THE FTC HAS spoken, and the words came down on Intel with a force that is harder and more nuanced than almost anyone realizes. The settlement over a lot of alleged ills is brutal to Chipzilla, and both Nvidia and Via come out big big winners.
The settlement can be found here, and it is a masterwork of outlining all the accusations against Intel, addressing them all, and doing so without ever saying what the problem was. In the end, Intel ends up with pried open technical docs, handcuffs, blinders, and a mother hen to cluck over them. With a baseball bat. And lots of caffeinated beverages.
While the penalties may be harsh, they are much better than the alternative, a long protracted and public court case. Intel admitted no wrongdoing, and leaves no open doors for lazy lawyers to lob errant grenades through. Anyone who still wants a piece of Intel will have to work for it. Intel did the smart thing in settling, an unquestionably correct call.
So, lets look at some of the subtleties, penalties, and what handcuffs Intel will be wearing. In addition to what is explicitly written, we will also attempt to delve into the background on some of the statements. Most of them directly address some of the allegations of Intel’s ill behavior of late in one way or the other.
Neither Intel, Nvidia, Via or anyone else in the complaint were consulted for their opinion, joy or sorrow about this document. It is solely the author’s opinion based on years of following the players, products, and industry that this settlement covers.
The first point of interest is Part II on page 9 of the PDF, and it deals with PCIe busses. This would require that Intel include an interface on all of it’s “Mainstream Microprocessor Platforms” to include a PCIe bus for at least the next six years. Intel may pick what version of PCIe it uses, but it may not do anything to the implementation that will “intentionally limit the performance or operation of any Relevant GPU”.
More on the technical aspects of this later, but this is a bone to Nvidia. Nvidia is deathly afraid that Intel will lock them out of the GPU market by coming up with a proprietary bus that they can’t legally interface with, or worse yet, no external bus at all. This ruling ensures that there is at least one standard interface that Nvidia can plug their GPUs into, should consumers still care about the company in a year or two.
This is problematic for Intel because it prevents them from doing anything interesting with x86 SOCs. The whole point of an SOC is that you don’t need any external cards to get the intended job done, it is all on one chip. This is quite efficient and cost effective, but now Intel has to include a PCIe bus, the attendant pins, and likely route it on reference designs. For a desktop chip, this is mostly irrelevant, but it could hurt them badly on the Atom side. The FTC definition of Mainstream Microprocessor Platform includes Atom, so this could hurt that line.
Then again, there are two holes big enough to drive a truck through. The first one is II. A., it says, …”to a standard PCI bus”, and II. B. says “Standard PCI Express Bus”. PCI – no e – is a standard PCI bus. If Intel wanted to be bastards to Nvidia, they could put pinouts for PCI 1.0 on all their platforms and watch Nvidia choke on lack of bandwidth. This would comply with the letter of the law.
Update: As several readers noted, section HH covers this by defining “Standard PCI Bus” as meaning any PCI Express Base Specification. That means someone writing the settlement has decent technical knowledge.
Secondly, Intel doesn’t need to be bastards, they can just continue with the bog-standard half-speed PCIe 2.0 link that they have on their Atoms. This doesn’t provide enough bandwidth to run a retired analog cigarette vending machine let alone a modern GPU. If Intel doesn’t want a GPU on their platforms, it is trivial to abide by the letter of the law and still screw Nvidia. Won’t this be fun to watch?
Rumors have it that Intel was making changes to their chipsets that detected Nvidia GPUs and hamstrung performance on them. Having the GPU not work at all would be too obvious, but performance losses are a bit of a “he said, she said” argument. These changes broke the PCIe spec, but are basically impossible to prove without a lot of specialized equipment, trained engineers, and time. Given that it was Nvidia complaining, it is more likely that it was simply bad engineering by the GPU (formerly) giant. Either way, it is a moot point now.
Licensing and Fabbing
The things dealt with in part III are all about who can make what, and where. Intel has been accused of playing games around this for years, allegedly threatening third parties, or simply not answering questions posted to them. A company can do a lot with veiled threats, or simply not confirming what has been agreed to in other venues.
Part III. A. is interesting, but was mostly covered in the AMD/Intel license agreement from last year. It more or less says that anyone who fabs a product covered by AMD, Via or Nvidia cross-license agreements can be told about the patents and items covered by the agreement. The game that was being played behind the scenes seems to be that Intel would not allow companies to talk about cross-license agreement even when it was necessary for the ordinary course of business.
III B. gets more interesting, the FTC extended the Via license agreement with Intel for a total of 15 years, it now ends on April 7, 2018. Via now has clear sailing until the patents in question are expired, a huge win for the boys in Austin.
In III C. 3., the FTC legislates that Intel must state publicly when asked that Via can make x86 parts. This is the “Y’all smile and nod there boy” clause, and was prompted by some games Intel has played in the past where they didn’t respond to such requests, leaving the questioner in legal limbo. It is, or was, a nasty game, but easy enough to do without technically saying anything wrong.
Part III basically removes Intel’s ability to pull the “Nice place you got here, shame if it burned down accidentally” style of sales. Not that Intel does this, just ask them, however I have personally heard several customers say Intel personnel did just that. Proving it is an entirely different matter, but now there is no need. An FTC sanctioned mother hen with a baseball bat and an empty six pack of Red Bull is watching.S|A
Part 2 to follow, covering Sales, MCMs and Apple.
Editor’s note: This analysis is based upon the proposed FTC settlement document. There is a 30 day review and comment period after which modifications may be made, rare, and the document finalized and implemented.
Latest posts by Charlie Demerjian (see all)
- SemiAccurate has Skylake-SP die shots, sizes, and more - Jul 21, 2017
- Skylake-SP has a diverged core - Jul 19, 2017
- Intel’s Purley platform architecture is a step forward - Jul 13, 2017
- Intel launches Purley aka metal Xeons - Jul 11, 2017
- AMD’s Epyc has lots of connections - Jun 26, 2017