NVidia announces x86 chip?

You call that announcing an x86 CPU? Gosh, Charlie must really be out of headlines. rendezvous' summary is pretty much spot on. The only thing they say is that it would make sense for MIDs and Netbooks in 2-3 years. I completely agree it implies they're interested in the market, but that doesn't even necessarily mean they're investing in it today - especially from an in-house R&D point of view. Heck, I suspect their handheld group would disagree that market makes any sense anyway given the R&D budget required.

It feels much more logical to me that this would indicate they've been talking with VIA about collaborating on a SoC on 28nm with a revenue sharing scheme or something like that. It's not impossible that they've got a project at the very early concept stage, but I'm not convinced. I'm pretty sure they had something in an early concept stage in 2H06/1H07 based on what Jen-Hsun said once, but given their change in tone later they probably canned it long, long ago.
 
I'd rather see them license those new out-of-order execution SMP ready ARM cores and integrate one (or more) into GPU architecture. Cell, your time is out.:LOL:
 
Charlie is so antagonistic that he makes himself look like a total fool.

Charlie said:
In any case, the real question is whether or not Nvidia will survive long enough for this to see the light of day. We have our doubts. Given that it was anounced at an investor conference, it is likely a shallow attempt to prop up the stock.

So he spends the first half talking about how he is right they are going to build an x86 chip, then the seconds saying they are just saying it to prop up stock and implying they are not going to build the chip.
 
Bit-tech is more thorough

http://www.bit-tech.net/news/hardware/2009/03/04/nvidia-reveals-plans-for-x86-cpu/1

In my view this piece makes it quite clear that NVidia is going to do x86. Indeed for a 3 year timescale NVidia would have to have the project underway already, I'd say.

An alternative, of course, is simply to licence x86 from Intel, say, and build a SoC with x86/Geforce all on the same bit of silicon. All done at TSMC.

The fact that Intel has just launched this capability sounds like quite an interesting match, no?

Jawed
 
Somehow, I doubt that Intel will give nv an x86 ip core or otherwise license. Going with VIA seems a more probable route
 
BTW, just thought I'd emphasize this: these statements are like 10x less conclusive than what Jen-Hsun said in early 2007 in a CC. And apparently not much came out of that, did it? And an in-house R&D program (versus a deal with VIA) makes no sense given Michael Hara's claim in the same CC that lowering OpEx further would compromise revenues, implicitly in the short/mid-term rather than speculative just long-term. This is all BS, sorry Charlie.
 
R&D into a possible entry of Nvidia into the x86 market would certainly make sense when combined with earlier statements by Jen-Hsun that they would greatly increase R&D funding this year.

And as Jawed said, if they are trying for something, they would have to have already started R&D on it...

Of course the counter to that is all the vitriol that Jen-Hsun has been saying towards CPUs in general and how their importance was waning.

Regards,
SB
 
Of course the counter to that is all the vitriol that Jen-Hsun has been saying towards CPUs in general and how their importance was waning.

On the other hand, unified shading was teh suck according to Nvidia all the way until they were first to the market with such a GPU.
 
On the other hand, unified shading was teh suck according to Nvidia all the way until they were first to the market with such a GPU.

Same thing with AA back then, remember all the "would you rather have AA and low res or a very high res with no AA" statements.
 
Follow-up.
A way around the Intel x86 license ?

If i understand it correctly, all the Cyrix-derived x86 IP is Intel licensing-free because it's clean reverse engineering work, and VIA can order production to any foundry with a previous x86 license (IBM, TI, etc), so... ;)
 
VIA has been using all kinds 'a foundries, TSMC, Fujitsu..

Maybe a Cel..GeForce with a Nano core (or two), some shader clusters, a HyperTransport bus and some GDDR5 controllers on it? :oops::D
 
@inkster wouldnt they then have to reverse engineer all the sse instructions
and how willing would amd be to license amd64
 
@inkster wouldnt they then have to reverse engineer all the sse instructions
and how willing would amd be to license amd64

VIA's Nano/CN x86 CPU already supports those at least up to SSSE3 and x86-64.
Nvidia could (like AMD with the only partially SSE4-compatible future SSE5, vs Intel's future AVX) just fork it from here, and use their experience in small floating point units to their advantage, i suppose.
The main concern, proper x86/x64 software compatibility, has been addressed already by Centaur/VIA.
What amazes me the most is that the Centaur subsidiary can design such a complex thing as the Nano CPU with a "mere" 100 or so engineers...
 
It's more understandable given how long it's taken that team to make a moderate-performance OoO x86 CPU, and how that design was rarely seen at all in the months (a year?) since it's been announced.
 
It's more understandable given how long it's taken that team to make a moderate-performance OoO x86 CPU, and how that design was rarely seen at all in the months (a year?) since it's been announced.

I believe they have been having issues with Fujitsu's 65nm process technology. Therefore, a move to TSMC for the 40/45nm dual-core version seems like a reasonable choice, given enough time to port it.
 
Back
Top