There is a new front in Intel’s war with everyone, customers not withstanding, and that is the so called microserver market. Intel had no answer for the ARM based devices from Marvell, Calxeda, and others until earlier this week when they admitted the existence of Centerton.
The silicon for the ‘new’ Centerton Xeon S1200 is actually not new at all, it is just the plain old, delayed, buggy, and generally mocked Oak Trail Atom with a new badge. And price tag. Intel won’t actually release real specs on the consumer variant of the CPU, and the same holds true for their new one. The press conference where Intel announced Centerton/S1200 was devoid of anything but fluff and self-congratulatory pabulum, in other words, typical Intel of late.
One thing that make us sit up and take notice however was the claim that Intel made for being, “the World’s First 6-Watt Server-Class Processor”. The 6W part is pretty obvious, the S1200 SoC has far higher energy consumption than all of it’s ARM based competition, and that doesn’t include the chipset. Intel Atom chipsets of late have been far more power hungry than the CPU itself, so 6 Watts is only the start. That is not a big deal, but it does explain quite a bit about Intel’s reticence to give out any technical details.
One can’t say the same about the claims surrounding the server-class processor portion however, that is so far past disingenuous that you have to question Intel’s desperation in trying the line. Why? Well, they are not the first server-class processor in that wattage range, unless you consider higher to be a virtue, which even Intel doesn’t in their presentation. They seem to be hinging the notion that server-class=64-bitness, and while it is a good thing for servers, it is hardly unique. S1200 is basically a tarted up widget SoC that Intel decided not to turn off ECC and 64-bit ISAs on, but it otherwise lacks the necessary server basics.
Marvell and Calxeda on the other hand have purpose built server parts that have all the connectivity, expansion, and controllers that Intel’s lacks. The Atom/Centerton parts have a crippled memory interface, purposefully hobbled PCIe links, and many necessary server features just not available. Why? To hobble the competition, PCIe was hamstrung to shut Nvidia out, and most of the rest was done to keep netbooks constrained to the low end of the market. In short, Cedar Trail was designed to thwart OEMs from doing the very things Intel is promoting Centerton as in order to protect big chip margins.
Calxeda is rightfully annoyed by these claims and put up a blog about it. Marvell seem to simply not understand the press, so we have no idea what they think, but you can be sure that they are not pleased with the terminology. That said, you can buy systems with both Calxeda and Marvell CPUs in them, and you could do it for months before Intel decided to claim the title of world’s first. Would it be gratuitous to point out that both of the competitors have all the server components that Intel lacks on board? The 5W number for a Calxeda node includes 4GB of DDR3 in the tally too, Intel does not in their 6W number.
On the hardware front, Intel is blown out of the water on performance per watt, they are painfully out of line here. Performance per watt is the only measure that matters in the microserver market, if your software is amenable to a 32-bit memory space. Before you dismiss this number, it is quite simply a make or break metric for most microserver customers, it is either a non-issue or a complete dealbreaker for deployment, there is really no middle ground. SemiAccurate went in to the problem at length here, and the same holds true for any 32-bit ARM device.
That would be all of them, full stop. The workhorse A9 class chips are pure 32-bit, the newish A15 cores have a memory controller that ups the total significantly, but each thread can only see 32-bits of memory. In short, A15s may expand the TAM a tiny bit, but they aren’t anything close to a fix. The A64/V8 cores that will come out in 2013 completely remove this restriction, and put the Atom/Centerton parts to shame for memory capacity, but they are not close to release yet. Intel is right in this regard, but only if you need that capability.
Many do, many don’t, and several companies have a thriving business built on the don’t category. That said, Intel does just about the minimum possible to exploit their 64-bit capabilities, the S1200 is capped at a mere 8GB, twice what A9s are limited to, and less than what some A15 implementations can see. What could have been a clear advantage for Intel is once again artificially hobbled to protect margins of their other devices.
So where does that leave us? The author feels that the Centerton S1200 is just not a competitive part. They address a market that the ARM vendors can’t for another 12 months, but that is about it. If you need large memory spaces, Centerton can’t deliver unless large means above 4GB and less than 8GB. An 8.0000001GB memory space requirements mean the Intel parts are faced with the same problem that ARM has, how much TAM do you think that adds?
In just about every other category, the Intel chips are woefully out of contention, power and performance per watt key among them. This makes the launch seem more like a PR move than a product move, and the paucity of details reinforces this. Intel has their back to the wall, and it is showing.
The upcoming 22nm Avaton part doesn’t change this either, it fixes none of the fatal flaws that Intel has, and the competition will have surmounted theirs before Avaton is available. x86 is not a factor in this space either, the microserver market is dominated by FOSS software that is ISA agnostic and just as at home on ARM chips as x86. Intel has no answer so they are handwaving, banging a drum, and trying to distract anyone not technical enough to see through the mounds of fluff.
That brings up one big question, what do you think? Beauty is in the eye of the beholder, and so are the deal breaking metrics. Do you think the title of “the World’s First 6-Watt Server-Class Processor” is fair, much less deserved on this part? Are the ARM vendors simply whining? Please discuss, we really do want to hear your opinion.S|A
Latest posts by Charlie Demerjian (see all)
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016
- Intel’s Xpoint is pretty much broken - Sep 12, 2016
- ARM adds 2048-bit vectors to v8A with SVE - Sep 7, 2016