refreshes of DX11 GPUs in the works

[josiahsuarez]

first Fuad talks bout a Fermi refresh:

http://www.fudzilla.com/content/view/17387/1/
Nvidia works on second generation Fermi
Written by Fuad Abazovic
Monday, 25 January 2010 11:51

To come in 2010


If all goes well, the second generation Fermi architecture should be coming in late 2010. Since Fermi is Nvidia’s big transition that got delayed, it looks like Nvidia plans to have two new major generations in a single year. Traditionally the new chip comes in one year and roughly 12 months after the company is ready to launch faster version of the same chip, with more muscle.

We also learned that the mainstream Fermi won’t be significantly delayed from the original plan and we hope to see GF100-based GT300 single-chip card in March of this year.

By November Nvidia should be ready to announce its new architecture, but this time we only know that Fermi is the first step in new architecture development that should stay with us for a few years. G92 foundations lasted almost three years and we expect that Fermi and future Fermi chips in next few years to feature the same fundamental design.

After all this is what ATI and Nvidia have been doing for quite some time now. If Nvidia doesn't push second generation Fermi to a later date, you should see these chips by Q4 2010.

and a few days ago xbit mentioned what seems to be a 5 series refresh from ATI:

http://www.xbitlabs.com/news/video/display/20100121233418_ATI_s_Next_Generation_Graphics_Proc essors_on_Track_for_the_Second_Half_of_2010_AMD_s_ CEO.html
ATI’s New-Generation Graphics Processors on Track for the Second Half of 2010 – AMD’s CEO.

ATI Looking Forward to Refresh Entire Graphics Lineup in the Second Half of 2010
[01/21/2010 11:33 PM]
by Anton Shilov

Chief executive officer of Advanced Micro Devices said during a conference call with financial analysts that the company’s graphics division ATI was on-track to refresh its lineup of graphics cards in the second half of calendar 2010. The mystery, though, is with what the family of graphics processors will be renewed.

“We are ramping the ATI Radeon HD 5000 series now and look forward to refreshing the entire lineup in the second half of next year*,” said Dirk Meyer, chief executive officer of AMD, during quarterly conference call with financial analysts.

A little earlier this month AMD’s worldwide developer relations manager said that ATI hopes to maintain leadership position in terms of product performance “for the majority of 2010”. Still, it is not completely clear with what does AMD plan to refresh ATI’s lineup going forward.

The company has two ways of getting even faster products to market: it can redesign existing ATI Radeon HD 5000-series graphics chips so that to boost clock-speeds tangibly or it can introduce a brand new family of ninth-generation Radeon chips code-named Northern Islands, which is widely believed to feature a new architecture.

Not a lot is known about Northern Islands. Some sources claim that the new chips will be made using 32nm fabrication process, but the others believe that the new chips will be made on 28nm node. There are also reports that Northern Islands will have richer feature-set compared to Evergreen. According to some other reports, ATI’s next-generation family is code-named Hecatonchires and will feature code-named Cozumel, Ibiza and Kauai chips, which are not northern, but southern islands. As a result, it is possible that ATI may be working on two new architectures.

Earlier this week TSMC said that it would start mass production of chips using 28nm process technology in Q4 2010.

* Considering the fact that the call was dedicated to AMD’s financial results in the Q4 of fiscal 2009, the AMD executive referred to the second half of fiscal 2010 as to the “next year”, which is basically the second half of calendar 2010.

[Serrasalmus]

looks like my 285's are gunna hang till nov then......

[Viral]

Both articles sound very confused. I don't think we'll see any new architectures this year, just refreshes. It makes sense for a Fermi refresh in Nov, but 2nd half for a R5xxx lineup refresh? Hopefully a 5890 will come before the end of Q2.

[Johnny C]

Both articles sound very confused. I don't think we'll see any new architectures this year, just refreshes. It makes sense for a Fermi refresh in Nov, but 2nd half for a R5xxx lineup refresh? Hopefully a 5890 will come before the end of Q2.

5990.........(lee)

[LydianKnight]

Would be weird for NVIDIA to be working on a next-gen part when they already have one near releasing onto the public hands, sound more like 28nm process shrink with better clocks, lower consumption and a 512bit memory controller, more like that... adding the bits they currently lack (or could be lacking).

On the other hand, sounds far more logical coming from ATI, given their current series is a refresh of the previous one, with some bits added, for them now it's time to move next-gen, at least that's what I'm thinking right now...

[nekrosoft13]

Would be weird for NVIDIA to be working on a next-gen part when

why? they are probably working on 2-3 future cards

[LydianKnight]

why? they are probably working on 2-3 future cards

Based on what? Fermi is their latest R&D effort, and the logical thing to do is to extend that current technology with enhancements, like the ones I mentioned, which sounds logical at least (whether this is what NVIDIA is going to do or not, that's another completely different story).

The usual development cycle for both companies has been something like 18-24 months, not 12. What kind of new technology can you bring to the table if you're recently done with your R&D efforts?

Making huge leaps in the shortest ammount of time is a principle for bad scientists/engineers, and the industry/market is full of those examples, who usually end going vaporware or just being completely useless...

So I'll stick to what I said... feasible objectives for this coming process shrink (28nm) are memory controller widening, lower consumption levels and higher clocks. Other than that, I find very difficult to know what kind of step NVIDIA could be doing.

[cschueths]

Based on what? Fermi is their latest R&D effort, and the logical thing to do is to extend that current technology with enhancements, like the ones I mentioned, which sounds logical at least (whether this is what NVIDIA is going to do or not, that's another completely different story).

I don't think you realize how hardware companies work. They work on multiple generations at a time. Their R&D is already brainstorming improvements that we won't see for several years.

It keeps the ball rolling, plus it increases the production window for system software guys (like moi) to write device drivers way ahead of its release.

Even though next gen (after Fermi) hasn't even taped out yet, I'd be willing to bet money that the engineering dept at nvidia has been working out bugs and testing an emulated version of it for several months now.

[LydianKnight]

I don't think you realize how hardware companies work. They work on multiple generations at a time. Their R&D is already brainstorming improvements that we won't see for several years.

It keeps the ball rolling, plus it increases the production window for system software guys (like moi) to write device drivers way ahead of its release.

Even though next gen (after Fermi) hasn't even taped out yet, I'd be willing to bet money that the engineering dept at nvidia has been working out bugs and testing an emulated version of it for several months now.

Yes, I get your concept, but still... brainstorming, deep analysis, whatever, etc., isn't that fruitful for any company to produce a next-generation architecture just out-of-the-hat in 12 months, they're talking about having their second generation Fermi for late 2010, even if it's early/almost-mid 2011, it's still just 12 months, and both NVIDIA and ATI usually have a 18-to-24 month development cycle for any generation coming out (usually, that doesn't mean it has to be that way forever, but up until now, it has been like that), with 6-to-8 month cycle for refreshes, which actually makes sense for a Fermi refresh (March/April = 3/4+6/8 = 9/12 = from september to december = late 2010).

REFRESH, not next-generation. For NVIDIA it's too early to talk about their next-gen part when they're just finished designing the first lot of Fermi products, now comes the refresh. This could be seen as a mimick of Intel's Tick-Tock model. Fermi is NVIDIA's Tock, now comes the 'Tick' :P

So my point is still valid (in my opinion), if you can provide me with a link with any media revealing shorter development cycles for current technologies, please, provide me with it and I'll gladly rectify my own statement, but for now... I still think it's the planned time for Fermi (as in Fermi 1, let's call it that way), refresh-wise.

[onmikesline]

they have video cards that will blow anything out of the water for the next 10 years "prototypes", but they dont release them, because!!! thats now how company's work, they will not make money that way

[Redeemed]

they have video cards that will blow anything out of the water for the next 10 years "prototypes", but they dont release them, because!!! thats now how company's work, they will not make money that way

Not entirely accurate.

I mean, if that were the case they'd have a perfectly functioning model of Fermi well in advanced- and could have then just mass produced it in time to compete with the HD5800s.

We've seen that is not the case. :)

[fivefeet8]

Second Generation Fermi architecture sounds analogous to the changes from G8x to the current GTX2xx series. Basic same architecture, but with more performance and tweaks to the hardware in some cases. This should take far less time then a full redesign.

[Johnny C]

Second Generation Fermi architecture sounds analogous to the changes from G8x to the current GTX2xx series. Basic same architecture, but with more performance and tweaks to the hardware in some cases. This should take far less time then a full redesign.

I think nV should release the damn architecture....manufacture it for a while, see how it performs in the field before the silly ass marketing dough-heads even mention a refresh. Right now people there is abso****inlutely nothing to refresh......

It's like talking about a refresh of the toothfairy....or the easter bunny or a damn unicorn....

In fact we have more "facts and information" on all of the above make believe things than we do on Fermentation.....

[LydianKnight]

Second Generation Fermi architecture sounds analogous to the changes from G8x to the current GTX2xx series. Basic same architecture, but with more performance and tweaks to the hardware in some cases. This should take far less time then a full redesign.

That makes more sense than pretending having something completely new. It's perfectly feasible to tweak their current design within their current architecture. Higher clocks, some bits of improvement in other areas, less power consumption in less size due to the upcoming 28nm process shrink.

Maybe even not going that aggresively for lower power consumption but to keep it on lower levels and adding more shaders, like... 768 VS 512 (or even 1024 but double the number looks like too much for a refresh, at least right now in my mind).

Quite reasonable. +1

[mailman2]

they have video cards that will blow anything out of the water for the next 10 years "prototypes", but they dont release them, because they are too hot to cool

fixed.

[JasonPC]

From the sounds of it Fermi was in development well before GT200 was released to consumers... probably while GT200 was still in development. But drastic changes were made to the architecture during development which made some pretty big delays. We know they are already working on the next generation that will use 28nm fabrication. With technology if you aren't working on multiple generations at the same time, you're going to fall behind.

[gstanford]

Thats the way it always is. nv40 and g80 started development around the same time, nv35 was taping out as nv30 launched.

[shadow001]

I think nV should release the damn architecture....manufacture it for a while, see how it performs in the field before the silly ass marketing dough-heads even mention a refresh. Right now people there is abso****inlutely nothing to refresh......

It's like talking about a refresh of the toothfairy....or the easter bunny or a damn unicorn....

In fact we have more "facts and information" on all of the above make believe things than we do on Fermentation.....


Unless they know that the current performance of this first generation of Fermi isn't all that,and have accelerated the timetable for the refresh part,since ATI is on the record as having refreshes planned for the entire lineup in the second half of 2010?


It seems that only the 448 SP version of Fermi will be available in any sort of decent volumes,while the 512 SP version will be a rare sight,but these are just rumors for the time being,and if we look at the past,going back to the dark days with the NV30 as an example here,but given that it was also released very late and ultimately didn't live up to expectations,Nvidia only produced a relatively small amount of them,and released the NV35 based cards just a 2~3 months later,and if i remember correctly,one of the main improvements for the NV35 was a 256 bit memory bus,so that it would be on even terms with the 9700 pro and 9800 pro cards on a memory bandwith basis.


Things only really got back on track for Nvidia with the NV40 cards,but that was well over a year later after the NV30 was released.

[Lyme]

I feel for nVidia pulling a Intel with the little bee. Intel went through three+ refreshes before killing it, nVidia really has no choice but to release.

[kaptkarl]

Let see NVida actually release the Fermi cards BEFORE we start discussing a refresh, shall we?? Talking about putting the horse before the carraige.....geez.......ATI can talk of refresh because they already have the 5800 series released....NVidia w/ Fermi....not.

Kapt