Looks Like July 12 For Power10 Announcements, Maybe July 27 for Shipments
June 13, 2022 Timothy Prickett Morgan
Those of us who have been waiting for the entry and midrange Power10 machines to be announced and start shipping, and thus beginning a new hardware cycle and a new opportunity to get IBM i shops on current hardware and software, will not have to wait long.
According to the latest rumors, it looks like the announcements of the entry Power10 machines – to be known as the Power S1014, the Power S1022s, the Power S1022, the Power L1022, the Power S1024, and the Power L1024 – and the Power E1050 midrange machine will happen on July 12. Which, as we wrote in early May, is about when we expected them. As far as we know, these machines will ship on July 27, but there is an outside chance that they can ship earlier.
The word we hear from business partners and resellers who are familiar with the lineup is that these machines will not pack as many cores as you might be thinking a Power10 server would have based on the fact that the Power10 chip has 16 cores laid out on it. For the high-end Power E1080 machine launched last year, IBM had 15 out of the 16 cores available to do work, and knew that at least one of the cores was going to be a dud no matter how well Samsung did with its 7 nanometer processes. This is Samsung’s second server-class processor, the first being the “Telum” z16 processor that IBM shipped in April in its System z mainframe line. We expect the yields to be pretty bad, to be honest.
The point is, given the high performance of the Power10 architecture, with its new implementation of the instruction set, it just will not matter if only have of the cores on the Power10 die are working properly, as is the case with the entry Power S1014 single socket box that will be the logical follow-on for the Power S914 and Power S924 systems. It looks like the Power S1022s, a new machine in the Power Systems lineup, will be a two-socket machine with only eight Power10 cores per socket. The regular Power S1022 and Power S1024 machines, will have a maximum of 10 cores and 12 cores per socket out of the 16 total on the Power10 die, if the word we are hearing is true, and the L Series Linux-only variants, the L1022 and L1024, will have the same core counts but presumably lower prices on compute, memory, and storage to promote the use of Linux on Power.
The Power E1050 will be a dual-chip module, with two Power10 chips in a single socket, and up to four sockets in a single system image, but it looks like each chiplet die in the Power E1050 package has only 6, 9, or 12 of the cores on the 16-core die active. I know that IBM has said that the Power10 has had good yields, but we expected a lot better than this, if it turns out to be true.
We will know more in a few weeks, and will do the usual thorough analysis.