[Lukfi]

So, it's a bit early, but still, NordicHardware claims to have some info about next-gen Radeon chips, which are supposed to be truly next-gen this time (not even loosely based on R600 anymore). So I took the liberty of creating a thread for it.

http://www.nordichardware.com/news,9997.html
http://www.nordichardware.com/news,9998.html

NH refers to the generation as "Radeon 100", there are supposedly three chips codenamed Cozumel, Ibiza and Kauai (as islands).

Launch date is Q3 2010 if AMD goes with TSMC and Q4 if they choose Globalfoundries.

[fellix]

Wow, that was fast one! And the current R800 speculation thread is still lurking around.

[IRQ Conflict]

Meh, 5870 was so last week.

[Voxilla]

Seems not very likely to me.
Next year there will be a shrink from 40nm to 28nm, but a complete new architecture seems kind of wild.
I don't expect this until DX12, maybe start a speculation thread about this as well

[fellix]

I don't think ATi spend that much R&D on R800 -- mostly a DX11-upgraded 2*RV770 single ASIC implementation.
Surely something much more grandiose is in the workshop behind the curtain.

[Silent_Buddha]

I suppose it could be possible they are doing a Rv8xx refresh in 6 months and then 6 months later launch a new card. But that seems really fast.

I suppose other possibility is they skip the refresh and launch a new card in 12-15 months, but that's a pretty long gap between cards.

Either way the timing just doesn't seem right...

Then again if Rv870 was delayed due to 40 nm problems and/or deliberately delayed to make the launch line up with Intel's Lynnfield and closer to Win7... Then a roadmap with refresh in 3-6 months (5890?) and then new product 6-9 months after that seems plausible.

Regards,
SB

[fellix]

There will be at least one 32/28nm node R800 "try-out" SKU before anything next gen to pop out on a finer manufacturing process, and if the competition environment is heating up that would be full-blown RV890, not just some mainstream puppy, I hope.

[Lukfi]

Originally Posted by Silent_Buddha

Then again if Rv870 was delayed due to 40 nm problems and/or deliberately delayed to make the launch line up with Intel's Lynnfield and closer to Win7... Then a roadmap with refresh in 3-6 months (5890?) and then new product 6-9 months after that seems plausible.

Regards,
SB

Reminds me of R520 and R580. If R600 had been ready, it would have been launched in August 2006. While it didn't, the gap between R600 and RV770 was roughly 13 months with a refresh (RV670) in between.

[rpg.314]

Bergman said a while ago that they'll have something revolutionary around that time, 2010, so the time line is about right.

I looked around, but the linky seems to have gotten lost somewhere.

[MarkoIt]

It believe they will revise the architecture, but it will be still based on the current VLIW architecture, which still has lots to show more and more games get shader intensive.
They will probably greatly increment the ALUs (looking forward a 3x), remove some fixed functions (RBEs and tessellation?) and they have to increment the bw efficiency as well.
The mainstream chips of R9xx will be probably be build for Fusion, so they will use a SOI manufacturing processes, but i have no idea how this can impact over the architecture.
But most important.. 28nm or 32nm? What kind of transistors density can we expect from both? And what about power savings?

[phenix]

Why so fast? Do they expect big things from Fermi?

[Scrat]

I'll stick to an optical die shrunk Cypress to 32nm in Q3 2010...

[Tchock]

According to nApoleon (still reliable as hell) originally R8XX was supposed to be the brand new (ugh) architecture from daamit. A while ago however that was pushed to R9XX.

But as we can see it didn't really affect their execution abilities though, and they might even be getting better in that segment (although it's easy improving from R600 piss poor )

[Silent_Buddha]

Originally Posted by phenix

Why so fast? Do they expect big things from Fermi?

Well if that single rumor is true, there's no way it can be a response to Fermi. You don't design a whole new architechture in half a year.

Regards,
SB

[ShaidarHaran]

Originally Posted by Tchock

According to nApoleon (still reliable as hell) originally R8XX was supposed to be the brand new (ugh) architecture from daamit. A while ago however that was pushed to R9XX.

But as we can see it didn't really affect their execution abilities though, and they might even be getting better in that segment (although it's easy improving from R600 piss poor )

R600 was delayed because of a bug in the 90nm design library. Not at all ATi's fault.

[fellix]

Originally Posted by Tchock

According to nApoleon (still reliable as hell) originally R8XX was supposed to be the brand new (ugh) architecture from daamit. A while ago however that was pushed to R9XX.

I wonder at what point (in time) that decision was made? It's known that the R&D process [both for NV & ATi] is virtually an uninterrupted time-line, and the work on each new design begins with a little overlap with the old one being just finnished...

[seahawk]

Originally Posted by phenix

Why so fast? Do they expect big things from Fermi?

You make such decisions long before you know about the chip your competition is going to deliver. And as ATI is executing quite well at the moment, I would not surprised to see them move quickly.

[no-X]

Originally Posted by ShaidarHaran

R600 was delayed because of a bug in the 90nm design library. Not at all ATi's fault.

That was R520.

[ShaidarHaran]

Originally Posted by no-X

That was R520.

Doh!

LOL, ok, R600 sucked then

[air_ii]

Originally Posted by seahawk

You make such decisions long before you know about the chip your competition is going to deliver. And as ATI is executing quite well at the moment, I would not surprised to see them move quickly.

Decisions about the design, yes, about leaking the info, not really .

[wingless]

Originally Posted by Voxilla

Seems not very likely to me.
Next year there will be a shrink from 40nm to 28nm, but a complete new architecture seems kind of wild.
I don't expect this until DX12, maybe start a speculation thread about this as well

Why doesn't it seem likely? From the 2900XT, to 3870, to 4870 and to 5870, the timspan has been almost exactly the same. The ATI division gets new chips out once a year.

[neliz]

tagline "DX11 done righter"

[phenix]

Originally Posted by Silent_Buddha

Well if that single rumor is true, there's no way it can be a response to Fermi. You don't design a whole new architechture in half a year.

Regards,
SB

Originally Posted by seahawk

You make such decisions long before you know about the chip your competition is going to deliver. And as ATI is executing quite well at the moment, I would not surprised to see them move quickly.

Originally Posted by air_ii

Decisions about the design, yes, about leaking the info, not really .

I am not saying it is the response to Fermi but it just seems rushed out. Does it leaves enough room for RV870? Besides ATI might had the information about Fermi long before Nvidia announcement. Well, everybody guessed that Nvidia will come up with a new architecture in 2010. Maybe they predicted that the market will be much more competitive in a year or so.

[MfA]

I don't expect huge architectural changes in the SIMDs ... slightly narrower would be nice for GPGPU but I don't think it would matter much for graphics, so I don't expect it happening. They'll probably chase NVIDIA in getting a writeable L2 and loosening up branching limitations.

[Arty]

I think its a bit premature, if we're speculating then we should be discussing some thing more sane like this -> AMD RV770 refresh -> RV790. So I guess either we turn this thread into a RV870 refresh thread or lock this baby down, which ever the mods feel is the better path.

As for this bit of "news", here is another shot at hardware-info's and nordichardware's credibility:

Originally Posted by Gipsel

I'm quite sure Cozumel, Ibiza and Kauai are not the code names for the mobile chips. They have actually no clue what the names stand for. They simply got it from a forum post (German) I've done on September 28th

The events were as follows:
I posted about some strange island code names (I wanted to tease some people) and a possible feature of the upcoming R9xx series and that there are already references to it in the driver. Furthermore that it took from December last year until now that we got the hardware after first references to RV870 und RV830 in the drivers. NordicHardware made up a news that R9xx is coming early in Q3/10. The German hardware-infos rumor site took this news and spiced it with the three island code names (claiming they are from the .inf files, which is wrong) and now NordicHardware just adds the code names to their rumor with the words that those names are coming from some beta driver (which is wrong, too). :rolleyes:

To say the truth, I don't know for for sure what these names mean either, but at least I think I may have a clue :rolleyes:

Anybody cares to know when these names actually appeared? Maybe Charlie has heard them already quite some months ago.