Intel Settles With the FTC

The Settlement

The resulting settlement is, in the FTC’s words, “not punitive but rather remedial.” Going in to the suit the FTC was not asking for a fine or some other method of punishing Intel, and nothing like this is in the final settlement either. Everything in the settlement is geared towards undoing the damages from Intel’s past actions and/or preventing future damages by disallowing Intel from engaging in specific anti-competitive actions. Ultimately the only money that this will cost Intel is $10 million for a reimbursement fund to pay misled buyers of Intel’s compilers and libraries, and another $2 million to pay for Technical Consultants to evaluate Intel’s compliance over the next 10 years.

Overall the FTC got the vast majority of the terms they were requesting when they filed the suit back in December. However they did not get everything, and we’ll be touching on what they didn’t get.

The settlement covers several areas of Intel’s business: CPUs, Chipsets & GPUs, and Compilers & Claimed performance. The following are the settlement terms, roughly grouped by what business they impact:

CPUs

The first and most important requirement being placed on Intel is that they are barred from engaging in any further rebate schemes or punishment schemes to discourage OEMs from using AMD processors. This means that Intel cannot offer any kind of benefit or rebate to an OEM based on the percentage of Intel processors they use, and at the same time they cannot punish an OEM by taking away marketing dollars or limiting their chip supply if they do use AMD or Via processors. There is an exception to this however: this does not impact most volume discounts. Intel is still free to offer volume discounts so long as they don’t end up selling CPUs below cost. Furthermore Intel is free to break the above terms and offer benefits to OEMs if they reasonably believe that AMD is already doing the same thing – this effectively keeps AMD from abusing rebates and payments in a manner similar to what Intel was accused of, by allowing Intel to resume those rebates if AMD does.

The biggest benefactor of rebates, Dell?

The second requirement is very similar to AMD’s settlement with Intel last year which paved the way for AMD to spin-off its fabs in to Global Foundries and then in turn outsource the fabrication of their CPUs to GF. Intel is required to let all of its x86 licensees outsource their x86 CPU fabrication to a third party fab. As Intel’s agreement with AMD already allowed AMD to do this, this requirement effectively means that Intel also allows Via to do this.

The third CPU requirement is a bit more interesting. It’s well known that Intel’s x86 cross-licensing agreements with AMD and Via place strict requirements on what these companies can do while still maintaining their x86 licenses, largely to keep these companies from selling off their license or sub-licensing other companies to design x86 CPUs. Or to put this another way, Intel’s x86 license agreement is designed to keep AMD and Via as the only other x86 CPU designers and to prevent anyone else from becoming an x86 CPU designer by buying the license or the company.

The FTC has not gone so far as to require that Intel drops these provisions, but it does weaken them. If either AMD or Via has a “change of control” (i.e. a buyout/takeover/merger/joint-venture), Intel cannot immediately take the resulting company to court to terminate the license. Intel is required to enter in to good-faith negotiations with the new company to continue x86 CPU design and can only begin court proceedings after a certain period of time. As far as we can tell this does not require that Intel extend a license to a buyer of AMD or Via, but it does require that they consider it. If Intel does not act in good-faith in these negotiations, then the FTC can sanction Intel over it.

The Foundry Dilemma: A problem no more. x86 CPU fabrication can be outsourced

The big question of course is whether this will lead anywhere. x86 CPU development is a uniquely expensive and time-consuming endeavor – just because AMD or Via could work with another company doesn’t mean there’s anyone else out there that wants to. NVIDIA has long been considered a candidate for entering the x86 market, but as far as we can tell these terms are to protect the x86 market as a whole, and are not just there to allow NVIDIA to enter the market.

The final requirement of the settlement specifically pertains to Via. Via’s x86 license was scheduled to lapse in 2013 – Intel is required to offer a 5 year extension to Via. Note that this doesn’t compel Via to take the extension or under what terms Intel must offer it, but ultimately Via must be given the option to extend their x86 license to 2018.

Chipsets & GPUs

The next group of requirements relate to Intel’s chipset and GPU businesses, and also how they interact with competing chipset and GPU manufacturers. Thus these terms largely dictate how the company interacts with NVIDIA and AMD’s GPU business.

The first requirement is that Intel must continue to support PCIe on all of its CPUs/chipsets for the next six years. Ultimately this is to prevent Intel from releasing CPUs that can’t be used with a third party GPU, although technically speaking this can apply to any peripheral that uses PCIe. Notably this only applies to PCs, which automatically excludes any device with a screen smaller than 7”. This means that Intel’s SoC platforms such as Moorestown are not required to include PCIe connectivity (which would hamper the platform’s energy saving abilities) while netbook platforms such as Pine Trail are required to include PCIe connectivity. Similarly, designs over 7” such as tablets are not automatically defined as PCs, but this is where the FTC’s definition gets murky.

It’s worth noting that while this requirement means that Intel has to support PCIe, it does not specify a revision or the required number of lanes. Intel is free to choose PCIe 2.1, 3.0, 4.0 (if there ever is such a thing), etc., and we don’t expect that this will change Intel’s plans to move to newer versions of PCIe in the future. Meanwhile in lieu of lane requirements, there’s a second, more general requirement towards limiting the performance of the PCIe bus.

The second requirement is that Intel is not allowed to engage in any actions that limit the performance of the PCIe bus on the CPUs and chipsets, which would be a backdoor method of crippling AMD or NVIDIA’s GPUs’ performance. At first glance this would seem to require them to maintain status quo: x16 for GPUs on mainstream processors, and x1 for GPUs on Atom (much to the chagrin of NVIDIA no doubt). However Intel would be free to increase the number of available lanes on Atom if it suits their needs, and there’s also a clause for reducing PCIe performance. If Intel has a valid technological reason for a design change that reduces GPU performance and can prove in a real-world manner that this change benefits the performance of their CPUs, then they can go ahead with the design change. So while Intel is initially ordered to maintain the PCIe bus, they ultimately can make changes that hurt PCIe performance if it improves CPU performance.

PCI Express: Intel is required to support it for six more years

Finally, six years is a not a number that’s set in stone. If the industry moves away from PCIe sooner than this, then the FTC can cancel this requirement early at their discretion.

Ultimately the fact that this is a six year (or shorter) requirement is quite interesting in the face of the fact that most of the other requirements are for five or ten year periods. Since the FTC has the power to cancel this requirement at any time, why didn’t they go with a full ten years? With the oncoming merger of the GPU and the CPU in Intel’s Sandy Bridge and AMD’s Bulldozer, it’s not a stretch to question whether the PCIe bus has more than six years’ life left in it as a CPU-GPU interconnect. The way this requirement was structured would seem to indicate that it was NVIDIA and AMD driving it, in which case we’re left wondering what the two GPU juggernauts have planned for 2016 and beyond.

Finally for chipset & GPU requirements, the FTC is requiring that Intel accurately represent its roadmap. One of the FTC’s charges was that Intel mispresented its roadmap to NVIDIA which in turn lead to the spat between NVIDIA and Intel over chipsets, buses, and licensing rights, so this would forbid Intel from offering false roadmaps in the future. Since the terms of this settlement don’t involve the renegotiation/reinstatement of NVIDIA’s chipset license for DMI and QPI however, we’re not sure where this is going to lead beyond clarification over what version of PCIe future Intel chipsets/CPUs will support.

Compilers & Claimed Performance

The final group of requirements revolve around Intel’s compiler, libraries, and any performance claims made about their products which involve those compilers.

As we mentioned previously, Intel was accused of sabotaging their compiler to use suboptimal code paths for non-Intel CPUs, such as by using an x87 code path instead of an SSE2 codepath on an Athlon 64 processor. As far as we know this practice ended some time ago, but we’re still trying to get a more conclusive answer here. In any case there are a few different requirements related to this.

The first requirement is that Intel needs to disclose when their compiler is favoring their CPUs over AMD or Via’s CPUs. Notably this doesn’t require that they treat other CPUs equally (such as by picking code paths based on CPU feature flags), only that if they discriminate based on the CPU that they disclose this discrimination.

This leads in to the $10 million reimbursement program that Intel is being required to offer. This fund will be used to cover the costs encountered by mislead customers who choose to move their software to a non-Intel compiler and/or library. Since Intel now has to disclose any Intel-only optimizations in their compilers, this only applies to existing customers who used Intel’s compilers ahead of Intel’s compiler disclosure.

The rest of the requirements relate to Intel’s advertising of their compiler’s performance, and the performance of their products when using those compilers. Intel is not allowed to claim their compiler is faster on AMD/Via CPUs when this is not the case. Finally Intel will be required to disclose that benchmarks may not provide an accurate performance comparison between their processors and AMD/Via’s processors whenever they are making a performance claim involving benchmarks. Or to put this another way, it’s a Your Mileage May Vary clause for CPU advertisements.