Slashdot Mirror
Nvidia Rumored To Be Readying X86 Chip Release
jdb2 writes with the (honestly labeled) rumor from the Inquirer "that Nvidia is preparing to release an x86 microprocessor with its guns targeted directly at its two major rivals — Intel and AMD/ATI," and excerpts from the just-linked Inquirer article: "THE HOT RUMOR going around IDF ... [is] that the company will do an x86 part. The background whispers say that the part will be announced next week at Nvision ... Nvidia's men in white coats certainly have the brainpower to do it, but they also most certainly don't have a license to sell such a part. NV is basically locked out unless Intel and AMD both decide to be magnanimous, and we would not recommend holding your breath waiting for this to happen ... That leaves the lawsuit option open ... Any attempt to enter the market without a license would bring down Intel legal on them like flying monkeys blackening the sky. It would get ugly. Really ugly. Expensive too.""
How is it that AMD is able to release x86 chips, but nVidia can't without a license from Intel? Why would nVidia need AMD to be gracious?
Does VIA has a license to make x86 processors?
And last week The Inquirer reported that nvidia was getting out of the chipset business.
http://www.theinquirer.net/gb/inquirer/news/2008/08/02/nvidia-chipsets-dead
So according to the inquirer nvidia is getting out of the chipset business, but is going to produces an x86 processor. I guess they'll have to hook up to an intel chipset...
For some reason I do not believe what the inquirer is writing...
They could pull a Transmeta and build a RISC/VLIW core or six and package it with an x86 interpreter or JIT translator, basically do the front end in software instead of hardware. Crusoe was using the same core to do the translation and execution, but with a multi-core CPU that pipelines the translator and interpreter on separate cores they could end up with quite a nice design.
Sure, if Nvidia tried selling x86 chips in the US or Europe, the company would get its ass sued off. But what about China? What about India? What about the third world? Merely because Intel has a rock solid patent portfolio in the US does not mean diddly squat in Bangladesh.
If someone says he and his monkey have nothing to hide, they almost certainly do.
...like Transmeta had. If they have a few chips that can convert and run converted x86 instructions quickly, that would work too. Provided it runs better than Intel or AMDs stuff.
Doesn't IBM and VIA (from Centaur and Cyrix) have x86 licenses as well?
Cyrix, Texas Instruments, IBM, NexGen, amongst others.
Other companies made clone x86 CPUs as well (The list: IBM, NEC, AMD, TI, STM, Fujitsu, OKI, Siemens, Cyrix, Intersil, C&T, NexGen, and UMC). Intel has never been really successful at prosecuting anyone for creating their own x86 compatible CPU. They won't sue, unless the company is small enough to just give up (Hint: nVidia isn't).
The idea of Nvidia producing an x86 CPU might seem dubious but perhaps not in the light of the fact that Nvidia bought Stexar in 2006. Stexar was a little known and quite secretive startup composed of a large portion of ex-Intel engineers and higher-ups from Intel's Xeon team. Before being swallowed by Nvidia they were intimating that work was being done on some sort of x86 "DSP".
jdb2
Nvidia also denied that rumour vigorously, going so far as to demand a retraction of the story (from the news site, not Slashdot...). As typical, everyone seems to have caught the rumour and completely missed the denial.
Right. Because when companies issue a denial of a rumour, they're always telling the truth. They'd have no reason to cover up some new product they're not quite ready to release ...
Oh, never mind, this is Nvidia, not Apple. Carry on.
My blog
Look guys, here's the facts: Nvidia will be releasing a chip. It won't be x86, though, it will be ARM based (with an fpu and vector unit), running around 1GHZ or more. A couple months ago, ARM Holdings announced a major license agreement (but didn't provide any other specifics). There was a lot of speculation that it was Apple. It's Nvidia.
My source didn't tell me if it's going to be targetted at smart phones, internet tablet pcs, etc.
NVidia hired a guy who was SGI's Chief Engineer last November. He's also done some other very interesting things.
I've worked with him directly in the past (as well as indirectly at SGI), and he is one of the smartest people I know.
If NVidia management lets him be useful, then I think we'll see some interesting things coming out of NVidia. It beats me what their management is like though.
This is really disappointing, because I'm rooting for AMD and their graphics efforts. And because NVidia is well known for their binary blob approach.
Yes, but also a few months ago Nvidia was denying vigorously that raytracing has any future, right up until a week ago when they showed off raytracing on their GPUs.
Behavioral functionality needed to emulate an x86 is non-trivial. Ask Transmeta.
A ton of bricks awaits them, should they or others try. Look again historically at Transmeta.
---- Teach Peace. It's Cheaper Than War.
I would like to see them put an x86 in the chipset that could work as a High Performance RAID controller if you have a CPU installed. If there is no CPU detected then make the x86 in the Chipset function as the Primary CPU. Make the performance equivalent to a VIA C3 CPU or maybe less. Make it like a Pentium 3 500~1000 Mhz processor.
That way they could sell their motherboards/chipsets as both consumer end devices and users of embedded systems could use the same chipset/cpu without having to buy an additional CPU for their kiosks/terminals/industrial automation.
I would love to be able to boot without a CPU if I needed to flash BIOS, test hardware etc....
Since the Chipsets these days on some motherboards already work as a sound card/bus controller/network card/video card/RAID controller/USB controller/SATA Controller/IO Controller/Memory Controller. Why not add a rudimentary CPU while we are at it.
Tsukasa: All I really want, is to be left alone...
See for yourself http://www.engadget.com/2006/10/19/nvidia-has-x86-cpu-in-the-works/
Seems easy to bypass the X86 issue. Create a full CPU using the X86 instruction set. Remove anything and everything related to 286 protected mode (keep "real" and 386 "protected" modes). Optionally, remove ring 1 & 2 from 386 protected mode, but keep the register format the same (windows and unix only use ring 0 and ring 3). Then, add a new CPU instruction or two that would really boost the performance of Nvidia's graphics drivers, which Nvidia can autodetect and use in their shipping drivers (just like most graphics drivers used to detect SSE and the like). Naturally, no one else would use these instructions, but Nvidia could be a good citizen and document them.
The resulting chip wouldn't be X86, because all X86 code does not run. The result would be a new chip that isn't backwards compatible. Let Intel bark and moan all day long in their marketing that the chip isn't X86. All Nvidia has to do is make sure it runs Windows just fine without a new SKU from Microsoft (is it Intel's fault that MS doesn't use 286 protected mode? Is it Intel's fault that MS doesn't use ring 1 or ring 2?).
There would still be a lawsuit, and it would be *wise* to ensure that your legal team is well funded. But it seems most legal arguments are letter of the law these days, and the subset and extended X86 is definitely not X86 (you can produce code that works on X86, but would fail on this, you can produce code that works on this but fails on X86).
This would be a ballsy move for Nvidia, but seems right up their alley.
Why not combine Intel/Nvidia?
Ave Molech Setting
Unless someone with deep pockets thought they might have a case with illegal cartel price fixing as the basis. If AMD and Intel can dominate the market by exclusively licensing to each other but not to other parties, that would start to inch pretty close to market price fixing and collusion.
Not saying this is what is going on, but it might be. If they had approached them and tried to license and got told a flat no, at any price, they would have at least a good enough case to bring suit then.
AFAIK, the VIA Nano supports all of those four instruction sets (Intel 64 and AMD64 are basically the same ISA). None of them are specific to Intel or AMD anymore.
Check out my sci-fi/humor trilogy at PatriotsBooks.
Given legal and licensing issues, it makes sense to work around this issue with a RISC / VLIW core (and NVidia has already mastered this) with a JIT or x86 bytecode interpreter at the front end.
Intel (and everyone else) has been doing this, to an increasing degree, since the 486. It's one reason for the long long Pentium pipeline.
Transmeta made the JIT translator software. That turned out to be not such a good idea, with one core trying to execute translated code and translate new code at the same time. With multiple cores (and nVidia has a lot of experience with heavy parallelism like that) this could actually work well.
My comment here was idle speculation along these lines. But apparently nVidia has been speculating less idly: they've licensed the Transmeta technology.
Sure, NVidia don't have x86 patent leverage, but they have GPU patent leverage. What with the Intel's GPUs (current and futuer) and AMD's purchase of ATi, they most certainly have patent leverage. Is it enough? I would guess that it is.
SJW n. One who posts facts.
Here is an off the wall idea to address the x86 licensing issue. Why doesn't Nvidia (7.81B) simply buy AMD(3.46B)? Might be one interesting way to compete with Intel(131.5B). This would allow Nvidia to get the x86 rights and put some pressure (is that really possible) on Intel by doing what Intel did in the CPU space except coming at it from the GPU side (by combining ATI's assets). And since the cross licensing agreement between AMD and Intel doesn't cover GPUs (just guessing here), Nvidia would have access to what Intel does in the CPU space w/o having to give up it's GPU IP.
Of course this doesn't make sense if Nvidia has a better CPU than either AMD and Intel. Anyway ,CPUs are becoming less relevant than GPUs.
One hundred million dollars would not even cover the P4 design team's salary for 1 year. Intel's R&D costs were about 2.5 billion in 1998 and 6 billion in 2007. AMD's was almost 2 billion in 2007. The bar in this field is very high.
Sockets are somewhat easier to patent than instruction sets, though, as they represent an actual invention, while an instruction just represents a fact---if I see this bit sequence, I perform the following mathematically definable operation. While I don't doubt that there are plenty of patents on instruction sets, I do doubt that any of them would be upheld in court if push came to shove and the opposing counsel were competent.... The only reasonably strong instruction set patents are the ones that define the hardware needed to implement them. If, however, you can find a way to implement them with a hardware implementation that looks sufficiently different, though, you've just worked around the patent.... The ones that merely attempt to define what the hardware should do if presented with a particular instruction are pure comedy, as they are a pretty blatant example of purely algorithmic patents, which aren't allowed in the U.S.
Check out my sci-fi/humor trilogy at PatriotsBooks.
The Silicon valley chess game is always an interesting one to watch. Pretty much everyone agrees nVidia must make some strategic move in order to survive in the market. Discreet GPU cards and mobo chipsets aren't enough to drive the company in the long haul with Intel and AMD both trying to integrate good performing GPUs into the x86 CPU.
AMD made a strategic financial blunder acquiring ATI. nVidia is likely not working on their own x86 design, but watching, anticipating, the continued downward financial spiral of AMD, waiting for the right moment for a hostile takeover. That, IMO, is the only real likely way that nVidia will break into the x86 CPU business.
Not to mention if you looked at it logically getting out of the chipset business would be majorly stupid coming so soon after the mobile chip mess. They need the chipsets for SLI,and we all know that the margins on the high end cards(which is what most gamers use in SLI) is a lot better than the low and midrange markets. With SLI they have a chance to either sell two high end cards at once or to sell a second high end card when it drops down to mid tier. Either way it is a good market to be in,and quitting would also give a boost to AMD who could brag that only they can tame the most graphics intensive games thanks to their HD48XX series with Crossfire.
My guess,based on years of watching hardware,is simply this: Nvidia had a bad run,which included their newest chipset. This is no different that the bad run that Maxtor had in 2001,or the bad caps that plagued Dell in the late '90s. When you are cranking out that kind of volume and are always looking for ways to lower production costs there are bound to be missteps along the way. And it isn't like Nvidia has never had a bad product before,simply look at the FX5xxx series,also known as the leafblowers. With poor PS 2.0 support and an incredibly noisy fan they were loud and lousy. But just like any major company they learned from their mistakes and came back with the excellent 6xxx line. I believe we will see the same thing here. They will lick their wounds,be lower on the radar for a little while,and then come out with an 8xxx based motherboard NForce series that will in all likelihood kick butt. Whether it will come with support for a new x86 CPU from Nvidia is anybody's guess,but either way I just don't see them leaving the chipset business for awhile,not until and unless they can come up with an on chip solution that will allow any board to run SLI. But as always this is my 02c,YMMV
ACs don't waste your time replying, your posts are never seen by me.