The LAST R600 Rumours & Speculation Thread
- Thread starter Geo
- Start date
- Status
Jawed
Legend
278M is 91.7% of 302M. 90nm is 67% of 110nm. The two together makes 61.4%, or a reduction of 38.6%.
In theory the 110nm node at TSMC is "over-sized", because it's a "half-node". I don't know by how much. The exact rescaling from 110 to 90nm would depend on the proportion of memory and logic (in the most basic terms) because they have different densities on die. And the pipeline change (25M transistors cut) might have affected the relative proportions of memory and logic. Hey, I'm guessing :smile:
So a miracle? it doesn't seem like it really.
Jawed
Obviously not a miracle, can't a guy have a slight poetic license? 
I figured it was something to that extent, some amount of logic/memory cache using less space therefore requiring less power therefore less transistors. Thanks for your guesstiplaination though, it does help put things in context for me.
I never figured it like that with the percentages (I never knew how much a die could be shrunk going 110->90nm...Straight math says ~82%, and I knew that wasn't right from knowing 90-80nm was roughly 15-20% not ~11% ), but that way it makes more sense...although it still is a 42% shrink, so there's a little more to it than that, although your explanation on differences in nodes and logic/memory shrink-scaling could take care of that. Once again, thanks for your synopsis.
Regardless, who's to say a similar thing couldn't occur, if not more-so, in a G80 shrink especially to 65nm? Obviously the likelihood of a 65nm G80 being smaller than a straight R600 shrink is good, considering 90->80 could bring a 15-20% shrink (say ~400mm), and a 65nm shrink perhaps 35-40% (~250mm) from there? Granted I don't know how it would shrink (basing guess on G73/rv535), nor R600 either, but nonetheless, ATi only has one step it can go through. I suppose it's very possible they could end up roughly the same size, come to think of it, on the same process if there are no changes to the refresh. That would make the original thought I opposed correct, but the question remains will ATi actually get the products out to compete with them before nvidia is ready to turn around and launch their next product(s)? I certainly hope so.
I figured it was something to that extent, some amount of logic/memory cache using less space therefore requiring less power therefore less transistors. Thanks for your guesstiplaination though, it does help put things in context for me.
I never figured it like that with the percentages (I never knew how much a die could be shrunk going 110->90nm...Straight math says ~82%, and I knew that wasn't right from knowing 90-80nm was roughly 15-20% not ~11% ), but that way it makes more sense...although it still is a 42% shrink, so there's a little more to it than that, although your explanation on differences in nodes and logic/memory shrink-scaling could take care of that. Once again, thanks for your synopsis.
Regardless, who's to say a similar thing couldn't occur, if not more-so, in a G80 shrink especially to 65nm? Obviously the likelihood of a 65nm G80 being smaller than a straight R600 shrink is good, considering 90->80 could bring a 15-20% shrink (say ~400mm), and a 65nm shrink perhaps 35-40% (~250mm) from there? Granted I don't know how it would shrink (basing guess on G73/rv535), nor R600 either, but nonetheless, ATi only has one step it can go through. I suppose it's very possible they could end up roughly the same size, come to think of it, on the same process if there are no changes to the refresh. That would make the original thought I opposed correct, but the question remains will ATi actually get the products out to compete with them before nvidia is ready to turn around and launch their next product(s)? I certainly hope so.
Last edited by a moderator:
G80 = 681 million transistors “090nm tech”
R600 = 720 million transistors “080nm tech”
R580 = 384 million transistors “090nm tech”
G71 = 278 million transistors “090nm tech”
R520 = 321 million transistors “090nm tech”
G70 = 302 million transistors “110nm tech”
NV40/NV45 = 222 million transistors “130nm tech” G6800Ultra
R480 = 160 million transistors “110nm tech” X850XT-PE
R420 = 160 million transistors “130nm tech” X800XT-PE
NV30 = 130 million transistors “130nm tech” G5800 Ultra
R300 = 107 million transistors “150nm tech” R-9700Pro
Between G80 and R600 the different is 39 million transistors.
Between G71 and R580 the different is 106 million transistors.
Between G70 and R520 the different is 19 million transistors.
Between NV45 and R480/R420 the different is 62 million transistors.
Between NV30 and R300 the different is 23 million transistors.
Based on the history for the GPU’s transistor count; for me it does not mean much, because FPS does not increase if one GPU has more transistor count then another.
My guess R600 is going to be better vs. G80 in term of quality area’s. But performance is under the question? Mark. “Either is going to be another R520 or future proof R580” I dough ATI/AMD could pull another R300 as a surprise for Nvidia. “Those days are over”
R600 = 720 million transistors “080nm tech”
R580 = 384 million transistors “090nm tech”
G71 = 278 million transistors “090nm tech”
R520 = 321 million transistors “090nm tech”
G70 = 302 million transistors “110nm tech”
NV40/NV45 = 222 million transistors “130nm tech” G6800Ultra
R480 = 160 million transistors “110nm tech” X850XT-PE
R420 = 160 million transistors “130nm tech” X800XT-PE
NV30 = 130 million transistors “130nm tech” G5800 Ultra
R300 = 107 million transistors “150nm tech” R-9700Pro
Between G80 and R600 the different is 39 million transistors.
Between G71 and R580 the different is 106 million transistors.
Between G70 and R520 the different is 19 million transistors.
Between NV45 and R480/R420 the different is 62 million transistors.
Between NV30 and R300 the different is 23 million transistors.
Based on the history for the GPU’s transistor count; for me it does not mean much, because FPS does not increase if one GPU has more transistor count then another.
My guess R600 is going to be better vs. G80 in term of quality area’s. But performance is under the question? Mark. “Either is going to be another R520 or future proof R580” I dough ATI/AMD could pull another R300 as a surprise for Nvidia. “Those days are over”
You are making your life difficult by making simple things too complex.
R600 is 720mil and ~434mm². A G80 shrink on 80nm would be less than 10% smaller. So yeah thats exactly Anon Lamer's point, its no where near an 80% die advantage that Nvidia had in this gen in the high end segment.
EDIT
Dave Orton from AMD Analyst Day about a month back
Dave Orton said:
Last edited by a moderator:
You are making your life difficult by making simple things too complex.
R600 is 720mil and ~434mm². A G80 shrink on 80nm would be less than 10% smaller. So yeah thats exactly Anon Lamer's point, its no where near an 80% die advantage that Nvidia had in this gen in the high end segment.
Isn't it more likely that ATI will finally be able to increase their margins while Nvidia keeps them steady? (At least initially...)
Historically, equally performing cards come a roughly equal prices. When there is a large imbalance, the producer of the smallest die can set the price and the other one has to follow. When there's no imbalance, it's possble that both will want to go for more money instead getting into a pricewar.
Or maybe it's just trivially obvious that I'm not an economist.
You are making your life difficult by making simple things too complex.
Yeah, you're right.
Still, rv530->rv535 and g73-A2->G73-B1 point more in the 15-20% catagory. I know, still semantical compared to last gen. Still, who's to say Nvidia won't surprise us and go to 65nm for G81, skipping 80nm, and have it out the door months before ATi's 65nm high-end refresh. If so, it might not be quite the same situation all over again, as ATi will undoubtabley get there, but it could happen...and those months could be crucial. I know I know...Lots of 'ifs' in there. As for Dave's quote, I'm almost certain he's referring to Rv610/Rv630, as they are expected shortly after R600 on 65nm, although who knows...rv660/rv670 and even possibley R6-whatever-0 could come earlier in H2 rather than later.
silent guy said:
I'd agree with that for the most part. When a product sucks ass (see x1600) there are always times when a company with a larger die has to lower the price and the other company follows, although that does follow the 'equal performing cards come at equal prices', although that didn't happen initially.
Just curious lurking level505 admin (I know you're out there):
Where are those promised pics of R600 for today? How about those benches that were do a few days ago? Did you spray your case with compressed air and/or find a HSF yet? I know...they're pretty hard to come by.
Last edited by a moderator:
Yes, 80nm -> 90nm ~ 19% size decrease.Yeah, you're right.
Still, rv530->rv535 and g73-A2->G73-B1 point more in the 15-20% catagory. I know, still semantical compared to last gen. Still, who's to say Nvidia won't surprise us and go to 65nm for G81, skipping 80nm, and have it out the door months before ATi's 65nm high-end refresh. If so, it might not be quite the same situation all over again, as ATi will undoubtabley get there, but it could happen...and those months could be crucial. I know I know...Lots of 'ifs' in there. As for Dave's quote, I'm almost certain he's referring to Rv610/Rv630, as they are expected shortly after R600 on 65nm, although who knows...rv660/rv670 and even possibley R6-whatever-0 could come earlier in H2 rather than later.
Using that as a factor, comparing R600 and G80 gives us a die differential of ~10%. And thats hardly significant to call it an advantage.
I wouldnt even start debating on who gets their flagship gpu on 65nm, that is a still some ways off.
True. Touche'.
Pardon the probably dumb question, but is that with NVIO encounted or not?
zgemboandislic
Newcomer
So...we're gonna have to wait for DX10 through Vista Update?
Because? Even if you don't think they are going to hard launch, you really think they'd launch without knowing when they could ship? Look at the R520 launch. Not a hard launch (for XT anyway), but they still committed to when they would ship (and I think missed it by a little bit anyway, but still. Edit: Well, some dependent milestone was missed on the way to availability date by a touch; it may not have been "ship date". The point tho is they still had internal milestone dates lined up, including "ship date", to reach an available timeframe for XT which was provided at the time of launch).
Huh? DX10 has been in Vista since forever (can't even remember the build it was included in, but it was loong ago), and is included in final build (artificially lifted build 59xx to 6000, available @ MSDN since.. uhm.. sometime in last November I think), too, as final version.
Well obviously there's no surprise for Nvidia, as the biggest surprice with the R300/NV30 launch was that NV30 was so utter shite and we know that G80 isn't. However, that doesn't tell anything about R600.
- Status