ANOTHER SHOE HAS DROPPED on the heads of Intel’s legal team. This time it is an FTC lawsuit saying that the company is anticompetitive. Time for the fun to begin again.
Update 1: Fixed the logo
The FTC complaint alleges that Intel has abused it’s ‘dominant market position’ to stifle competition. Anecdotally speaking, that much is obvious. Talk to any of Intel’s competitors and they will say the same thing. That said, separating the facts from bitterness and abuse from simply having better products is going to be a tough slog for the FTC legal team.
The FTC alleges a few things – anticompetitive inducements and threats from a CPU near-monopolist, compiler tricks, and smothering GPUs. All of these have been gone over in fairly rigorous detail in the past few months, especially with the recent EU Commission and AMD legal tiffs getting a great deal of publicity.
The first one – inducements and threats to maintain dominance in the x86 CPU market, has been done to death recently. The EU Commission pretty well slapped Intel silly over that, and the $1.25 billion check that Intel just wrote out payable to AMD should have enough zeros to drive home the point that there was some credence to the allegations.
Intel is fond of saying it has never been found guilty of anything, and while that is true, it has effectively lost in Japan, Korea, the European Union and to AMD in a civil case. There is spin and then there is a lack of victories anywhere, across four disparate legal systems – decide which you want to put more weight on. Then again, even if things are settled, if Intel ever ‘admits’ wrongdoing, it opens itself up to new rounds of fun and courtroom visits, so I expect it to say nothing more or less than, “we are completely innocent”. That’s Intel’s story and it’s sticking to it.
Update 2: Intel is rather keen to point out that it has never been found guilty of anything, nor really ever ended up in a courtroom, and that is quite true. Legal and regulatory actions have been brought against it in four different coutries under four different sets of laws. Intel quite wisely settled three of them, and is still appealing the most recent action. It has not walked away without some sort of restriction, fine or payment in any of these legal proceedings, but some have barely been a slap on the wrist.
Given the FTC has enough documents to build a fort with, its case will be one of finding a needle in a haystack. No, that is not meant to say that there are no needles, just lots and lots of hay – the needles are quite likely there. Luckily, there have been several years of litigation that drew some very precise maps as to where they can be found.
Compiler tricks are the really obvious one, and this will be a slam dunk for the FTC to prove. I have long been annoyed by the Intel compiler tricks, and you can see it in a few articles like this and this. Intel vehemently denies any tricks, but they are painfully easy to prove. Just take a commercial program compiled with Intel’s ICC (the Intel compiler), run it on an AMD CPU and get a result/benchmark/scores. Change the CPUID string on the CPU to Intel and re-run the same test. Compare.
If the numbers differ at all there is something funny going on. Before you jump in and say that is not possible to change a CPUID string, you can do so with ease if you have the right boards. This is a painfully easy test, and the results are really conclusive.
To their credit, Intel compilers have gotten much better over the past few generations, and now are not playing tricks, or at least far fewer of them, but the older compilers did some very shady ‘optimizations’. They are supposed to check certain flags to see if a CPU has a given capability like MMX, SSEx or any other extension.
Programs compiled with older versions of ICC would instead check the CPU string, and if it said ‘GenuineIntel’ or whatever else the compiler was looking for, it would then check status flags, and run the faster code paths.
If it found a chip that was something other than ‘GenuineIntel’, it would just ‘assume’ that there were no, or very few, extra capabilities, and run vastly slower code paths. You can read some of the gory details here, and there are many others out there. This is one that Intel should be very afraid of ever reaching court.
The last allegation, cutting out GPUs, is a new one, and it is undoubtedly brought on by Dear Leader and his crew of screwdriver wielding lawyers that can’t read a legal contract. If you go over the Intel versus Nvidia history and the follow on Nvidia v. Intel lawsuits carefully, you can see that Nvidia made a grave error in the contracts the lawsuits cover, fatal in fact.
Unless this allegation is about something more than Nvidia feeling pouty because it botched drafting a contract, it is unlikely to go very far. Intel has little to no incentive to block GPU sales, and given that Larrabee doesn’t even exist yet in the market as a GPU, it is going to be very hard to prove any damages there.
In the end, the FTC should have a cakewalk on two of the three counts, and until the third is detailed much further, who knows? On the surface, it has no legs. That said, we are in for a long wait before anything more than press releases happens.
The complaint says. “Under the recently implemented rule expediting the Part 3 administrative hearing process, this matter is tentatively scheduled to be heard before an Administrative Law Judge on September 15, 2010, at 10:00 a.m.” This is the ‘expedited’ hearing, so it is only about a year out, but that is before the inevitable delays. The wheels of justice turn mighty slowly for things like this, just look at the SCO soap opera for a taste of how one long side can drag out litigation.S|A
Latest posts by Charlie Demerjian (see all)
- Redhat put Calxeda based ARM servers in to production - May 15, 2013
- Why would Nvidia release an unlocked GTX680 to GTX770 BIOS? - May 14, 2013
- Get a Nvidia GTX770 free! Bios flash and GTX680 not included but required. - May 13, 2013
- XBox Next/720 architecture only adds to Microsoft’s problems - May 13, 2013
- Intel talks about their 22nm P1271 SoC process for Silvermont - May 10, 2013