you think ATI R400 will double the transistor count again?

[megadrive0088]

Is it likely or even possible that ATI's R400 (design team East right?) will double the transistor count/ complexity over R300? meaning another doubling of transistors like R300 (almost) did over R200, and R200 did over R100.

Could we see a 200M+ transistor VPU in 2003, with perhaps 16 pipelines?

Then there is the R500, which is being designed by the same team that made R300 (design team West) I would not imagine R500 to be out until mid-to-late 2004. And if R400 is a 200M+ transistor chip, I would NOT expect R500 to double that, but perhaps be much more efficient rather than going for all out size and brute force. then again I might be wrong about that, but I cannot imagine a 400M+ transistor chip. but who knows, with the very rapid pace of development and continued fierce rivalry between ATI and Nvidia, we could see an ever escalating chip war by two evenly matched opponents.

 

[megadrive0088]

of course, there is the more immediate future, with NV30 coming, ATI will surely produce something to counter that. There's little doubt that will be in the form of R350, with DDRII as standard (an option in R300) as well as 16 TMUs in a 8:2 configuration. I like that idea alot since ATI would have 256-bit bus AND DDRII, while NV30 only has 128-bit bus with DDRII while being equal in the number of pipes and TMUs.

 

[RussSchultz]

I don't think that 200M transistors will be economically feasible until .09 is mainstream.

 

[megadrive0088]

Sony is going for 500M transistors on .10 micron process with production to begin in 2004 at the latest (in time for PS3 in 2005) that is the main CPU chip I believe. The PS3 graphics chip could very well be even larger than 500M transistors, as PS2's GS was about 4x larger than the EE (43M vs 10.5M)

 

[Tagrineth]

Sony is going for 500M transistors on .10 micron process with production to begin in 2004 at the latest (in time for PS3 in 2005) that is the main CPU chip I believe. The PS3 graphics chip could very well be even larger than 500M transistors, as PS2's GS was about 4x larger than the EE (43M vs 10.5M)

Most of the GS's transistor count is its cache.

 

[MfA]

Sony is going for 500M transistors on .10 micron process with production to begin in 2004 at the latest (in time for PS3 in 2005) that is the main CPU chip I believe.

Reference?

 

[Kristof]

Lets concentrate on what is currently available and feasable rather than dream about some magic number pulled out of thin air. Reality is that companies are having trouble to get .13 going, so lets not go overboard quite yet. IMHO of course.

 

[DadUM]

Let's see, Intel's Northwood P4 is 55 million transitors at .13um and 127mm^2. So since we are at the exact same process size, I will just assume a linear relationship, which would put a 200 million transitor chip at around 461 mm^2. Wow. What is r300's die area? Athlon XP's is around 80 mm^2, Barton is around 115 mm^2, couldn't find Parhelia quickly. That's a very large complex chip.

 

[Mephisto]

Let's see, Intel's Northwood P4 is 55 million transitors at .13um and 127mm^2. So since we are at the exact same process size, I will just assume a linear relationship, which would put a 200 million transitor chip at around 461 mm^2. Wow. What is r300's die area? Athlon XP's is around 80 mm^2, Barton is around 115 mm^2, couldn't find Parhelia quickly. That's a very large complex chip.

You can't compare GPU and CPU complexities. CPUs include huge amounts of caches, around half of the die is covered with simple logic. GPU caches cover only ~1/5 of the die.

 

[Entropy]

Reality is that companies are having trouble to get .13 going, so lets not go overboard quite yet. IMHO of course.

Sources that are typically far more trustworthy than me on these matters argue that the transition from .13 to .09 should be less nasty than the transition from .18 to .13
So while the progression of GPUs from .15 to .13 can't really allow any dramatic increases in gate counts, the transition to .09 is more likely to allow architectural innovation. And we have reasonable hope to get there relatively quickly.
Wider parallellism would be the quick and dirty way of utilizing a larger transistor budget. It could be interesting to hear what alternatives people could come up with.

Entropy

 

[Prometheus]

Who believed Ati could do 110million trans on 0.15?

 

[Vince]

Sony is going for 500M transistors on .10 micron process with production to begin in 2004 at the latest (in time for PS3 in 2005) that is the main CPU chip I believe.

Reference?

He doesn't have one Because, it's not true.

Speaking of the 0.13um -> 0.09um shift and beyond, I recall reading on perhaps EETimes that as you shrink the lithography, the inherient errors multiply geometrically - something which will require a pretty large change in how chip designing is thought out in the early stages and completed.

 

[Nappe1]

Who believed Ati could do 110million trans on 0.15?

who would believed year ago what Matrox had on it's sleeves?
ATI doubled their transistor count, matrox more than tripled.

 

[megadrive0088]

the 500 million transistor count for Emotion Engine 3 was a statement made by SCEA's Ken Kugari in 1999, when Sony first said that PS3 would have 1000x the power of PS2. look it up on Google, its around. I've got a headache right now, I can't be bothered. maybe later. but it's there, and I didn't make it up. if it happens or not, is upto Sony-IBM-Toshiba, but KK did say 500M transistors. weather he was blowing ____ out his arse or not, you'll have to ask him. besides others around here and elsewhere know about this supposed 500M transistors for EE3.

 

[megadrive0088]

http://www.geek.com/news/geeknews/2001jan/gam20010123003936.htm

there's a link with a reference to 500M transistors for EE3.

Also a correction, the EE in PS2 has 13M not 10.5M as I mentioned before.

 

[MfA]

Hearsay's reliability does not really increase with the frequency it occurs, an attributable quote from some more credible source than your memory would be nice.

If you have never with 100% certainty remembered something which turned out to not have happened as you remembered it you make me feel old ...

 

[OpenGL guy]

Who believed Ati could do 110million trans on 0.15?

who would believed year ago what Matrox had on it's sleeves?
ATI doubled their transistor count, matrox more than tripled.

Which is more impressive: Going from 1 to 3 or from 100 to 200? Not that this represents ATi and Matrox, but I think your comment says nothing by itself.

 

[V3]

Don't people already planning what 1 billion transistor CPU going to be like. I read over Toms Hardware, that Intel is going multi core as well for their 1 billion transistor CPU.

Do you think 1 billion transistor GPU will go multi core ? with huge cache ?
My hunch, GPU will hit 1 billion transistor count first before CPU.

 

[Nappe1]

Who believed Ati could do 110million trans on 0.15?

who would believed year ago what Matrox had on it's sleeves?
ATI doubled their transistor count, matrox more than tripled.

Which is more impressive: Going from 1 to 3 or from 100 to 200? Not that this represents ATi and Matrox, but I think your comment says nothing by itself.

OpenGL guy... don't take everything so personally man... relax... I was trying to show that everything is depending scale.

and besides, I don't know the transistor count G550, so that tripled was just a plain guess really. But I know that Parhelia has something like about 79 million Transistors and I am pretty sure that G550 had less than 30 million...




EDIT: and of course, we have to remember that in 1999 (or was it 2000? I need to check this one.) Rendition / Micron already demoed with V4400 chip having more than 128 Million Transistors. (most of it was for 9MB RAM on chip. manufacturing process was 0.15 µm for memory, 0.18µm for logic.) They decided not to bring it to market because of unannounced reasons. (propably heating problems.)

so IMO, limits are made to be broken in this market. Afaik, at least one company is already taking a risk and skiping 0.13µm process and jumping straight to the 0.09µm. This of course menas that we won't see this chip very soon. Only time will show if they success on it.

 

[KimB]

If the R400 doubles the transistor count again, it's not coming our next summer/fall.