NVIDIA Developing x86 CPU?

Nvidia has to have a total platform to compete with Intel and AMD. Without a platform, Nvidia would be at a serious disadvantage. The only real big issue with this "rosy" picture is : "How will we get a license from Intel?"
 
I think nVidia buying out Via is the likely route. It gives them something to start with and build up on. The reason Via CPUs are...well, not so good, is because Via purposely focuses on the low-end, low-power market. With nVidia at the wheels they could likely ramp up the power enough to fit their plan (weaker but still powerful CPUs and extremely powerful GPUs) and have a platform to work with. Hell, they'll even get a bit of S3s technologies, not that they really need them...but who knows, S3 is aimed in the same direction all of Via is, so perhaps they have a diamond in the rough and just don't even know it.
 
Is this the can of whoop-ass they're going to open on Intel? Considering maybe making an x86 cpu in 2 - 3 years? If not maybe someone can clear up exactly what that "can of whoop-ass" remark was referring too.

That can is NVIDIA filling for Bankrupcity in 4 to 5 years. Hope not but It's what I am seeing.
 
That can is NVIDIA filling for Bankrupcity in 4 to 5 years. Hope not but It's what I am seeing.

I'm rather ignorant on the matter so I'd appreciate some info. How did they get into a state where bankruptcy is 5 years away?
 
I'm rather ignorant on the matter so I'd appreciate some info. How did they get into a state where bankruptcy is 5 years away?

I think the question is how can they get there? maybe by making a long series of dumb decisions and products that end up taking a successful company and running it into the ground.
 
Of course it would be nice to see something new by nVidia. That would also make more competition to the market and other possibilities. I came to think about something more like a GPGPU thingy.
 
The biggest hurdle in that scenario: NVIDIA will have to figure a way to make CUDA a widely-used language, or they will have to abandon it and concentrate on OpenCL.
Trying to push the use both languages seems counterproductive, unless there's a way to combine the two seamlessly.
Again, I'm not a programmer, but it's my understanding that OpenCL is a lower-level language than CUDA.

Wrong actually.
OpenCL is based on Cuda, but is less hardware-dependent, and as such it is less low-level.
Basically the OpenCL people took Cuda as their guideline.
It's very easy for nVidia to support both OpenCL and Cuda, and this is what they're going to do. Expect beta drivers with OpenCL support within a few months.
ATi on the other hand was not so lucky. Their initial CTM + Brook+ stuff was not that hot. So what they did recently, was abandon it and use new technology they dubbed Stream. Basically it was based on a draft of the OpenCL standard, and now that OpenCL is finalized, ATi will adapt its Stream to make it OpenCL-compatible. So ATi just started over from scratch a while ago.
 
hrmm, wonder which x86 licence holder they are going to buy out.. :)

Won't work, the x86 license cannot be transferred to another company in case of a buyout or anything (in fact, technically what AMD wants to do with splitting R&D and manufacturing into two companies is not even allowed).
 
AMD licensed out their x86-64 extensions to Intel correct? Without them, their CPUs aren't 64-bit compatible. So at least its not a TOTAL one-sided relationship.

Well, technically it is.
The x86-64 extensions are just that: extensions.
You need x86 in order to have any use for these extensions. And AMD can only get x86 by licensing it from Intel. Which is done through a cross-licensing scheme (basically it says "You can use x86, but in return we get to use any extensions you make to it"), so if AMD licenses x86, Intel will automatically get free use of x86-64.

So basically there's no way for AMD to leverage their x86-64 extensions.
The only way AMD can stop Intel from making 64-bit compatible CPUs is to completely stop making x86-compatible CPUs altogether.
It still favours Intel heavily. Whatever AMD does, it hurts AMD more than Intel.
 
It's possible that Intel has a completely separate implementation of x86-64 from AMD's implementation.

It's pretty obvious that Intel's implementation is completely separate from AMD's.
Intel and AMD never share architectural/implementational information. This is why a Core2 or Core i7 doesn't perform the same as a Phenom clock-for-clock. They're different architectures, different implementations (but all three implement the same instructionset).
In a way, x86-64 is little more than extending your execution units and registers to 64-bits and extending your instruction decoder to understand the new 64-bit machinecode. It is a variation of your 32-bit architecture/implementation by definition. The CPU basically still works exactly the same, it just operates with 64-bit words rather than 32-bit.

In the Pentium 4 days Intel actually literally did this. They took their Northwood core, and extended it to 64-bits, resulting in the Prescott. I recall a die analysis where they actually pointed out that the ALUs were implemented in two 32-bit parts, in slightly different locations on the die. So it really was the original 32-bit ALU of the Pentium 4, with an extra 32-bits tagged on later. In a way it's amazing the Pentium 4 works so well in 64-bit :)
 
Hasn't Nvidia spent the last 9 months or so very publicly ranting about how Intel is clinging to a decaying market and that GPUs are the future of computing?

If I was in Intels Marketing dept I'd be rubbing my hands together with glee right about now.

Then I'd kill myself for being a marketing sleaze.
 
Hasn't Nvidia spent the last 9 months or so very publicly ranting about how Intel is clinging to a decaying market and that GPUs are the future of computing?

If I was in Intels Marketing dept I'd be rubbing my hands together with glee right about now.

Then I'd kill myself for being a marketing sleaze.

Actually do your homework. It started with an Intel engineer at IDF claiming that they were going to make GPUs obsolete. So of course NVIDIA attacked back.
 
Actually do your homework. It started with an Intel engineer at IDF claiming that they were going to make GPUs obsolete. So of course NVIDIA attacked back.

Intel might very well make GPUs obsolete, if their Larrabee becomes a success.
That does not mean that nVidia should attack back, let alone that they should talk smack about CPUs.
If Larrabee takes off, it will redefine both GPUs and CPUs as we know them, and it will make both 'obsolete' in a way (although technically it just seems to be the next evolutionary step).
 
They can just design and build a new CPU/GPU that can execute x86-64 instructions in a hardware/microcode virtual machine at near full speed. No need to make an actual x86 cpu.

Why would you, its crap, always was. Don't confuse market success with technical dominance. Alot more went into Intel's platform coming out the winner than the technical/engineering aspects of the device.

At the point of the early 80's the 68000 was a vastly superior DESIGN, but in the hands of a totally disfunctional company with zero managment interest in the CPU business in reality.

They put no real effort into the game, and effectively gave up by 1988, and Intel won by default. Intel won most importantly due to the worst technical/engineering desicion in the history of mankind... IBM Boca's selection of the 8088 over the 68000 in 1978 to base the PC on... because Motorola's managment having ZERO insight and vision didn't bid competitively for the contract, so Intel said they'd make the chip available for a reduced high volume price, and Motorola said... it's $X, and that's the price, take it or leave it.

The Motorola mangament and board should have been shot for incompetence blowing that opportunity. The rest is history.

If even a fraction of the R+D money Intel put into the x86 had been unleashed on the 68000 family, it would have crushed anything the x86 trash has accomplished.

Hell, you could use 45nm tech today, and squeeze THOUSANDS of original 68000 cores onto a die, and run them at 4-6 Ghz and redefine "SuperComputer".

Most of the foundation patents have expired, so NVidia can make a clean-room clone workalike of the x86 any time they want to bother.

Like I said, when Steve put up the story of Intel's guy yapping about NVidia not being complete by not having a CPU, all NVidia needs to do is use thier design expertise (which we can all agree is fucking hardcore) to make a competing CPU.

The only reason not to start from scratch is so much x86 software out their, but if you wanted to make it a dual commitment and take Linux under their wing to...make it a real consumer OS and run it on a seriously improved CPU platform... they could kick the legs out from under Intel.

Again, Intel exists and thrives due to far more circumstances than the TECHNICAL aspects of thier product. The technical end can be defeated by anyone with sufficient resources.

Google could join forces with NV and literally take over the computing world in a couple years if they really wanted to start a war with MS Intel etc.
 
David Kantor at RWT made an interesting point: nvidia has already acquired an x86 design when it purchased ULi (386 class). http://www.realworldtech.com/forums/index.cfm?action=detail&id=97058&threadid=97044&roomid=2 It's not really suitable for embedded or any kind of modern use though.

Read the whole analyst question and nvidia answer that's quoted in that post. In context there doesn't seem to be a current nvidia plan for making any x86 CPUs. It's more of an eventuality. Nothing to see here. The broken telephone game seems to only need 2 people now.
 
Back
Top