As we get closer to the end of the year, product releases tend to slow down. However, that trend has been ignored by Intel. This is because the company is trying to combat a resurgent AMD product portfolio that could well nuke a lot of market share, particularly in the lucrative server arena. To that end, Intel has launched Cascade Lake Advanced Performance (AP) Xeons with up to 48 cores. Will this be enough to combat the soon to be released third round of Epycs (codenamed Rome) from AMD?
The new Cascade Lake AP Xeons are the lead announcements. They feature 48 cores (and presumably 96 threads) in a multi-chip configuration. Ironically, Intel pooh-poohed this arrangement as being simply “glued together” recently.
Given the specs provided, it’s likely that the new chip will be a monster. The headline CPU will likely be based on the current Intel Xeon Platinum 8186, which has 24 cores and 48 threads, runs at 2.7GHz with 3.7GHz boost and has a power rating of 205W. If Intel is to use this Xeon as the basis for the multi-chip Cascade Lake AP, then expect a number changes, including frequency to drop in order to manage power. Otherwise, you’re looking at a CPU that consumes more than 400W. This is all speculation however, as Intel didn’t release too much information on the newly announced parts.
Intel confirmed that two of these CPUs can be installed per server, making for 96 cores and 192 threads in total. This would be a massive increase in the number of threads available, and Intel claims Cascade Lake AP is up to 3.4 times faster in Linpack and up to 1.3 times faster in Stream Triad than AMD’s (current) Epyc. Against its own chips, Intel claims Cascade Lake AP would be up to 17 times faster in DL Inference than the first Xeon Platinums at launch.
Shipment of the initial CPUs to customers will begin in early 2019. The rest of the Cascade Lake CPUs will also be launched at around that time, with (hopefully) more details on precisely what will make the chips tick.