The R7xx Architecture Rumours & Speculation Thread

Yeah well, high-end this gen was the G80 GTX/Ultra, GTS was high-midrange and that's where ATI landed with R600.

I don't know how you can call $399 "mid-range" in any current or historical sense. Maybe you can stretch "mid-range" to $299 (which is indeed a stretch, btw), but not a dollar past. $149-199 is the historical meat of the mid-range, with $249 the usual outer reach where your typical mid-range buyer can feel a little naughty/proud for having splurged.
 
Okay, I moved the R600 vs G80 discussion to a new thread since this was running off-topic. I hope you enjoy the thread title, too! ;) (no offense intended to those who posted some of that originally, I just thought it sounded awesome).
http://forum.beyond3d.com/showthread.php?t=43523

There was some content that ideally have should be copied to both threads, but I'll just quote it here instead:
Jawed said:
Let's guess at R700: say it is 2xR600 configuration on 55nm, each die with 256-bit memory bus (70GB/s?), with an additional 140GB/s connection between them and performs 120%+ faster than R600 on "CF compatible" games (clocks should increase from where they are now). Which part of the architecture and technology of R600 are you expecting to be redundant? I can't think of anything.

e.g. I'm guessing the two dies' L2 caches will share data and that CF will suffer none of the "traditional" 2x "distinct pools of memory" problems that SLI and original CF suffered from.

It doesn't mean I like the idea of a 2 die R700, but I'm trying to correlate aspects of R600 with that direction as well as think of function points in D3D11 that steer the architecture. R700, conceptually, hasn't just popped out of nowhere. After they got it running, ATI didn't go "oh shit, R600 is a dead end, what are we gonna do? Oh, we could put two of them together on one board."

I dislike the idea, because I think game compatibility will go right out the window. I'm also pessimistic about the compatibility of AFR with the more intricate rendering algorithms. So CF-incompatible games will be just as wasteful (if not more so) than they are on a single R600.
And then, in reply to that...
Blazkowicz said:
but maybe, as an extension of your L2 sharing idea, we can think of the R700 as a single processor on multiple dies? like a 386+387, a voodoo 1 or a pentium pro (with one or two L2 dies). one of the dies would be the master and speak to the PCIe bus and you'd effectly have a single GPU software wise. The on-package interconnects would have to be really fast (how is that made on the pentium pro? or the L3 dies on POWER chips?)

Is that feasible and could they still easily be able to use a single die for midrange boards.
 
I don't know how you can call $399 "mid-range" in any current or historical sense. Maybe you can stretch "mid-range" to $299 (which is indeed a stretch, btw), but not a dollar past. $149-199 is the historical meat of the mid-range, with $249 the usual outer reach where your typical mid-range buyer can feel a little naughty/proud for having splurged.

Semantics aside, you do get my point - they didn't make it to put out a top-performer, the chip leading the pack in all benchies etc. or at least equal to the competition's top offerings.
 
Yeah well, high-end this gen was the G80 GTX/Ultra, GTS was high-midrange and that's where ATI landed with R600. I don't think they'll do better suddenly after trailing nV since 3 generations in both timing and features.

Excuse me, but how can you claim ATI have been trailing nV in features for the last 3 generations ?!.

Timing, sure, features, no way.
 
Well, in R4xx vs NV4x, they did trail in features, because NV had support for VS/PS3.0, VTF, FP render targets, etc. Only in the R5xx generation did they get a lead in features. And the NV3x, although sucky at FP performance, had a more full featured PS pipeline (unlimited dependent lookups, 4096inst length, full swizzle, even more with the pack/unpack insts, etc)

Which just goes to show you that being early to introduce advanced DX features is not neccessarily a guarantee of better performance later. Nvidia essentially had to discard the NV3x-G7x architecture line in order to make a leapfrog, as legacy tweaks to previously frontloaded features in fact weighted them down.

That's why its no guarantee that DX11/12 features implemented now in the R600 will actually be the optimal way to implement DX11/12 features when the market is actually ready for them. You might find ATI jettisoning the work they did in the R600 by the time the R800 rolls along.
 
Yup, what DC said ^^

But the 3 last generations would be R420, R520, R600 (argueably R520, R580, R600), vs the Nvidia counterparts... which featurewise adds up to 1 with more features for nV, or possibly 0 if you go with the extreme version of 520, 580, 600 being the 3 last gens.

From the R350 to the R520 generation however i agree that nV had the upper hand featurewise.
 
But the 3 last generations would be R420, R520, R600 (argueably R520, R580, R600), vs the Nvidia counterparts... which featurewise adds up to 1 with more features for nV, or possibly 0 if you go with the extreme version of 520, 580, 600 being the 3 last gens.

From the R350 to the R520 generation however i agree that nV had the upper hand featurewise.

Is R600 vs G80 anything but a tie WRT features supported?Ignoring the tesselator, which is likely to simply be an inexpensive thing that adds perceived value, but probably won't ever get used on an R600 board outside of ATi demos.
 
But the 3 last generations would be R420, R520, R600 (argueably R520, R580, R600), vs the Nvidia counterparts... which featurewise adds up to 1 with more features for nV, or possibly 0 if you go with the extreme version of 520, 580, 600 being the 3 last gens.

From the R350 to the R520 generation however i agree that nV had the upper hand featurewise.

Well they also had a DX10 part way before ATI. I'm not dividing between features and timing here, but looking at the combination of both. Same or more features 9 months later are worth nothing for the sales at the specific point in time (as in "now" when G80 came out, for example).
 
Semantics aside, you do get my point - they didn't make it to put out a top-performer, the chip leading the pack in all benchies etc. or at least equal to the competition's top offerings.

I think arguing backwards from results to intent on a next-gen architecture as to competitive positioning --and even more so in this case than next-gen architecture, but also a major market inflection point (Vista/DX10)-- is fraught with peril. We know the roots of R600 go back into 2002, ffs. To suggest ATI/AMD purposely "didn't make it to put out a top-performer" is, I think, baloney.
 
Why purposely? They made wrong decisions several times in a raw, that is hardly to blame on coincidence. They didn't make it and now they might realize that the race is lost. That is, unless all goes well from the get-go with the R700 and it is a real performance and efficiency monster. Frankly, I doubt their ability (both executional and technical) to pull it off. Call me pessimistic, I'm just observing things and think I'm (ruthlessly) realistic.
 
I'm sorry, but I still can't get the thrust of your argument. Are you arguing they're incompetent, and thus won't compete at the top going forward? Or are you arguing they don't want to play in that market anymore and thus won't compete at the top going forward?

Tho frankly you also seem to want to make "compete" = "win", which is a fallacy of its own, in my book. But first things first!
 
Winning is the reason we have competition to begin with - every company wants to win, they're all in it for the money.

And frankly, yes, I think they're incompetent and that they now slowly realize that. Not necessarily technically, but in terms of planning/execution definitely. I already said several times that I'd axe the ex-ATI management completely if I was AMD (although they have their own share of problems as well).
 
Winning is the reason we have competition to begin with - every company wants to win, they're all in it for the money.

And frankly, yes, I think they're incompetent and that they now slowly realize that. Not necessarily technically, but in terms of planning/execution definitely. I already said several times that I'd axe the ex-ATI management completely if I was AMD (although they have their own share of problems as well).

Incompetent is a harsh word, but ATi always was somewhat like a company run by engineers, which is not necessarily what you want. They're always ending up with wacky, overengineered solutions, that are scarcely used in the real world because they don't do evil things like TWIMTBP and aggresive developer interaction, they rather let it float and what'll be will be. The focus on math and computational power and on fixing through them some problems that may or may not materialize in the future is also illustrative of this.

I remember that hopes were high with Orton to bring a more business/real-world focused attitude there, after the huge R300 success, but in hindsight it simply seems that was merely a happy accident. Coincidentally, it was also one of the parts that lacked overengineering/WTF contorted features, and simply cared about staying close to spec and running stuff fast.
 
They're always ending up with wacky, overengineered solutions, that are scarcely used in the real world....... The focus on math and computational power and on fixing through them some problems that may or may not materialize in the future is also illustrative of this.

Exactly the reason why I think they're incompetent. Because they repeated the same failure after R520.
 
Hi;

after reading this thread I'm wondering why everyone assumes that R700 is an two-chip architecture.

Why no four-chip architecture?

IMHO it would be far more logical to have an 4 chip highend implementation; a 2 chip midrange implementation and a single chip low-end/mainstream implemenation of the same basic architecture.

And with a chip with maybe (using the RV630 as basis) 8TMU's; 120-160 Shaders (24-32x5) and hopefully also 8 ROPs with MSAA functionality and a 128bit GDDR4 memory interface such an system could produce quite nice highend, midrange and mainstream products.


Manfred
 
Hi;

after reading this thread I'm wondering why everyone assumes that R700 is an two-chip architecture.

Why no four-chip architecture?

IMHO it would be far more logical to have an 4 chip highend implementation; a 2 chip midrange implementation and a single chip low-end/mainstream implemenation of the same basic architecture.

And with a chip with maybe (using the RV630 as basis) 8TMU's; 120-160 Shaders (24-32x5) and hopefully also 8 ROPs with MSAA functionality and a 128bit GDDR4 memory interface such an system could produce quite nice highend, midrange and mainstream products.


Manfred

There may be one little aspect:how do you distribute the load between the chips?Ignoring possible PCB size limitations, how would you distribute the rendering load between the 4 chips?AFR would mean significant latency, and the split rendering schemes that we know aren't all that effective. And unless it's a 100% hitter(meaning no scenarios in which multi-chip fails), which may or may not be possible, you'd want your base chip to have a little oomph by itself...if you pile on low end GPUs, if you break down to a single one because you don't have a profile/the game does something wacky/whatever, you're in for some serious pain, IMHO.
 
Hi;

after reading this thread I'm wondering why everyone assumes that R700 is an two-chip architecture.

Why no four-chip architecture?

IMHO it would be far more logical to have an 4 chip highend implementation; a 2 chip midrange implementation and a single chip low-end/mainstream implemenation of the same basic architecture.

And with a chip with maybe (using the RV630 as basis) 8TMU's; 120-160 Shaders (24-32x5) and hopefully also 8 ROPs with MSAA functionality and a 128bit GDDR4 memory interface such an system could produce quite nice highend, midrange and mainstream products.
I agree, it looks more "scalable". But I wouldn't be surprised if R700 is a gentle introduction that uses only two chips.

Also we need to take into account the buses to the outside world on each chip. e.g. I'm assuming that the bandwidth to the chip's attached memory, say 70GB/s on a 256-bit bus, or 35GB/s on a 128-bit bus needs to doubled for a chip-to-chip bus.

The pad counts and density required to achieve this (remembering that we're talking 55nm chips) are up in the air (power consumes a lot of pads too). Hey, I was one of the "R600 doesn't have 512-bit memory bus because they can't get all the pads on the chip!" crowd, for a while anyway.

This old thread is good:

http://forum.beyond3d.com/showthread.php?t=35804

It's amusing to see Silent Guy at the end there arguing against R600's soon to appear ring bus.

Alternatively, perhaps RV670/R670 is the "gentle introduction" and R700 is four core?

Jawed
 
Alternatively, perhaps RV670/R670 is the "gentle introduction" and R700 is four core?

Jawed

I don't know about that one. You never know with ATi though.

My thoughts on R670 is a dual package gemini like board. I think the similarities between RV670 and R700 will be what is inside the chip rather than what on the outside.:p


Did not the Inquirer say a about a half a year ago that R700 was going to be scalable from top to bottom by just ajusting the amount of chips?:oops:
I know it's the Inq, but maybe their is some merit in the 4 chip idea.

Each chip consisting of say 24 super scalar ALU's, 2 TU's, and 1 RBE?
 
Back
Top