R800.

JPB

Diamond Member
Jul 4, 2005
4,064
89
91
R8x0 next gen ATI is again a dual chip

No surprise

Two graphics Vice Presidents at AMD, Matt Skynner, the Vice President of Graphics Marketing, and John Byrne the Vice President of Worldwide GPU Sales, have confirmed to Fudzilla that the next generation high-end graphics card will again be dual chip.


We're not talking about R700, as this card is almost at our door step, we are talking about the R8x0 card that should show up at some point next year.

ATI has decided that performance per watt is very important and it wants to continue to make very fast mainstream / performance chips and use two of them for the high-end card. This is the way the company wants to go, as this makes more financial sense.

At this time we don?t know many details and we can only suspect that R800 is the current codename, but at least we know the direction where ATI wants to go. Nvidia at the same time still believes big bulky chips are the answer to the highest performance part question.

Kind of early for anyone to be talking about this isn't it ?
 

AmberClad

Diamond Member
Jul 23, 2005
4,914
0
0
Yeah, it seems a bit premature to be announcing plans so far in advance. One year is a long time as far as GPUs, where it seems like plans change from day to day sometimes.

Still, performance-per-watt and fast mainstream GPUs is a good general goal to work towards.
 

tuteja1986

Diamond Member
Jun 1, 2005
3,676
0
0
Well , he hasn't said anything R800 architecture. All he said was that the highest end gpu will be dual chip. RV770 team wasn't the same team that worked on R600 but it was Xenos team. R600 team was given Axe :(.

http://www.rage3d.com/reviews/...itecture/index.php?p=2 Read this page to understand their new design philosophy.

Anyways , the Xeonos team has alot of new feature they are designing for R800.
 

BurnItDwn

Lifer
Oct 10, 1999
26,066
1,550
126
If we're lucky we might even see a die shrink too, though that may be a bit premature I'm thinking....
 

tuteja1986

Diamond Member
Jun 1, 2005
3,676
0
0
Originally posted by: BurnItDwn
If we're lucky we might even see a die shrink too, though that may be a bit premature I'm thinking....

45nm to 40nm is AMD target with R800.
 

Rhino2

Member
Jun 19, 2008
59
0
0
Originally posted by: KingstonU
Originally posted by: tuteja1986

45nm to 40nm is AMD target with R800.

Do you perhaps mean 55nm to 40nm? or 55nm to 45nm

I'm not sure but I think he meant 45-40nm as a target range, not from current size to future size.
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
I think that he meant that their goal is either 45nm or 40nm. imho they'll shoot for 40nm to keep that 1/2 step advantage over nvidia, but it will in large part depend on whether tsmc is ready for it.

imho nvidia is going to come out with all guns blazing in 2009. unlike amd's pathetic whimper when c2d came out, nvidia has a LOT of $$ and a strong focus on their core competency. Yes, yes, I know that they're branching out with cuda/tesla/chipsets/etc, but they still know that their primary source of income over the years has been gpus. In fact, I wouldn't be surprised to see them release all their 55nm gpus over the next 6 mos as g92(b), then see the 45nm shrink of gt200 about this time next year as the "midrange" that is competitive with high-end. If amd is truly using the same engineering team for r8xx then nvidia has a huge leg up on them for the next round, no matter how awesome r770 is.
 

Kuzi

Senior member
Sep 16, 2007
572
0
0
But ATI uses TSMCs process, and right now TSMC are testing 45nm, I guess TSMC may be able to produce chips on this new process this year. Maybe ATI can use the 45nm process for an RV770 refresh, in say 4-6 months and next year R8xx on the same 45nm process. Kind of like what they did with RV670/RV770 on the 55nm process. First the refresh of the old GPU on a new process, then the new GPU architecture after the process is more mature. Works very well this way.

I could be wrong, but I don't think TSMC can produce 40nm chips next year.
 

Tempered81

Diamond Member
Jan 29, 2007
6,374
1
81
http://www.pczilla.net/en/post/16.html

R870 is 40nm with 2000sp?
2008-5-21 13:44:35

We heard from chiphell forum that TSMC now tries 40nm process to produce ATI RV740 graphics chip in the labs. It?s said that RV740 would be the vanguard of 40 nm graphics chips.

We heard 40nm RV740 would be announced at the end of this year. Also at the first half of 2009, ATI would introduce R870, the last graphics chip based on R600 architecture. R870 would be 40nm chip with 2000 sp inside.


 

Munky

Diamond Member
Feb 5, 2005
9,372
0
76
I'm not sure why anyone would consider these rumors even remotely credible. Anyone would have guessed the dual chip prediction, and even if the rumor came directly from AMD, it could have been a misdirection leak to confuse people about the real nature of r800. At this point, you might as well expect the unexpected, and have better chance of being right than trying to predict based on current trends.
 

Sylvanas

Diamond Member
Jan 20, 2004
3,752
0
0
Originally posted by: munky
I'm not sure why anyone would consider these rumors even remotely credible. Anyone would have guessed the dual chip prediction, and even if the rumor came directly from AMD, it could have been a misdirection leak to confuse people about the real nature of r800. At this point, you might as well expect the unexpected, and have better chance of being right than trying to predict based on current trends.

QFT.

I would, however, imagine that AMD will be jumping on board TSMC's 40-45nm process as soon as possible- I can see a low end card based on this process coming perhaps early next year which would be the 'trial' test for 40-45nm before they build the high end chips on this.
 

JPB

Diamond Member
Jul 4, 2005
4,064
89
91
Originally posted by: munky
I'm not sure why anyone would consider these rumors even remotely credible. Anyone would have guessed the dual chip prediction, and even if the rumor came directly from AMD, it could have been a misdirection leak to confuse people about the real nature of r800. At this point, you might as well expect the unexpected, and have better chance of being right than trying to predict based on current trends.

Yep, not all is credible, IF any.

BUT....just imagine if it DID have 2000sp :Q
 

Kuzi

Senior member
Sep 16, 2007
572
0
0
Seems I was wrong and 40nm GPUs next year is very possible, even maybe at the end of this year if ATI decided to skip the 45nm process.
 

Janooo

Golden Member
Aug 22, 2005
1,067
13
81
Originally posted by: munky
I'm not sure why anyone would consider these rumors even remotely credible. Anyone would have guessed the dual chip prediction, and even if the rumor came directly from AMD, it could have been a misdirection leak to confuse people about the real nature of r800. At this point, you might as well expect the unexpected, and have better chance of being right than trying to predict based on current trends.

What's wrong with the story?
2000sp's? Having learned the lesson with 800sp's on such small die it wouldn't surprise me if they squeezed 2000 on a die @40nm.
Never say never. :)
 

Kuzi

Senior member
Sep 16, 2007
572
0
0
Originally posted by: Janooo
Originally posted by: munky
I'm not sure why anyone would consider these rumors even remotely credible. Anyone would have guessed the dual chip prediction, and even if the rumor came directly from AMD, it could have been a misdirection leak to confuse people about the real nature of r800. At this point, you might as well expect the unexpected, and have better chance of being right than trying to predict based on current trends.

What's wrong with the story?
2000sp's? Having learned the lesson with 800sp's on such small die it wouldn't surprise me if they squeezed 2000 on a die @40nm.
Never say never. :)

Never :)

I mean, more than doubling of the SPs again would go against what ATI is trying to do. At 40nm the core clocks can be pretty high maybe around +1000MHz. But that is of course if they don't increase the transistor count too much (over RV770). So for example, adding 50% more SPs (to 1200), 50% more TMUs (to 60), 50% more ROPs (to 24), would allow them to improve performance by a good margin, while keeping the die size small, and allowing higher clocks at 40nm. May also allow them to have a three GPU card, HD5870x3 anyone? :)

A smaller, less complex die with higher clocks is probably the path ATI will take for future GPUs. Though I don't think a 1600 SP R8xx would be out of the question.
 

Munky

Diamond Member
Feb 5, 2005
9,372
0
76
Originally posted by: Janooo
Originally posted by: munky
I'm not sure why anyone would consider these rumors even remotely credible. Anyone would have guessed the dual chip prediction, and even if the rumor came directly from AMD, it could have been a misdirection leak to confuse people about the real nature of r800. At this point, you might as well expect the unexpected, and have better chance of being right than trying to predict based on current trends.

What's wrong with the story?
2000sp's? Having learned the lesson with 800sp's on such small die it wouldn't surprise me if they squeezed 2000 on a die @40nm.
Never say never. :)

I'm not denying the fact that it could be 2000 or even more SP's. But this "story" sounds like someone is just throwing around numbers and reporting it as credible news. When the r800 is so far away from release, the rumor mill has a greater chance of being wrong than correct.
 

bryanW1995

Lifer
May 22, 2007
11,144
32
91
I know this guy who knows everyone. really. we were talking about kobe bryant the other day...my buddy had lunch with him 2 years ago. ron paul? check. hillary clinton? he's from arkansas. So I asked this buddy about r8xx. here's his prediction:

1841.3 sp's
1200mhz gddr5 (4800 effective)
256 bit
1100 mhz core clock
39.1 nm process


am I qualified to write for the inq or fudz yet? ;)
 

Kuzi

Senior member
Sep 16, 2007
572
0
0
Originally posted by: bryanW1995
I know this guy who knows everyone. really. we were talking about kobe bryant the other day...my buddy had lunch with him 2 years ago. ron paul? check. hillary clinton? he's from arkansas. So I asked this buddy about r8xx. here's his prediction:

1841.3 sp's
1200mhz gddr5 (4800 effective)
256 bit
1100 mhz core clock
39.1 nm process


am I qualified to write for the inq or fudz yet? ;)

You forgot to mention a 100k 3DMark Vintage score you got from an R8xx engineering sample you had personally tested. And of course that your dog ate the Laptop containing the screenshots. (it was an easy to swallow MacBook Air) :p