Intel Confirms HyperThreading Returns Starting With Coral Rapids In 2028
Speaking to industry analysts on Intel's earnings call yesterday, Tan was asked how he intends to regain marketshare in the server space versus AMD and Arm. His response was kind of rambling and didn't really answer the question, but in the first half, he said:
Clearly, on the server side, data center side, I think we have a strong roadmap. We have Clearwater Forest, Diamond Rapids, and also Coral Rapids with the SMT involved. I think we continue working on improving the competitiveness against our competitor. I think clearly it’s very important to drive the improvement in the single thread and also multithreading. Multithreading will be coming in the Coral Rapids.
Hopefully multi-threading is coming in all of your CPUs, Mr. Tan! But kidding aside, he clearly meant Simultaneous Multi-Threading, or, of course, Hyper-Threading, answering the question we asked almost exactly a year ago: which CPUs? It's definitely strange that Intel is dropping Hyper-Threading for a single generation of Xeon processors. This warrants explanation, so let me explain.
For the last two generations of Xeon processors, there are fundamentally two categories. Aside from a few desktop chips that have been rebadged as Xeons, Intel doesn't sell Xeons with hybrid cores, so the two categories are Xeons with P-cores and Xeons with E-cores. Intel's "E-cores" haven't had Hyper-Threading since Saltwell in 2012, but the P-cores always have, and indeed, the company's current-generation Granite Rapids CPUs keep Hyper-Threading, but it's going away on the next-gen Diamond Rapids chips, only to return in 2028.
We don't know a lot about Coral Rapids yet. As far as anyone knows, they're planned for 2028, and they're going to start with chips that have 8-channel memory, although that could be changed depending on the reaction to Diamond Rapids (which is only going to have the big 16-channel chips.) They're bringing back hyper-threading, but it's not really known what core architectures the chips will use nor exactly which fabrication process they'll be on; most likely either Intel's 14A or a late-stage 18A-P refinement.
Hyper-Threading was removed in Intel's consumer CPUs because when you have some ten cores on even entry-level CPUs, it's a net loss. The extra silicon that goes into supporting SMT, which is significant, is largely wasted most of the time as most client users don't really run the kinds of heavy, throughput-oriented tasks where Hyper-Threading helps out. Removing it saves power and die area as well as granting a slight improvement to single-threaded performance due to the elimination of thread contention for CPU execution resources.
With that in mind, we'd be surprised to see it come back to Intel's client CPUs any time soon. However, server customers absolutely do enjoy the benefits of Hyper-Threading, which can offer as much as a 30% boost in multi-core performance in medium-load scenarios. Intel's absolutely going to need a boost like that if it wants to compete with the monster processors that AMD just unveiled yesterday.
