Epyc CPU which are a fraction of the price.”
“Later this quarter AMD will release their ‘Rome’ based Epycs which the company has stated will double performance per socket over Naples. The same set of numbers which correctly called Cascade’s performance had Rome beating Cascade by 50%+. We can now reveal that those numbers had Rome coming in at ‘only’ a little more than 190% of Naples’ performance. The make or break thing for Cascade Lake is where AMD prices Rome, the rest is rounding error.”
This is a great article by Patrick Kennedy. AMD had been promoting the one socket Epyc (Naples) as a powerful yet cheap configuration, while Intel was pushing customers toward their two socket and thus more expensive solutions. Now belatedly Intel is offering their version of a heavily discounted one socket skus. Intel has acknowledged the AMD advantage here. However, Rome is coming out soon. Additionally, how will these discounts hurt Intel’s gross margin? Market share or gross margin, either way Intel will lose something.
“AMD still holds a clear benefit. Its single socket CPUs have more than twice the PCIe connectivity of a single Intel Xeon Gold U series CPU. For form factors like popular 1U 10x NVMe (plus NICs) or 2U 24x NVMe (plus NIC), the AMD EPYC solution still has a clear benefit.”