IT LOOKS LIKE some people are finally starting to catch on to what we have been saying for almost four years now, Nvidia is building an x86 chip. The story is long, complex, and it is SemiAccurate’s opinion that the CPU will never make it to market for technical and legal reasons.
First off, what is this mystery chip? That is the easy part, it is Tegra 5, code named T50. As you might recall, Tegra 3 just taped out, and since the chips are on a yearly cadence, Tegra 5 should tape out in mid to late 2012 with products following in 2013. We have previously covered it here, here, here, here, and here.
That is if all goes well, and there is only one major technical wrinkle that we can see, the new core. What new core? The new ARM Eagle core, you can see it in the slide below as the successor to the Coretex-A9 line. Sources tell SemiAccurate that this will be a 64-bit quad-core capable CPU that has quite a bit more power than than the current A9s.
Eagles soar above Coretex-A9s
The problem we mentioned above is not unique to Nvidia though, it is a new core that the company has never worked with before, and institutional knowledge about it will be pretty thin. ARM delivers cores as a black box, but there is still an associated learning curve. TI is slated to be the first one out of the gate next year with an Eagle product, and the floodgates open from there.
Nvidia is more than capable of implementing the Eagle core in an SoC, but if anything will delay a part, a new core is near the top of the list. There will have to be a lot of tweaking done to the core, or at the very least, a lot of work done on the surrounding components to enable an x86 translation engine. All of this carries a lot of risk, and that risk tends to show up as delays in product shipping.
That said, lets assume that Tegra 5, code named T50, will be on time. How is an ARM core related to x86? That is easy, Nvidia is going to use Transmeta-esque code morphing firmware to make the CPU run x86 code. Firmware x86 has two associated problems, one technical, one legal.
On the technical side, the problem is simple, speed. ARM A9 CPUs are great for phone level applications, and can reach into the current tablet space, but hit a glass ceiling there. If Eagle doubles the performance per MHz and doubles performance per watt, it will basically be on par with the low end of the Atom-class CPUs, and woefully behind the Nano/Bobcat level of performance.
On top of that, you strap a translation engine to change the x86 code to ARM instructions, and you lose more speed, add latency, and drop performance. It also costs power. Transmeta managed to drop power levels greatly while doing the same thing, so why can’t Nvidia?
When Transmeta attacked the problem, it was competing against K7 and K8 CPUs, and, wait for it, Pentium 4s. Being more efficient than those CPUs is like boasting to your friends that you are a better gymnast than a quadriplegic, technically true, but….
This isn’t to say that Transmeta wasn’t revolutionary, but the bar has been raised by orders of magnitude since then, and every milliwatt that could be squeezed out of a design has been. If you think that there is a wasted gate in an Atom design, you are wrong, Intel and AMD take power very seriously now.
Nvidia is trying to shoot for the same hole as Transmeta, but that hole closed two years ago. It will be interesting to see what they come up with, but hardware is more efficient than firmware, and everyone else is doing it in hardware. To top things off, Intel, for one, effectively has a two process node lead on anything Nvidia can buy from a foundry.
While some of this may change in the intervening three years, I can’t technically see how Nvidia will pull this chip off. If the company does everything they are planning perfectly, the end result won’t be enough. Why are they wasting the development resources and engineering bandwidth?
That one is easy, management ego. Nvidia has never let technical challenges stop them from attempting the impossible. Unfortunately for them, this is not a play on words, they try to do things that are clearly impossible. The laws of physics inevitably triumph, just ask the lawyers on the receiving end of the Bumpgate settlement checks.
Management is also the source of the other problem. Everyone contacted by SemiAccurate for the past several years has stated unequivocally that Nvidia does not have the legal right to make an x86 CPU. The problem is not the instruction set behind the chips, aka x86, but the underlying patents.
It seems Intel, AMD and VIA have some very strong patents that cover the basic functionality necessary to implement an x86 CPU. The patents are about how the instruction is carried out, not what the instruction is called. The right to use these patents are often referred to as an ‘x86 license’.
Intel and AMD have come to terms with those, the latest round settled late in 2009. VIA beat Intel with the patent stick as well, and the result was that VIA has the right to make x86 CPUs. While the rest of the agreement was confidential, the FTC just extended it for several more years.
Speaking of the FTC, it is strongly rumored that the one thing Nvidia wanted out of the FTC settlement was the right to make an x86 CPU. Sources at Nvidia tell SemiAccurate that this was the reason they complained to the FTC in the first place,hoever, we have not been able to independently confirm that claim.
In any case, Nvidia did not achieve that goal, and do not appear to have any right to make a CPU that uses certain of Intel’s patents. Even if they managed to negotiate use of those patents, there is still AMD and VIA to contend with. Anyone want to bet that Intel, AMD, and VIA aren’t keen to let a new player into the niche that they occupy?
One big clue about how badly Nvidia lost is in the FTC settlement, under Section I. F., Other Definitions. It defines, “Compatible x86 Microprocessor” as, in part iii. as “that is substantially binary compatible with an Intel x86 Microprocessor without using non-native execution such as emulation”. Our bold. Hear that? It is the sound of the door to Nvidia’s dreams slamming shut.
This won’t stop Nvidia though, their one managerial tactic when confronted with something they’re not allowed to do is to simply do it anyway then figure out how to make it legal later. This is what happened with Rambus, the recent 2% DDR royalty rate seems to be quite a bit higher than more tactful companies pay, and that still only covers items shipping from this point on. Things already shipped are still in dispute with Rambus and in court.
Intel is unlikely to let something like an Nvidia x86 chip hit the market, much less stay on the market for a few years, if they feel patents are violated. Nvidia will point out that Transmeta was never sued by Intel, so their approach is ‘legal’. Unfortunately for Nvidia, Intel does not share either opinion; that Transmeta was ‘legal’, or that Nvidia using the same approach would be ‘legal’ either.
From what we are told, Intel had a strong case against Transmeta, but the company went under before it was filed, if it was worth filing at all. Patent litigation has a lot of risks involved, the largest one is invalidation of those patents by a jury of laymen. It is a minefield, and losing is not the only down side.
If Nvidia uses the same techniques, and we are told that that is exactly what they are doing, Intel likely has the case against them already built. In fact, all Chipzilla needs to do is dust it off, search and replace a few names and dates, and off they go.
One other tactic Nvidia IR and PR uses to ‘sell’ their spin to investors is that since Intel didn’t go after Transmeta, they won’t or, if the analyst is gullible enough, can’t go after Nvidia. Nvidia made this fight personal, and Intel isn’t about to miss any opportunity to put a boot in.
More importantly, any litigation will be patent based, not copyright based. Copyrights and trademarks can be invalidated if they are not defended, but patents can’t. If Nvidia is actually trying to produce and then market CPUs based on these theories, they are in for a painful surprise.
The take home message of this whole story is that Nvidia is building an x86 based on the next generation ARM core. There are technical reasons why it’s gestation will be problematic, but the legal roadblocks are more than likely fatal. That won’t stop Nvidia, it’s management is more than happy to waste hundreds of millions of dollars on a futile effort to boost their own egos. At the very least, now you know why their R&D expenses are so high, the bright engineers working on it aren’t cheap.S|A