Please, try to be more discreet about your findings. You might embarrass someone over there.
Heh, well done Pete
AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
Heh, well done Pete
From the "just for the hell of it" dept, based on these rumors of 4870 being 25% faster than 9800GTX,
I used this review (Tom's hardware 9800GTX review) to attempt to calculate how much faster 4870 is than 3870 using the definition 4870=1.25 9800GTX. I just used values for the 3870/9800GTX from the lowest rez tested (usually 1680X1050, and not CPU limited for these demanding games) with AA+AF. The only exception was Crysis, where the AA/AF score was 18 FPS on both 3870/9800GTX. Clearly this was an anomalous result, so I used the no AA/AF result for both cards in that one case.
Edit: Ah crap, it didn't format right. Anyways, the bottom line result was a hypothetical 1.25X 9800GTX 4870 was on average 82.3% faster than a 3870. The lowest improvement was 66.1% (Crysis) and the highest improvement was 123.9% (UT3).
So, 82.3%. I suppose that's in line with what I'd expect out of a 32 TMU part, a near 100% increase, minus some overhead, perhaps occasional shader bottlenecking (since the TMU:Shader ratio is lower in the new part, shader bottlenecks become more likely) etc.
That's actually better than the initial rumors we got of 4870 being 40-50% faster than 3870. So maybe it isn't true. But the 1.25 9800GTX thing came from CJ, who seems to be a trusted source here from what I gather.
So if correct, you're talking about a near doubling of power for AMD high end, which is more impressive than "25% better than 9800GTX" sounds at first glance, and I'd wager a heck of a lot better than they pulled off last time with the initial X1950XTX>HD2900XT transition. Perhaps that will be my next statistical "analysis".
I used this review (Tom's hardware 9800GTX review) to attempt to calculate how much faster 4870 is than 3870 using the definition 4870=1.25 9800GTX. I just used values for the 3870/9800GTX from the lowest rez tested (usually 1680X1050, and not CPU limited for these demanding games) with AA+AF. The only exception was Crysis, where the AA/AF score was 18 FPS on both 3870/9800GTX. Clearly this was an anomalous result, so I used the no AA/AF result for both cards in that one case.
Code:
3870 FPS, 9800GTX FPS, 4870=1.25 9800GTX FPS, % 4870>3870
Flight Simulator X: 21.5, 31.1, 38.9, +80.8%
Call of Duty 4 32.5, 43.7, 54.6, +68.0%
Test Drive Unlimited 48.4, 65.0, 81.2, +67.7%
Crysis 24.0, 31.9, 39.9, +66.1%
World In Conflict 14.0, 22.4, 28, +100%
Supreme Commander 38.4, 53.7, 67.1, +74.8%
Quake Wars 53.0, 75.1, 93.9, +77.1%
UT3 33.6, 60.2, 75.2, +123.9%
Avg. across all games +82.3%Edit: Ah crap, it didn't format right. Anyways, the bottom line result was a hypothetical 1.25X 9800GTX 4870 was on average 82.3% faster than a 3870. The lowest improvement was 66.1% (Crysis) and the highest improvement was 123.9% (UT3).
So, 82.3%. I suppose that's in line with what I'd expect out of a 32 TMU part, a near 100% increase, minus some overhead, perhaps occasional shader bottlenecking (since the TMU:Shader ratio is lower in the new part, shader bottlenecks become more likely) etc.
That's actually better than the initial rumors we got of 4870 being 40-50% faster than 3870. So maybe it isn't true. But the 1.25 9800GTX thing came from CJ, who seems to be a trusted source here from what I gather.
So if correct, you're talking about a near doubling of power for AMD high end, which is more impressive than "25% better than 9800GTX" sounds at first glance, and I'd wager a heck of a lot better than they pulled off last time with the initial X1950XTX>HD2900XT transition. Perhaps that will be my next statistical "analysis".
Last edited by a moderator:
Fuad admits he was mistaken about 4870's bus width. He also still doesn't have much to say, which is surprising as he hasn't provided much of interest lately (except maybe the fan sizes for 4870 and 4850, if that can be considered interesting). Maybe some of his sources were let go.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=7455&Itemid=34
But honestly, it was all very weird to have Pro @ 256 bit and XT @ 512 bit. Some of those news sites have caveats when disclosing less plausible information, like "our sources told us, but take it with a grain of salt", I wish he was a little more circumspect instead of boasting.
http://www.fudzilla.com/index.php?option=com_content&task=view&id=7455&Itemid=34
But honestly, it was all very weird to have Pro @ 256 bit and XT @ 512 bit. Some of those news sites have caveats when disclosing less plausible information, like "our sources told us, but take it with a grain of salt", I wish he was a little more circumspect instead of boasting.
Well, revisiting some old HD 2900XT launch reviews, it's performance was all over the map. In a few cases it did seem to provide that 66%-100% jump over X1950XTX, but most times seemed nearer to 50% when it did offer a consistent improvement, but more befuddling was many cases when depending on AA/AF settings and resolution and the individual review it offered almost no improvement at all over it's predecessor. I'd wildly guesstimate if you were to average it out, and again the results seem all over the map depending on review and settings, it would come out a lot closer to 40% improvement if that though.
The HD2900XT at launch often suffered a MASSIVE hit with AA/AF enabled (being say, 50% faster than X1950XTX with no AA/AF, and only 5% faster with AA/AF). It seems they must have gotten that a bit straightened out over time, because I dont see the massive disparity in HD3870 AA/AF versus none scores.
A bit strange to look back at a not so distant time when games like "Prey" were a benchmark staple :smile:
Of course it's easy to remember what went down, 7900GTX and X1950XTX were fairly equal at that time (slight edge for R580). G80 came out and it was that 80-100% increase for Nvidia. R600 came out and it was more like 30-50% over it's predecessor for AMD. Thus it had to be slotted down in price to compete with 8800GTS, which could be thought of as very roughly halfway between 7900GTX and 8800GTX.
The HD2900XT at launch often suffered a MASSIVE hit with AA/AF enabled (being say, 50% faster than X1950XTX with no AA/AF, and only 5% faster with AA/AF). It seems they must have gotten that a bit straightened out over time, because I dont see the massive disparity in HD3870 AA/AF versus none scores.
A bit strange to look back at a not so distant time when games like "Prey" were a benchmark staple :smile:
Of course it's easy to remember what went down, 7900GTX and X1950XTX were fairly equal at that time (slight edge for R580). G80 came out and it was that 80-100% increase for Nvidia. R600 came out and it was more like 30-50% over it's predecessor for AMD. Thus it had to be slotted down in price to compete with 8800GTS, which could be thought of as very roughly halfway between 7900GTX and 8800GTX.
Last edited by a moderator:
Here is another post I found that claims RV770XT=~80% over RV670XT
http://forum.beyond3d.com/showpost.php?p=1164052&postcount=1846
http://forum.beyond3d.com/showpost.php?p=1164052&postcount=1846
This is RV770 Pro version .
leoneazzurro
Regular
Well, the first results in the launch reviews were certainly disappointing (but I remember that first results for the G80 8800 GTS too were also on the line of X1950 XTX). But IMHO it was also a question of immature drivers vs mature ones. I made a little check on digit-life in the last September's video digest , and in Call of Juarez and Company of Heroes, with AA, HD2900Xt was above X1950 XTX by respectively a 25% and 40% average. If we look at the 3870 with latest drivers, the percentage in the same tests increases to 40% for CoJ and 44% for CoH.
Of course, there could be lower deviations in other games (but also higher ones in other cases), but I think that this kind of improvement is what ATI engineers were expecting from R600. Of course, now we know that they were aiming too low...
PS: and of course optimizing for R600 architecture could have given the driver team a lot of headaches :smile:
Yet another reason why no one should believe any technical rumors Fuad posts unless you've seen it elsewhere first, or it's just plain common sense. He may have a source or two which gives him legitimate AMD business info, but his tech info is b.s. more often than not.
I mean, I knew HD 4870 wasn't going to feature an external 512-bit memory interface.
1) HD 4850 and 4870 use the same GPU - why would any GPU manufacturer disable half of the memory controller/channels on a native 512-bit part?
2) How the #^% did ATi manage to fit a 512-bit MC in a 256mm^2 part on 55nm?
Common. $%&*ing. Sense.
wireless technology
ATI may have small number of RV770 Graphics Cards for Announcement next month
http://www.pczilla.net/en/post/18.html
http://www.pczilla.net/en/post/18.html
Of course. Thanks for pointing that one out.
Hey, while we're at it - why not add #4 to the list - 16 ROPs + 512-bit MC =
I'll believe it when I see it. ATi has a strict no paper launch policy in the post-R300-R520 era.
Thanks, I was getting affraid I was missing some fundamental part of R600 design.
Just out of curiosity, how big would a die need to be for 512bit? Does that depend on the process used, or is that always the same? And is there some rough (linear) equation to check how big the bus can be for a given size? Or am I asking too many questions? :smile:
On a completely different note:
This came from AMD's Rick Bergman. It really got me wondering if AMD was just spreading FUD because they can't get to NVIDIA (could this be similar to Intel's raytracing hype?), or if they really see the future so differently from NVIDIA. My own thinking got me this far:
The most often heard argument for smaller chips (or against large monolithic chips) would be the higher yields it would give, thus saving cost. There are a lot of arguments against that though, the most important being all the extra costs associated with a multi-chip solution, such as the double memory, extra packaging, more complex PCB, and of course the die-size for double logic (eg both chips have PCI-E communication). I can hardly imagine all these cost weigh up to the gain in yields a smaller chip would give.
I also read a post from Arun a few days ago saying die size wasn't any problem "unless your design team is composed primarily of drunken monkeys". From what I've read about yields and newer smaller processes the yields are more and more depending on design faults. Still, isn't that an argument against large chips, which are inherently more complex?
The second most heard argument for AMDs new multi-chip appraoch might well be that AMD will somehow connect the two chips together, so the two GPUs could essentialy work as one, also saving the costs and problems with the double memory. I wonder how this could work though. Connecting the ringbus from both chips is what some people say, but I doubt they could make any chip big enough to fit two 512bit connections (R700 still has a 512bit internal ring bus right?)
Also, if AMD wants to make the two chips work with each other like that, why not just put two chips in the same package?
And to add to that, why is chip size becoming such a problem all of a sudden? Yes, GPUs are getting bigger and bigger in the race for top performance, but it's only a gradual increase. I would think power would become a concern long before chip size would be the limiting factor. If so, yields and design complexity would seem to be the only advantages to the whole multi-chip story, which leads me to the question why, if multi-gpu is the future, it is not also the past? Why wasn't this used earlier. Yields and complexity would be problems of all times right? Why don't we have a G80 with four chip with each four SIMDs and two extra chips with the memory controllers on it?
Or is multi-chip just an excuse for AMD to get at least a bit closer to NVIDIA in its post-G80 rampage?
BTW, I'm sorry for all the questions, I'm just not as knowledgable as some (most?) people on this forum are. If the mods think this post belongs in the beginners section rather than here, please go ahead and move it.
Last edited by a moderator:
As I understand your quote, I'd put it this way:
The days of monolithic mega-chips are gone.
AMD has taken the route that there really is only one chip produced and packaged both for performance and mainstream class, the only difference being final clockspeed and type of memory on board. For highend they take the same two chips and stick them on same board.
I think producing only one kind of chip without further physically crippling it greatly simplifies everything from development to logistics and reduces related costs for 90% of cases.
The days of monolithic mega-chips are gone.
AMD has taken the route that there really is only one chip produced and packaged both for performance and mainstream class, the only difference being final clockspeed and type of memory on board. For highend they take the same two chips and stick them on same board.
I think producing only one kind of chip without further physically crippling it greatly simplifies everything from development to logistics and reduces related costs for 90% of cases.
Anarchist4000
Veteran
Would there be any reason why a HTX or some form of serial interface couldn't be added onto the chips? That would at least be capable of cutting down on the number of pins needed. As an added benefit it might be an interesting way of interfacing directly with the CPU for some integrated designs.
Hyper transport seems to work well for communication between CPUs so I'd think it could be adapted to work well with GPUs. The only question would be getting enough bandwidth to pull it off and not adding to much latency.
Hyper transport seems to work well for communication between CPUs so I'd think it could be adapted to work well with GPUs. The only question would be getting enough bandwidth to pull it off and not adding to much latency.
jimmyjames123
Regular
IMHO, it is a bit risky and leaves little margin for error when trying to tackle the high end with only multi-die approaches. If the multi-die approach doesn't give better performance than a larger single die, then it won't gain any traction in the market.
The biggest problem for AMD is the ability (or lack thereof) to fuel R&D costs required for big "monolithic" GPU's. There's no way they can afford to do that, so they have to tackle the high end with multi-die solutions. NVIDIA has a big advantage in the sense that they will have two options: either single larger die or smaller multi-die approach, and they can choose whatever approach gives them the best performance at any given point in time.
If ATI had not merged with AMD to become a combined corporation in much need of cash, then perhaps ATI would have been able to keep going the route of larger single die monolithic gpu. Would have been interesting, but ATI simply has to do the best they can now with whatever resources they have at their disposal, and the monolithic approach is simply not feasible with AMD unwilling or unable to feed billions of dollars into R&D for graphics technology.
The biggest problem for AMD is the ability (or lack thereof) to fuel R&D costs required for big "monolithic" GPU's. There's no way they can afford to do that, so they have to tackle the high end with multi-die solutions. NVIDIA has a big advantage in the sense that they will have two options: either single larger die or smaller multi-die approach, and they can choose whatever approach gives them the best performance at any given point in time.
If ATI had not merged with AMD to become a combined corporation in much need of cash, then perhaps ATI would have been able to keep going the route of larger single die monolithic gpu. Would have been interesting, but ATI simply has to do the best they can now with whatever resources they have at their disposal, and the monolithic approach is simply not feasible with AMD unwilling or unable to feed billions of dollars into R&D for graphics technology.
Last edited by a moderator:
Multi-die doesn't need to give better performance than a single die, it just needs to be competitive performance with a competitive price.
Really? So NVIDIA can choose to throw away all the work they did on a large die in favor of two smaller dies? All the R&D work for the large die is irrelevant? No.
-FUDie
- Status
