PART TWO OF SemiAccurate’s look at the Intel/FTC settlement looks at sales, rebates, MCMs, bumps and cracking open an Apple. There is a lot of dirt in this one, and some business practices that will make your head hurt.
Sales and Rebates
In part IV, the topic is sales tactics, and was mostly covered by the previous AMD agreement. Intel merrily pointed out that they didn’t mind abiding by the agreements because they never did the alleged behaviors in the first place. Touché.
The behaviors that are now officially forbidden, again, are the normal exclusivity, rebates, discounts, and other things that are used to exclude competitors. This time around, punishing OEMs and customers who stray from the fold is also frowned upon. There are some gems here though.
The first one is IV. A. 7. where Intel is forbidden from making discounts in the way that most people think is obvious. Intel seems to only be allowed to give discounts when a certain sales level is reached. If they give discounts on orders of more than 50K CPUs a year, then another break at 100K, and another at 150K, making those levels retroactive is not really allowed any more.
Under IV. A. 7., Intel for example must price CPUs 1-49,999 at one price, and then give the discounts on CPUs 50,000-99,999, and another break on CPUs 100,000-149,999. What they can’t do is retroactively give the OEM the same price the first 49,999 CPUs that they get on the others.
While this may seem petty and pointless, it prevents Intel from doing things like offering an OEM an extra 10K CPUs and discounting the entire month’s purchase enough to make those added chips ‘free’. Several people tell SemiAccurate that Intel would see that AMD would account for 10% of an OEM’s sales, and offer that OEM enough steeply discounted CPUs to make it unnecessary to buy any AMD products.
The discounts would be retroactive, so Intel could price those new CPUs at basically zero, and amortize it over the other 90% of sales. If the first 90% are sold at $100 per CPU, an Intel sales rep comes in and says, we will discount all of the month’s purchases by 10% if you just reach a sales level that is 10% more, you would have to be an idiot to say no.
If an OEM moves 100K computers a month, 90K Intel units at $100 per CPU costs them $9M. If AMD comes in and offers their CPUs for $50 each for the last 10%, that is $500K for those CPUs, and a total cost of $9.5M for the OEM. Fair enough.
In comes Intel and offers the OEM an additional 10K CPUs at $90 each, more than AMD is charging, and a 10% discount on the list price. That 10% discount extends to the other 90% of the customer’s purchases, so an OEM ends up with 100K CPUs for $9M. Magic. Intel effectively locks AMD out, and keeps ASPs up to almost twice what AMD can get.
Intel representatives will look you in the eye and say that they sold those 10K CPUs to the OEM at $90 each, which they technically did. Effectively though, they gave the OEM 10K CPUs to exclude AMD. This may be sleazy but it is not illegal. In any case, that tactic is now forbidden.
MCMs and Bumps
Moving on to Part IV. B. 2. a. we get into some of the weirder Nvidia claims. It says that Intel will not “use the same model number for Computer Products containing a Relevant Product or Computer Product Chipset supplied by Respondent in conjunction with Computer Products containing a Relevant Product or Computer Product Chipset not supplied by Respondent”.
This may seem weird, but what it does is work in conjunction with some of the other ‘openness’ clauses, analyzed later, that allow Nvidia to force their way onto an Intel MCM. Imagine an Arrandale or Clarkdale MCM with an Nvidia memory controller and GPU under the heat spreader and you get the idea.
In Nvidia’s countersuit against Intel, case C.A. No. 4373-VCS, the redacted public document, smoothly titled 112_200903261953290462.pdf, this topic comes up in paragraph 35. It says, “In attempting to resolve the dispute with Intel in good faith, NVIDIA provided a detailed technical explanation as to why NVIDIA’s Nehalem-compatible chipsets would continued to be licensed. Nvidia explained and illustrated how it chipsets would continue to REDACTED even using Intel’s unduly narrow definitions. After initially admitting during an in-person meeting that it “had not considered” NVIDIA’s detailed, technical explanation, Intel sent NVIDIA a letter summarily dismissing NVIDIA’s position”.
What Nvidia is asking for here is access to replace one of the two chips on Intel’s MCM with one of theirs. If you think this is a sane technical proposition, read this three part story. The whole scheme relies on Nvidia’s packaging technology, something they still can’t get right. If you sense a mound of smoldering half-working MCM’s with cracked bumps in the near future, you are not alone.
What Nvidia failed to achieve contractually, they are trying to strong arm. I wonder if the FTC will buy the sob stories when Nvidia can’t get the packaging technology to make this all work on the open market? Intel develops a lot of tech, and they are masters of packaging. Nvidia can’t do the easy things adequately, but they yell about it a lot.
What Part IV. B. 2. a. is all about is preventing Intel from calling a CPU with an Nvidia chip on the MCM the same thing as one without the Nvidia part. While Intel naming is a blight on humanity there is an easy way around this. Any CPU with all parts Intel would be named something 1 higher than the equivalent part with Nvidia chips on them. Bigger is better, right? In this case, bigger is also likely more reliable.
I can also see a future with Intel iSomethingmeaningless CPU model number 785 with two Intel chips, and an equivalent Intel/Nvidia hybrid as model iSomethingmeaningless 7804ooOOO00O0Oo)(00OoO930729/i33%&ARZ2125-NV=-HT1946+++ATH. Nothing SemiAccurate can find in this agreement prevents Intel from denying marketing funds if a company doesn’t use the full trademarked product name everywhere.
Apple Broken Open
Another bombshell is in Part IV. B. 8., it opens Apple up to AMD CPUs. How? It states that any company that Intel provides “Extraordinary Assistance” to can have an exclusivity agreement only long enough to pay for the engineering, or 30 months from when the first product benefiting from said Extraordinary Assistance is sold.
One of the dark secrets of the current PC world is that Apple doesn’t make motherboards, they are the biggest Intel white (or silver) box vendor out there. Intel makes exceptionally great mobos, and Apple wraps them in a pretty package that drops the reliability by an order of magnitude or more. Remember the MacBook Air? Extraordinary engineering feat until you consider that it was nothing more than an Intel Metro laptop board with some features removed.
There hasn’t been an AMD CPU in an Apple product yet for two reasons. The first is that AMD CPUs aren’t all there in comparison to Intel’s, especially on the mobile side. While an iMac might be a good home for an Phenom II HE, the rest are fairly unsuitable. This is completely AMD’s fault, and won’t be rectified until they get a new core, or at least a compelling reason for Apple to overlook the core problems, pun intended.
That compelling reason would be Llano. The problem is that whenever you asked AMD about the possibilities of getting into Apple, they would politely respond about how they are working on it, then walk away muttering the words “exclusivity agreement”. When asked about those agreements, the topic was quickly changed.
Those types of agreements are effectively forbidden now, so will we see an iMac with a Llano in it next year? Unless Intel gets it’s graphics act together ASAP, we probably will. Don’t hold your breath for Intel to get serious about graphics until after the contract is publicly lost.S|A
Part 3 to follow, covering burdens of proof, compilers, and technical watchdogs.
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)
- Spin Transfer Technologies talks about their ST-MRAMs - Sep 27, 2016
- ARM adds CMN-600 interconnect and DMC-620 memory controllers - Sep 27, 2016
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016