Intel (INTC) is desperately trying to stem the tides of defection among server vendors moving to ARM (ARMH), but they are hamstrung by the company’s own market manipulation. The exclusion of competitors not only meant tepid chips that no one wants, it is now excluding the very partners that Intel wants.
The short story is that today, Intel held a press conference to announce that they discovered the niche of ‘microservers’, something the industry started making years ago, and there are several Atom based machines already on the market. Intel has had their ‘Bill Gates discovers the Internet’ moment, but it is too little, too late.
They want you to believe that Intel will take this market by storm, with the low end being taken by Atom, likely with ECC added, and the high end taken by clocked down Sandy or Ivy Bridge chips. It sounds good on paper, but there are several problems.
First, lets talk about the good, Intel’s Atom has an ISA that is common in servers, x86. This means the software stack that you run on a ‘real’ fat chip will run on the Atom. It should be familiar territory to server admins, more or less. The ISA is the good, singular, point here.
Moving on to the bad, we can start out with Intel’s legendary marketing games. The company has a penchant for fusing off features that are useful, needed, or both, and then trying to sell them back to the user under the pretense of yields. This pretense was true in the past, but has morphed into a naked, customer abusive money grab with no technical justification of any sort.
In the old days, you could take a chip with an error in the cache, and instead of a scrap chip with 512K of bad cache, fuse half of it off and sell it as a 256K cache Celeron. Fair enough, and it was a win/win for both Intel and the user. That all changed recently, with Intel fusing off just about everything under the sun, and selling it back piecemeal.
If you look at the current Core iSomethingmeaningless(TM)(R)(C)(P), you will see that some of the ‘justifications’ that add 2 to a CPU’s iMeaninglessMarks(TM)(R)(C)(P) are things like HT, Turbo, and even the AES-NI instruction set. This is in addition to the things that you actually bin for like clock speed, power use and cache. Intel is taking good functionality, and simply turning it off in order to force more money out of the customer by upselling them. Think of how you feel when you want to buy a car with air conditioning, but can’t without $900 floor mats, $300 paint sealer, and $700 upgraded engine bay chrome trim.
There is absolutely no technical justification for fusing off an ISA, none. You might have a painfully low split of chips that don’t have high turbo headroom, but these are much better sold as slower parts anyway, they are probably higher power consumption to begin with. Intel is simply using the technically correct idea of fusing things off, and using it as plausible deniability for abusive profit margins.
Going back to Atom, did you know that Atoms have full 64-bit capability? Did you know that Atoms have full vPro capabilities? How about full VT-x virtualization? It is in there, and has been from day one, but they won’t turn it on because they are/were afraid that the chips might be used in a way that cannibalizes their more lucrative Xeons. Want more than 2GB of memory for your server? That ‘capability’ was only allowed about a week ago with the launch of the Atom N570 parts.
This artificial, and downright stupid market segmentation opened the door for ARM servers to step in to, something that no one would have wanted if Intel hadn’t been so manipulative on the Atom front. Vendors don’t like the games, but have to shut up for fear of retribution on other fronts, so they smile in public, and start work on ARM based servers.
Latest posts by Charlie Demerjian (see all)
- Intel shows off 10nm 112Gbps SerDes - Mar 12, 2019
- Intel releases Compute Express Link spec - Mar 11, 2019
- Qualcomm rolls out a second gen 5G modem called X55 - Feb 19, 2019
- What is Intel’s Foveros tech and what isn’t it? - Feb 11, 2019
- Why SemiAccurate called 10nm wrong - Jan 25, 2019