that’s a statement refuted by the article in which it appears, though. The Ampere system is clearly not suited to latency-sensitive application services, unless you’re just looking to double your latency for no reason. If you looked at these specjbb results because you’re shopping for Java backend servers then you’re buying Xeon yet again.
It's not refuted, because Ampere isn't the only competitor. The statement is that with AMD EPYC and Ampere Altra as options, Intel Xeon loses everywhere.
I don't really think that's a supportable statement. The Xeon's outlier latency on specjbb is superior until it hits its throughput limits. A lot of machines are bought on the basis of latency, not throughput.
Also you really have to grind your teeth to get past AnandTech's habit of comparing, on a performance-per-dollar basis, just whatever CPU that happen to have laying around. The entire reason the Xeon Platinum costs $10000 is because it scales to 8 sockets, a level needed only by people who have backed themselves into a corner with Oracle or SAP and who now need a gigantic single host at any price. Anyone who was actually planning to build a 2-socket server would choose the Xeon Gold 6258R with the same core count, the same cache, higher clocks, and less than half the price. Suddenly when you make a comparison to a comparable product, you find that Intel has the cheapest product of the three, along with certain favorable performance aspects.
