The FTC Sues Intel Over CPU & GPU Competition
by Ryan Smith on December 16, 2009 12:00 AM EST- Posted in
- CPUs
AMD and Intel have had their differences. And by differences, we mean Intel engaging in anti-competitive actions that they’ve been found guilty of in the European Union.
But all of this was supposed to come to a close last month, when AMD and Intel buried the hatchet and made up for past offenses. In return for some cash, some good behavior out of Intel, and for Intel to stop trying to block the Global Foundries deal, AMD would drop all of their civil and regulatory complaints against Intel. And that would be the end of Intel’s legal problems with various governments, right? No, as it turns out that’s wrong.
The catalyst for Intel’s legal woes (besides their own actions, obviously) has been AMD complaining to various regulatory boards about anti-competitive actions undertaken by Intel. Based on those complaints, the European Commission, the South Korean FTC, and the American FTC have been investigating Intel for some time now over these alleged actions. Intel has been found guilty and fined in the EU and South Korea (with both cases on appeal) while the American FTC has continued to investigate.
In fact despite the FTC just now suing Intel, this is actually about half-way through the process. The FTC investigation is done, and they have been negotiating with Intel in private for quite some time to get the matter settled. A lawsuit is the next step for the FTC, when those negotiations break down. Those negotiations have in fact broken down, so here we are: the FTC has sued Intel, and the biggest court battle ever for Intel is soon to begin.
What the FTC Accuses Intel of Doing in the CPU Market
As the FTC’s investigation into the matter is already over, they have published a complete list of complaints against Intel which will be the basis of the coming trial. Based on these complaints the FTC case is a significant departure from the EU and South Korean cases, as the FTC is accusing Intel over not only anti-AMD shenanigans early this decade, but of continuing anti-AMD and anti-NVIDIA shenanigans right up to this day.
The Athlon, the processor that's at the root of all of Intel's legal troubles
The case fundamentally breaks down into two halves: what Intel did against AMD in the CPU market, and what they’re continuing to do against AMD and NVIDIA in the GPU market. Let’s start with the CPU-focused complaints:
- The usual complaints we’ve seen from the EU. Intel rewarded OEMs to not use AMD’s processors through various means, such as volume discounts, withholding advertising & R&D money, and threatening OEMs with a low-priority during CPU shortages.
- Intel reworked their compiler to put AMD CPUs at a disadvantage. For a time Intel’s compiler would not enable SSE/SSE2 codepaths on non-Intel CPUs, our assumption is that this the specific complaint. To our knowledge this has been resolved for quite some time now.
- Intel paid/coerced software and hardware vendors to not support or to limit their support for AMD CPUs. This includes having vendors label their wares as Intel compatible, but not AMD compatible.
- False advertising. This includes hiding the compiler changes from developers, misrepresenting benchmark results (such as BAPCo Sysmark) that changed due to those compiler changes, and general misrepresentation of benchmarks as being “real world” when they are not.
Interestingly enough, the FTC cites Intel’s reasoning for all of this being that the company was at a competitive disadvantage, and engaged in these actions to buy time to improve their products. The timelines given place specific emphasis on the Athlon (K7) launch in 1999, and the Athlon 64 (K8) launch in 2003. This is a somewhat different take than in past cases, where Intel was merely accused of attempting to keep AMD’s overall market share down rather than specifically bridging performance gaps.
The FTC believes that the effects of all of these actions have (besides limiting AMD): served to drive up CPU prices, driven up CPU distribution costs, limited CPU innovation, harmed AMD’s ability to market CPUs, limited the ability of OEMs to innovate and differentiate their products, and reduced the quality of industry benchmarking.
Ultimately all of the CPU accusations are for things long past; none of the FTC’s CPU-related allegations are for things that have occurred in the last few years. We would not take this as a sign that the FTC is happy with the current market situation, but that they have no proof that they wish to follow up on that would show Intel as having engaged in anti-competitive actions in the CPU market in the last few years. The FTC does want some significant changes at Intel, which we’ll discuss in a bit.
Finally, there’s also the matter of AMD. Since AMD and Intel have settled their matters, AMD is presumably not going to participate in these proceedings as an ally of the FTC. As the FTC is going ahead on these charges, it’s clear that they aren’t worried about what this means for their position.
What the FTC Accuses Intel of Doing in the GPU Market
When we were first reading the FTC’s suit, the thing that caught us entirely off-guard was that it wasn’t merely about anti-competitive actions in the CPU market, but anti-competitive actions in the GPU market as well. While the CPU-related accusations are all for things done well in the past, the GPU accusations are fresh, very fresh. These run right up to today, and include the Larrabee project and the anti-competitive actions Intel has taken in the GPU market both outside and inside that project. To get right to the point, the FTC believes that as things currently stand, Intel is likely to get a monopoly on the GPU market similar to the one that they have on the CPU market, and that this monopoly will be created by abusing their CPU monopoly.
In the complaints about the GPU market, both NVIDIA and AMD are mentioned as being the primary competitors for Intel. The bulk of the complaints however are related to NVIDIA and their chipset business, as while AMD stands to be harmed too by an Intel GPU monopoly, it’s NVIDIA that stands to be the most harmed. In effect Intel has finally gotten AMD off their back for CPU matters, only to now have NVIDIA on their back for GPU matters.
The GeForce 9400M: Intel's chief competitor in the integrated graphics market and a threatened product line
Just to note where things stand, the FTC already estimates that Intel has approximately 50% of the GPU market. This is consistent with the vast number of Intel IGP-equipped computers that are on the market. Depending on how you intend to count various user bases, this stands to grow in the future as Intel puts their IGP GPUs first on-chip, and then on-die with their CPUs.
The basis of the FTC’s complaint here is that they believe Intel is threatened by the rise of GPUs as programmable computing devices, and that using them in GPGPU situations threatens Intel by making CPUs less important (something NVIDIA has been trying to play for ages) and as a result less profitable. The FTC argues that Intel is seeking to establish a monopoly here to maintain their overall control of (and high margins in) the computing market.
As for the specific complaints:
- Intel eliminated the future threat of NVIDIA’s chipset business by refusing to license the latest version of the DMI bus (the bus that connects the Northbridge to the Southbridge) and the QPI bus (the bus that connects Nehalem processors to the X58 Northbridge) to NVIDIA, which prevents them from offering a chipset for Nehalem-generation CPUs.
- Intel “created several interoperability problems” with discrete CPUs, specifically to attack GPGPU functionality. We’re actually not sure what this means, it may be a complaint based on the fact that Lynnfield only offers single PCIe x16 connection coming from the CPU, which wouldn’t be enough to fully feed 2 high-end GPUs.
- Intel has attempted to harm GPGPU functionality by developing Larrabee. This includes lying about the state of Larrabee hardware and software, and making disparaging remarks about non-Intel development tools.
- In bundling CPUs with IGP chipsets, Intel is selling them at below-cost to drive out competition (given Intel’s margins, we find this one questionable. Below-cost would have to be extremely cheap).
- Intel priced Atom CPUs higher if they were not used with an Intel IGP chipset.
- All of this has enhanced Intel’s CPU monopoly.
The FTC believes that all of this will help Intel to establish a GPU monopoly. This is on top of all other effects of Intel’s actions, which are similar to the effects of their actions in the CPU market: driving up GPU prices, driving up GPU distribution costs, limited OEM differentiation, and limited GPU innovation.
There’s also one last complaint unrelated to GPUs, which has to do with standards.
- Intel used their market position to delay AMD and NVIDIA’s implementations of USB and HDCP by refusing to make the specifications accessible until Intel’s products were ready. We know that there has been some strife among Intel and virtually everyone else over Intel dragging its heels on the USB3 specification, but it’s not clear if this complaint is about that.
114 Comments
View All Comments
uibo - Thursday, December 17, 2009 - link
I thought their graphics drivers were not so great?Orangutan2 - Thursday, December 17, 2009 - link
I simply disagree with that.With integrated graphics Intel's chipsets are utterly abysmal. At a high level based off volumes you could conclude all Intel chipsets are rubbish.
Without integrated graphics they are sub par at the high end, relying on nVidia chips to support PCIe connectivity for RAID, graphics, SLI etc.
Sure they are ok for the average user or gamer. Is that what you want? Pure crap on the low end and lack of innovation at the high end?
Without competition we're back to 1997 with the Pentium II coming out with 66mhz bus on a slot = fail
mindless1 - Thursday, December 17, 2009 - link
You are aware the Pentium II, on 66MHz bus with a slot, was the right move?Do you not recall how much slower the old motherboard socket 7 L2 cache was? Do you not recall they were an evolutionary performance increase and did outperform the K6-2 handily, particularly in floating point?
Do you not recall the limitation was the process size and cache yields, that once they had the ability to get rid of the back-side bus and integrate the L2 into the die they did it and never looked back except a spattering of products to continue support for partners who had already invested R&D on the slotted platform?
I'm not quite sure who you're contrasting against when you look down upon their chipsets either. nVidia? I don't mind nVidia chipsets but there's the bumpgate scandal, a few issues with nForce 1 as well.
Who else? Sis or Via? Not even in the same ballpark as the aforementioned two. AMD only now has begun to polish their chipset drivers enough that they have the basic support claimed in the first place. Same could be said for Intel I suppose but they were always much better at documenting such errata.
Truth is, on average over time Intel had at least as good a product as anyone else, but it seems they tried to keep a bit too much of the pie because they weren't THAT much better and I can't condone their misguided attempt to force 945 chipsets on Atom platforms, a combo that makes no sense at all but for deliberate crippling and abandonment of the market's demands.
Diosjenin - Thursday, December 17, 2009 - link
Yes, Intel's IGPs are crap. But the quality of Intel's chipsets in general has absolutely nothing to do with the inclusion of the nForce 200 on high-end X58 boards.The X58 chipset has support for up to 36 PCIe 2.0 lanes. That's enough for two GPUs at x16 apiece, or for three GPUs, either in x16/x8/x8 or x16/x16/x4. nVidia allows two-way SLI to be enabled natively on X58 boards (after validation), but to enable x16/x16/x8 or x16/x16/x16 three-way SLI, you *need* an x16 to x16+x16 lane splitter.
That's exactly what the nForce 200 does - and that's more or less all it does. In fact, it's not even tied directly to SLI support at all. Boards like ASUS's P6T6 WS Revolution include *two* nForce 200 chipsets, just to get a higher number of PCIe lanes.
Everything else - native PCIe support, RAID, etc. - is all handled by X58, and is handled quite well. Intel's IGPs may suck, but they run a pretty tight northbridge ship.