AMD: R8xx Speculation

[dizietsma]

In every press and analyst session we've done since Septermebr 10 we've showed a strategy / roadmap slide that showed they are 2010 parts.

Can you remind me what they showed for 5890, I have seen them but can't find them at the moment.

 

[neliz]

In every press and analyst session we've done since September 10 we've showed a strategy / roadmap slide that showed they are 2010 parts.

Some OEM's beg to differ: http://twisted-reviews.com/ati-radeon-hd5xxx-shortage-leads-to-delay-of-hd5650-based-notebooks/

As we recently mentioned here, Acer had plans on releasing at least two notebooks based on Intel Core i7 and ATI’s newest Mobility Radeon HD5650. Few days later it became clear that there is heavy shortage of HD5xxx chips and videocards, which would apply also to the mobile graphic cards by ATI. Our guy at Acer commented that we can expect a delay of the new HD5650-based models, although he couldn’t say how much we’ll have to wait.

According to other sources the delay will be at least to Q1 2010, which is rather disappointing, considering how ATI has the edge in desktop graphics and still lags on the mobile market (although much has changes with Mobilty Radeon HD4xxx series, NVIDIA still seems to have better range of chips in late mid and high-end notebooks), where those new Radeons would shine.

And this roadmap has the Q3 RV870 as a Q4 part..

Maybe I'm just confusing the mobile parts and low-end parts. but Q4 for both of these was feasible.

 

[Dave Baumann]

Sept 23rd is as close to Q4 as makes no difference - its hardly worth putting an additional sliver of a Q3 line for the sake of 7 days. As I said, we've consistently been promoting these as 2010 parts.

 

[Groo The Wanderer]

FWIW, I have been hearing the same line as Dave is saying, from long before there was an official word. Simply no one has said these things are Q4.

-Charlie

 

[Silent_Buddha]

Yeah, I've been expecting the Rv810/830 (equivalent) parts in Q1 2010 ever since summertime. Not sure where anyone got Q4 2009 for those parts.

Still waiting on those as I've been wanting to upgrade my 3450 in my HTPC for quite a few months now.

Regards,
SB

 

[mczak]

In every press and analyst session we've done since September 10 we've showed a strategy / roadmap slide that showed they are 2010 parts.

These may be Q1 2010 parts, but can't you start leaking some information about them .
I think the only new card I was expecting this year which didn't show up so far is the HD5870 EyeFinity (or whatever it will be called) edition.

 

[Squilliam]

I was wondering, given that AMD produces their CPUs on an SOI process going forward and given that they have their Fusion line of processors due to release in 2011 whether they are likely to release a 32nm SOI refresh of their current R800 lineup Q1/Q2 2010?

According to wikipedia the two main benefits of the process are:

* Lower parasitic capacitance due to isolation from the bulk silicon, which improves power consumption at matched performance.
* Resistance to latchup due to complete isolation of the n- and p- well structures.

The extra cost involved in using the process shouldn't be problematic for margins as AMD really gets paid for the performance of the chip relative to the competition and if it allows them to produce a chip which uses less power on a smaller die size relative to TSMCs 40nm process they should net a competitive advantage using the technology relative to Nvidia as they should be able to extract more performance at a lower TDP which is perfect for their upcoming mobile parts and it should net them faster desktop chips to compete vs Fermi, especially as it may allow them to release a full speed Hemlock variant.

So are we likely to see them transition to this process node? Im especially curious for their mobile parts scheduled to release in 2010 based on the Juniper chip which uses a 128bit bus when many have speculated that it could have used a 192bit bus for better performance. Perhaps they chose to use the smaller bus because they have a shrink in the works which would have otherwise been pad limited with a wider memory bus?

Juniper die size: 180mm^2 (126mm^2 @ 32nm with die shrink? @ estimated 70% scaling)
Cypress die size: 334mm^2 (234mm^2 @ 32nm with die shrink @ estimated 70% scaling)

 

[Silent_Buddha]

Depends on cost, and other than OEMs with low end budget chips, it's only really the tech savvy (IE - very small subet of the overall market) that will make note of power consumption numbers.

I don't think it'd be all that cost effective, especially since I believe the entire chip would have to be redesigned with SOI in mind?

Regards,
SB

 

[Grall]

Depends on cost, and other than OEMs with low end budget chips, it's only really the tech savvy (IE - very small subet of the overall market) that will make note of power consumption numbers.

In the portable marketspace it surely has to make a sizeable impact, there OEMs definitely care about power dissipation even if consumers are clueless about such issues.

 

[Kej]

I was wondering, given that AMD produces their CPUs on an SOI process going forward and given that they have their Fusion line of processors due to release in 2011 whether they are likely to release a 32nm SOI refresh of their current R800 lineup Q1/Q2 2010?

According to wikipedia the two main benefits of the process are:

* Lower parasitic capacitance due to isolation from the bulk silicon, which improves power consumption at matched performance.
* Resistance to latchup due to complete isolation of the n- and p- well structures.

The extra cost involved in using the process shouldn't be problematic for margins as AMD really gets paid for the performance of the chip relative to the competition and if it allows them to produce a chip which uses less power on a smaller die size relative to TSMCs 40nm process they should net a competitive advantage using the technology relative to Nvidia as they should be able to extract more performance at a lower TDP which is perfect for their upcoming mobile parts and it should net them faster desktop chips to compete vs Fermi, especially as it may allow them to release a full speed Hemlock variant.

So are we likely to see them transition to this process node? Im especially curious for their mobile parts scheduled to release in 2010 based on the Juniper chip which uses a 128bit bus when many have speculated that it could have used a 192bit bus for better performance. Perhaps they chose to use the smaller bus because they have a shrink in the works which would have otherwise been pad limited with a wider memory bus?

Juniper die size: 180mm^2 (126mm^2 @ 32nm with die shrink? @ estimated 70% scaling)
Cypress die size: 334mm^2 (234mm^2 @ 32nm with die shrink @ estimated 70% scaling)

It seems like there will be no production on GLOBALFOUNDRIES 32nm node until second half of 2010.
See page 6 on this PDF from AMDs Financial Day Earlier this month.
GLOBALFOUNDRIES Presentation

 

[Squilliam]

Well that settles that!

 

[Alexko]

I don't know, GloFo also said that they would only start production in 32nm in Q3'10 because AMD didn't need them to start any sooner, even though they could.

But what if AMD did need them to start sooner?

 

[rjc]

I think Acer has the first DX11 notebook out, the 8942G:

http://www.webdream.de/produkt-reviews/acer-aspire-8942g-das-flaggschiff/
Aber auch alle genannten Details bekommen jetzt noch ein Sahnehäubchen verpasst, auf das alle gewartet haben: Das neue Flaggschiff von Acer zeigt sich grafisch von seiner allerbesten Seite und brüstet sich mit einerATI Mobility Radeon HD 5850 Grafikkarte mit 1024 MB RAM.

It's a high end desktop replacement - 3 variants are supposed to come through December.

According to the rumors originally it was supposed to have a 5670 but Acer could not secure units in time to release the model before christmas...if it's true that the 5850 got substituted at the last minute imagine it will be downclocked to similar thermal profile/power usage of a 5670 so nothing else on the notebook needs redesigning.

(Note the 5850 mobile is Broadway = Juniper, 5670 mobile is Madison = Redwood)

 

[Kaotik]

rjc, has that broadway = juniper actually been confirmed somewhere?
I just don't see much sense in it; as it's the top end of laptop chips, yet it wouldn't push performance envelope up practicly at all compared to HD4800 Mobility series; power envelope in those desktop replacement machines can easily handle Cypress based Mobile chip, too.

 

[rjc]

rjc, has that broadway = juniper actually been confirmed somewhere?
I just don't see much sense in it; as it's the top end of laptop chips, yet it wouldn't push performance envelope up practicly at all compared to HD4800 Mobility series; power envelope in those desktop replacement machines can easily handle Cypress based Mobile chip, too.

There is a image half way down this page showing the three mobile gpus. Broadway is listed as a 30-60W part. Also the RV740 on the same process was about 30W for the 4830 and 44W for the 4860 mobile chips suggesting this new chip is just a bit above that in performance.

If for a second you assume Cypress = Broadway from above diagram which therefore implies Juniper = Madison and Madison is listed in the 15-30W range. This is a contradiction to the desktop performance where the performance and power usage of Juniper is above that of the RV740.

Edit:
Re: relative performance to 4xxx series.
Yes at max the Juniper chip is still below raw performance of the RV770, though i think the TDP of the 4870 board was ~150W and the 5770 was ~110W. Now looking at the mobile segment both the RV770 mobile and Broadway have to fit into the same 60W, that is all the OEMs will let them have(except maybe Alienware or someone else with more balls than brains). That tends to weight things more in the favor of the new chip, clocks should be relatively quite a bit higher on the mobile part compared to the desktop.

The above said most of notebook action should happen with the 2 chips below, they are more what OEMs seem to want and also their is competition from nvidia with their 3 GT21x chips. Redwood and Cedar both may be up to 50% more than RV730 and RV710 respectively at close to same TDP. I am guessing this extra performance is due to GDDR5 and more units, so might not show up fully as OEMs try to cut costs by substituting DDR3 if they can.

 

[sunscar]

Wondering if these will mannifest as MXM modules, and if any of them will be voltage/thermal comparable with a 4330, or if they'll even be compatible with the same interface (wanting to replace mine).

 

[Mindfury]

http://www.hardforum.com/showthread.php?t=1471938

System setup,
i5 750 @ stock
Gigabyte P55M-UD2
4GB OCZ 1333
5670 @ stock
4670 @ stock

Street Fighter 4: Max settings no AA 16xaf
Quote:
SCORE: 10473
AVERAGE: 93.35FPS
OS: Windows 7 Ultimate
CPU: Intel(R) Core(TM) i5 CPU 750 @ 2.67GHz
Memory: 4096MB
Graphics Card: ATI Radeon HD 5670
Display Setting: 1600x1200 60Hz

(c)CAPCOM U.S.A., INC. 2008, 2009 ALL RIGHTS RESERVED.
4670 Street Fighter 4: max settings no aa 16xaf
Quote:
SCORE: 8559
AVERAGE: 65.35FPS
OS: Windows 7 Ultimate
CPU: Intel(R) Core(TM) i5 CPU 750 @ 2.67GHz
Memory: 4096MB
Graphics Card: ATI Radeon HD 4600 Series
Display Setting: 1600x1200 60Hz

(c)CAPCOM U.S.A., INC. 2008, 2009 ALL RIGHTS RESERVED.

 

[Squilliam]

That looks excellent for the 56xx series! They are infact my favourite part because they represent usually the entry point a store bought computer has into gaming.

 

[neliz]

Redwood XT \o/

That's a LOT of work for GT240.

 

[Psycho]

Wonder how many texture units that is.. ofcourse the texture rate on the gpuz shot is wrong, but does it have 40? (got a lot of bandwidth, so could be).
I would have expected like 800mhz gddr5, and then maybe ddr3 for the 5650..