FTC-Intel Lawsuit Raises Possibility Of NVIDIA-Built x86 Processor - HotHardware
FTC-Intel Lawsuit Raises Possibility Of NVIDIA-Built x86 Processor

FTC-Intel Lawsuit Raises Possibility Of NVIDIA-Built x86 Processor

When the FTC sued Intel last week over the company's alleged anti-competitive behaviors, we noted that NVIDIA could be one of the main proponents (and beneficiaries) of such a lawsuit, particularly given the price structure of Intel's Atom products. It's now been alleged that NVIDIA's interest in the FTC's investigation goes beyond Atom's chipset; The New York Times claims the GPU designer may have been working on an x86 processor since 2007. This particular rumor has been kicked around a time or ten before, but the FTC complaint does raise the question. (If you want to see the full FTC complaint, you can view it here.)

According to the FTC, Intel's decision to sell Atom+chipset packages at a lower price than plain Atom CPUs could constitute illegal bundling as it prevents/inhibits would-be competitors from offering profitable alternative solutions

The FTC document addresses Intel's competitive position in the CPU, GPU, and chipset markets, and examines how the company's alleged behavior has impacted NVIDIA rather than strictly focusing on AMD. It raises several new competitive issues AMD's original complaint did not address, including accusations that certain benchmarks have been hand-tailored to favor Intel processors, and includes two propositions for relief that, if granted, would remove one of the primary barriers to entry on the x86 market. These are:
  • 17. Requiring Intel to make available technology (including whatever is necessary to interoperate with Intel's CPUs or chipsets) to others, via licensing or other means, upon such terms and conditions as the Commission may order, including but not limited to extensions of terms of current licenses.
  • 18. Prohibiting Intel from including or enforcing terms in its x86 licensing agreements that restrict the ability of licensees to change ownership, to obtain investments or financing, to outsource production of x86 microprocessors, or to otherwise partner with third parties to expand output.
The Legality of Licensing

Over the past 33 years, Intel has fought tooth and nail to keep an iron grip on the x86 family of patents. While it was forced to issue a license to AMD in 1982 due to IBM's requirement that it have a second source for all processor production, Intel attempted to terminate its agreement with AMD in 1986, and aggressively sought to prevent its one-time partner from producing an 80386-compatible design. AMD's launch of its own Am386 was delayed by several years thanks to this tactic; the California Supreme Court finally found in AMD's favor in 1994. To date, AMD's x86 license (the parts of it we can see, anyway), has been subject to the various restrictions and limitations listed in 17 and 18 above.

AMD isn't the only company that ever developed x86-compatible processors and then ended up in court with Intel, but it's the only competitor left standing with any degree of market share. VIA subsidiary Centaur Technologies has its own license, but the terms and conditions of the agreement aren't widely publicized. Historically, Intel has sued manufacturers for patent infringement, even if the processors themselves were developed entirely from reverse engineering and were later found to violate no Intel patents. An FTC decision that forced Intel to reduce or eliminate the restrictions it places on the few companies that even possess x86 licenses could have long-term implications for the market, but NVIDIA probably isn't planning to build an x86 processor anytime soon. It's true, as the New York Times points out, that the company has hired a number of engineers that once worked for Transmeta, but the new workers were probably brought in to work on ARM-based
Tegra solutions, not a brand-new x86 design.

Why x86 Doesn't (Yet) Make Sense

Even if NVIDIA was working on an x86 processor design, its ability to successfully bring that chip to market would depend entirely on the FTC's ruling, that ruling being upheld upon appeal, and, most likely, that ruling being upheld upon appeal, again. The entire process could drag on for years—when Intel and AMD tangled with each other over x86 licensing back in 1986, it took eight years for the Supreme Court of California to finally rule that AMD had the right to manufacture x86 processors.

Support from major OEMs or motherboard manufacturers would be all but nonexistent, thanks to factors the FTC points out in other parts of its complaint. Intel is capable of exerting an inordinate amount of control over the computer industry, thanks to its dominant position in the CPU and motherboard markets. Challenging Intel by selling parts made by a company Intel is suing (or threatening to sue) isn't recommended, and the vast majority of manufacturers simply won't risk it. NVIDIA said as much when it
acknowledged it was shuttering chipset development, and the fate of VIA Technologies chipset division is proof in and of itself.

For those of you who aren't aware, VIA shot to prominence in 1999-2000 as a provider of K6, K7 (Athlon) and Intel chipsets. At that time, Intel had no 133MHz, SDRAM-equipped chipset, while the new Athlon had been limited to an old platform of AMD design. VIA released products that filled both niches and gained a huge chunk of the Pentium 3 chipset market in a relatively short time. Intel, smarting from the loss, then announced that only companies with a license to produce Pentium 4 chipsets would be allowed to do so. VIA retorted that it already had a license, Intel said it didn't, and VIA launched its DDR-based P4X266 anyway. The chipset did well at first, right up until Intel sued VIA, at which point P4X266 solutions started to vanish. None of the company's later designs gained more than a sliver of traction in the market.

Conclusion

The FTC's decision to sue Intel has shown a spotlight on NVIDIA's interaction with that company, and led to a lot of speculation on what the GPU and chipset manufacturer might hope to gain. In this case, we believe the speculations that make sense are those that don't depart from NVIDIA's current core competencies and acknowledged major initiatives. Chipset licensing? Check. Allegedly illegal bundling and below-cost product discounts? Check. The right to produce an x86 core? Not so much. The first two complaints, if proven and addressed, would allow NVIDIA to design its own compatible products and compete more fairly in terms of price. The third implies that NVIDIA is risking the success or failure of a major initiative on a court case that could drag on for nearly a decade. The company may have had a rough 2009, but there's been nothing to suggest its willing to risk its future in such a mercurial roll of the dice. 
0
+ -

Intel may fight and win against this remediation, but they need to open their eyes: Java, .Net, Linux.... The x86 instruction set is slowly becoming irrelevant.

If Microsoft were to release Win7 for ARM tomorrow and put actual support for all sectors behind it, Intel stock would tumble.

0
+ -

How on earth do Java, .Net, or Linux make x86 irrelevant? You're comparing two different platforms with an entire operating system, and then comparing *that* to an architecture. Hopefully you realize that no modern processors actually execute x86 instructions--x86 instructions are broken into micro-ops (sometimes called uops) that are then executed.

Even the original "86" designation (meaning 8 registers, 16-bit processor) is long since obsolete--not only do we have 32-bit and 64-bit chips (obviously), Intel has used register-renaming to 'trick' architectures since the days of the Pentium Pro.

0
+ -

>> How on earth do Java, .Net, or Linux make x86 irrelevant?

Think about it:

Both Java, and .Net apps are compiled to Intermediate Languages. These distributed apps will run on any system with a Java VM or CLR implementation (and supporting libs). The JIT compilation will rewrite it to any architecture. I've written Gtk GUIs with Mono such that you can run the same .exe on x86-Windows or Cell-Linux... no changes necessary.

Linux will run on any system it's recompiled for. x86 instruction set or not. (Practically) All the apps are open-source and can be recompiled for the same new CPU.

These all destroy the dependence on the x86 instruction set for the purpose of running common apps.  If history shows us anything, there will be even more abstraction from the hardware in the future.  Intel's going to have to make money on the performance of their hardware, not artificial value of an instruction set.

Intel and Microsoft are trying the same trick:  Lock everyone into your API - be it the Windows API or the x86 instruction set.  Abstracting things up to higher levels is going to kill those two.  MS knows this, or their supporting CLR framework library wouldn't be proprietary.

Login or Register to Comment
Post a Comment
Username:   Password: