NVIDIA: New 90nm GF7s by January, Next-Gen in H1 2006, plus RSX NREs until mid-Q1.

[Geo]

I think Geo is about to blow a gasket with the all the possibilities that are rattling around his head.

About par for the course!

Changes don't cause me any particular concern --trying to resolve inconsistency drives me wild. :smile:

 

[dizietsma]

If I was nvidia I would not do too much with the 90nm top end part, at least in respect to increasing the number of pixel and vertex units. Instead I would try to get 90nm to make it a % faster whilst making it a % cheaper per die ( or per wafer, depending on how they pay TSMC, no idea ) and then just be happy counting the money rolling in.

 

[CMAN]

If I was nvidia I would not do too much with the 90nm top end part, at least in respect to increasing the number of pixel and vertex units. Instead I would try to get 90nm to make it a % faster whilst making it a % cheaper per die ( or per wafer, depending on how they pay TSMC, no idea ) and then just be happy counting the money rolling in.

That assumes the higher clock speed of the part would keep it performing competitively. They'd be better off targeting a particular yield/price per die and get the most performance out of that figure. Basically make the fastest processor for a given price point.

 

[kemosabe]

Edit: And I can't tell for the life of me whether there is another high-end G7x in 90nm coming in January or not. The first part of the answer says 'no'. The second part of the answer ("family of 7 in 90 nm") suggests 'yes'.

Another source that seems to confine NVDA's January 90nm launch to low-end and midrange parts. Could be that R580 will soon reign unopposed for a few months, as G70 did.

 

[Geo]

Another source that seems to confine NVDA's January 90nm launch to low-end and midrange parts. Could be that R580 will soon reign unopposed for a few months, as G70 did.

Nice one. Of course it also seems to say that R580 will only be announced in January and won't ship until March. Ish.

 

[Carl B]

Thanks for posting the synopsis Uttar - this is just what I needed (and quite timely as well) as I consider the branching of my upcoming PC/vid card upgrades. G70 derivatives into 2007, DX10 around mid-year 2006.

RSX's tape-out is also something of interest to me, and it amazes me to think that indeed it may not have taped-out yet. I mean that would make it a full year after Xenos in terms of tape-out, though the two really can't be compared in that sense since they come from different architectural paradigms. RSX better be *quite* the chip though.

With CES in early January and the PS3 rumored to be featured there, makes me wonder if they'll just have a mock-up or whether they might actually run on some early silicon. If tape-out hasn't already occured, the later seems unlikely.

 

[dizietsma]

Nice one. Of course it also seems to say that R580 will only be announced in January and won't ship until March. Ish.

Maybe that indicates the announcement will be in very late Jan and the rollout in very early March to give the 30 day lead ? The item seems to indicate they will have better availability than X1nnn

 

[_xxx_]

RSX better be *quite* the chip though.

Well, we know what it is - G70 on 90nm process.

 

[Jawed]

What's this on Digitimes?:

an upstream component shortage has led to tight supplies of ATI’s GPUs including the Radeon 9250, X300, X550 and its recently launched Radeon X1000 family.

Jawed

 

[Carl B]

Well, we know what it is - G70 on 90nm process.

I doubt it's a straight G70@90nm, y'know? There have to be at least some changes to the logic to make it work with the FlexIO and such.

Now I'm not saying that it will deviate from the G70@90nm greatly - traditionally I've said just the opposite - but I am saying that it better be quite the chip!

 

[Geo]

Maybe that indicates the announcement will be in very late Jan and the rollout in very early March to give the 30 day lead ? The item seems to indicate they will have better availability than X1nnn

Well, I'd had hopes that they were intending to crimp down the 30-day thing, based on Orton's comments over the summer. They seemed to indicate that something close to same day availability was now the goal, abandoned temporarily for the X1k family introduction by the fact that they felt they needed to do it all at once because of the delays to give people a view of the entire family's performance at the same time. Whatever you think of that reasoning, it is what he said.

 

[kemosabe]

What's this on Digitimes?:

Jawed

I've stopped trying to figure out the various substrate/component/other shortages which seem to have afflicted ATI's production chain since last year. DigiTimes offers these tiny morsels every few weeks but never elaborates, and nobody at ATI seems to discuss the matter much during the CCs so I suppose it's either not very serious or they prefer to keep it neatly tucked under the rug.

 

[Hanners]

What's this on Digitimes?:

an upstream component shortage has led to tight supplies of ATI’s GPUs including the Radeon 9250, X300, X550 and its recently launched Radeon X1000 family.

Jawed

Upstream component shortage? Sounds like they have some mystical river running through their factories that delivers the cores to them.

 

[dizietsma]

Well, I'd had hopes that they were intending to crimp down the 30-day thing, based on Orton's comments over the summer. They seemed to indicate that something close to same day availability was now the goal, abandoned temporarily for the X1k family introduction by the fact that they felt they needed to do it all at once because of the delays to give people a view of the entire family's performance at the same time. Whatever you think of that reasoning, it is what he said.

Yes, you'd think they'd want to get it down to match nvidia but maybe nvidia will go back to 30 days and the only reason why they managed it this year because of Ati's chronic delay.

I'm going to guess then that reviews will be the last week of January and the cards on the floor will be 3rd week of Feb .. lets see how close that will be.

Not many GTX 512MB's at present, looks like they were select chips and not just extra volts plus better cooling .. Ati's Powerpoint might be right on that topic,

 

[trinibwoy]

Not many GTX 512MB's at present, looks like they were select chips and not just extra volts plus better cooling .. Ati's Powerpoint might be right on that topic,

You're dismissing the memory shortage possibility?

 

[Sxotty]

What's this on Digitimes?:

Jawed

BS?

Or not, but what components does ATI need that Nvidia does not?

 

[Megadrive1988]

That also could mean the RSX is closer to their next gen part than previously thought.

every indication so far is that is not the case. RSX seems to be firmly based on NV47 / G70 with some custom pieces so it can work with Cell as effectively as possible, plus a speed boost over the original G70. the pipeline architecture, shader units, etc, are all pretty much the same as G70, AFAIK.

with that said, it would be nice if PS3's GPU was based on Nvidia's next-gen architecture (Nv50) with SM4.0 and the advanced features that i'm sure Nvidia is packing into Nv5x. that would somewhat future-proof the PS3, which is going to have to last until 2012.

in my ideal senareo, RSX would be a super tricked out, beefed up, souped up NV50. ...more specifically, an NV50 refresh. much like G70 is a refresh of NV40. heck, even the first Xbox GPU was a souped up NV20. it seems that PS3 GPU will be behind the curve whereas the original Xbox GPU was somewhat ahead (compared to the best PC GPUs i.e. GeForce 3 Ti500)

 

[Farid]

Well, we know what it is - G70 on 90nm process.

We do?

 

[Farid]

with that said, it would be nice if PS3's GPU was based on Nvidia's next-gen architecture (Nv50) with SM4.0 and the advanced features that i'm sure Nvidia is packing into Nv5x. that would somewhat future-proof the PS3, which is going to have to last until 2012.

What real visual differences is there between SM4.0 and Sm3.0 (Or SM2.0 for that matter)?

Other than the probable (or not) inclusion of a true Geometry Shader, I see nothing exciting about SM4.0 from a visual standpoint, that is.

 

[no-X]

We do?

We know, that nVidia prepares something based on G70 architecture. It will be faster than 7800GTX 512MB. This chip will be used as a competitor for R580 and will be based on 90nm or 80nm process. It will have more than 24 pipelines total. Am I wrong?