Qualcomm today unveiled the first 10nm server CPU/SoC called the 48-core Centriq 2400. If you were wondering when the first no-compromise ARM server chip would be out in the wild, now you have an answer
SemiAccurate thinks this is really a milestone but there was precious little technical data this time around, the real announcements were the name, the process node, and the core count. Starting with the name, Centriq was announced earlier as was the first generation part now called the Centriq 1400. That 24-core device was meant as a test chip, not for production, think software testing, bringup, and hardening. Xylinx and Mellanaox were also involved for reasons you will probably see in the 3rd generation parts.
Nothing to see at the bottom
One big difference between the two generations is the core count but what was not directly stated at the launch was details about the cores themselves. 1400 was based around Hydra or at least the server variant of it. Before you start having ideas about Qualcomm pulling in a phone SoC to server duties, remember that the 1400 was a development and test chip, not anything close to a production part. The 2400 has a completely new core that has nothing to do with Hydra or the 1400.
This one is called Falkor, that is both the code name and the marketing name. Refreshing eh? Back to the point, Hydra != Falkor, they have nothing in common, completely new, different designs. Once again, nothing in common, really. We realize this won’t stop the conspiracy theorists and astrotrufing from the usual suspects, but they would be wrong, trust us on this one.
What process is the 2400 on? Qualcomm was being a bit coy when they said 10nm but refused to say which 10nm process it was. When asked they pointed out that the Snapdragon 835 was on Samsung’s 10nm flavor but that does not mean the 2400 is on the same process, completely different core if you recall. More below but officially it could be anyone this time around.
Qualcomm demo’d the Centriq 2400 by pulling in and searching Twitter feeds, all of it not a single person, searching the hashtags, and putting them on a map with any related YouTube videos correlated to both. It is a lot of work but very well suited for a silly core count SoC. Luckily Qualcomm has just that to do the work. The boxes look like this but Qualcomm didn’t show them open, just the closed cases.
Qualcomm 1U server for Centriq 2400
Now lets move on to a caveat, a big one. That would of course be process node of 10nm. Qualcomm is comparing foundry nodes to Intel’s process nodes which might have held water in the past but now it is just a marketing game. This isn’t to say Qualcomm has any part of the process naming debacle of late, just that you can’t compare a foundry 14nm process to Intel’s 14nm without giggling. We would rate Intel’s 14nm process to be on par with foundry 10 or even some 7nm nodes, closer to 10nm than 7 though. Intel is still two years ahead or more in terms of real technology but marketing sadly rules this arena now.
Then there is the shipping versus announcement part of the launch. Qualcomm is launching the Centriq 2400 now but it won’t ship in production volumes to customers until 2H/2017 or about 2-3Q from now. Intel will have announcement their Skylake-EP/Purley lineup by then and it will be for sale to consumers on the day of the launch. Barring any production slips between now and the two launches, Intel’s Skylake-EP will be out first but it is ‘only’ 14nm.
On paper, Qualcomm unquestionably wins the race to the first 10nm server SoC. From our more technical point of view, our jaundiced eye would call it much more of a tie than either side is comfortable with. Think about that, Qualcomm’s first production ARM server SoC could tie Intel on tech, that is one hell of an achievement. All of this isn’t trying to say a hard launch vs a soft launch is good or bad, while we prefer hard launches as long as the schedules are clear at announcement, it is just a difference in corporate styles more than anything substantial.
Lets look at a few tidbits that we have learned unofficially over the past few weeks about the Centriq 2400 now, starting with the process node.
Note: The following is for professional and student level subscribers.
Disclosures: Charlie Demerjian and Stone Arch Networking Services, Inc. have no consulting relationships, investment relationships, or hold any investment positions with any of the companies mentioned in this report.
Latest posts by Charlie Demerjian (see all)
- HyperX ships it’s 60 millionth enthusiast memory module - Oct 15, 2018
- Bittware/Nallatech water cools 300W of Xilinx FPGA - Oct 12, 2018
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018