Nvidia trapped in x86 pipe dream
Intel foe wants Intel license
Nvidia is contemplating its very own x86-compatible processor core. in theory, this core could combine with Nvidia graphics technology to create system-on-chip (SoC) silicon that competes in the rapidly expanding low-power handheld-device market.
According to a report from the The Wall Street Journal, Mike Hara, Nvidia's senior vice president of investor relations, told analysts Wednesday that although the company's Tegra SoCs have ARM-based cores, "some day it's going to make sense to take the same approach in the x86 markets as well."
Rumors of an Nvidia x86 chip have been around since at least 2006, but Hara's comment is the first time to our knowledge that the company itself has mentioned the possibility in public.
Hara is right about one thing: x86-compatible SoCs are the New Hotness, with both AMD and Intel actively developing them. But it's unclear whether Nvidia has or could obtain the necessary x86 license from Intel. We called both companies to find out the status of their licensing position, but did not immediately receive an answer - and our assumption is that the situation is murky, at best.
After all, the two companies are not exactly buddy-buddy these days. They are, in fact, currently in court after negotiations broke down over whether their 2004 cross-licensing pact allows Nvidia to make chipsets compatible with the QuickPath Interconnect in Chipzilla's new Nehalem processors.
What's more, Nvidia's mercurial CEO Jen-Hsun Huang couldn't have made many friends at Intel when he said in reference to the lawsuit, "At the heart of this issue is that the CPU has run its course and the soul of the PC is shifting quickly to the GPU. This is clearly an attempt to stifle innovation to protect a decaying CPU business."
As overstated as Huang's comment may seem, there is a grain of truth in his rhetoric. Last December's firming up of the OpenCL 1.0 standard gave a boost to the long-sought dream of GPGPU computing (general-purpose computation on GPUs), in which GPUs take over from CPUs those tasks that would benefit from a GPU's massively parallel structure.
That said, the CPU market is far from "decaying." If anything, it's evolving - especially in the low-power space - into an SoC market. That is, one in which the CPU is reborn as the core of a single chip that manages computing, graphics, memory, and I/O.
This so-called "embedded graphics" scheme won't kill off discrete graphics, but integrated graphics are headed for a fall.
To compete in the embedded-graphics market, Nvidia would have to integrate its graphics expertise into an SoC. And to have an SoC be competitive in more than the limited handset market now dominated by ARM cores, Nvidia would have to be part of an x86-compatible SoC.
But to do so, perhaps Nvidia wouldn't have to have an x86 licensing agreement with Intel. It could, instead, work with a core-designing company that already has one - Via Technologies, for example.
There's plenty of time, however, for this drama to play itself out. According to Nvidia's Hara, if the company decides to develop an x86 core, meaning if it could settle on a licensing agreement with Intel or if it could form a partnership with a company that already has one, products wouldn't be available for two or three years.
And at the pace that AMD and Intel are moving towards x86 SoCs, that may be too late. ®
COMMENTS
@AJ Stiles -
"Why the fuck should they need a licence anyway, just to make an x86- compatible processor?"
Yes, because they'd use Intel IP, the x86 insctructions set. It's not the idea, but the implementation which is licenses, here, so there's no abuse.
But if they want, they could also do as others have done, come up with an entirely new instruction set, call it NV86 ! But then, they would have either to wait a decade before seeing customers adoption, or emulate x86 atop, which is back to the license.
Use OpenSPARC and pay NO ONE...
I don't think the world needs a 3rd x86 player, given Intel is about to switch to Nehalem family.
I think Nvidia mostly need an x86 system as a boot, compile and control environment for code they want running on their families of GPUs.
Why not use an OpenSPARC implementation for that ? In Sun's implementations (the Niagara family) they are relatively power efficient and low in the GHz stakes. The control OS could be Solaris (if Sun ever got their a** into gear and finished a binary Open Solaris distro that wouldn't cost Nvidia anything either). There *are* ports of Linux to SPARC but ironically the ecosystem of things people actually use like Apache, Java, compilers etc is far more mature and complete on Solaris-on-SPARC than on Linux-on-SPARC (with the reverse being true on x86).
So if the PLAYstation is where work stops, then Nvidia could reinvent the WORKstation as a true computing device, rather than today's bastard child of a general purpose traffic light controller.
@Gis Bun
The terms these days are synonymous, since any contemporary consumer-grade x86 processor is expected to be 64-bit-capable. Perhaps the one exception is the low-power chips such as the Atom where efficiency takes precedence over performance. Anyway, like I said before, compatibility relies on being able to perform certain techniques that Intel has patented. They sure as blank can't go to AMD since it's a direct competitor (they now possess ATI, nVidia's chief GPU rival). And Intel won't grant the license itself because they're horning into nVidia's market itself with Larabee, making them direct competitors also.
@BioTube
Each computing architecture has its own advantages. Current CPU designs are meant to handle almost anything computational and thus are best suited for general and unpredictable tasks, such as tasks involving human interaction. Whereas the parallel nature of GPUs are best employed on highly-structured, predictable jobs. That's why they're math whizzes (math is highly-structured and thus predictable).
x64 is a M$ invention.
The x86-64 specification was designed by Advanced Micro Devices (AMD), who have since renamed it AMD64.[1] AMD licensed its x86-64 design to Intel, where it is marketed under the name Intel 64 (formerly EM64T)..
