AMD Cayman info (or rumor)

[zerazax]

Uh, the 6850 is touching it, but is positioned higher. Difference between the 5770 and 5870 is the same as 5870 to 5970 and it's not about % performance. It's about it's placement....

That was exactly the point I was getting across: I'm saying its silly to use the pixels to extrapolate performance: for instance, its 130 pixels from the top of the 5870 to the top of the 5970, whereas its 120 from the top of the 5870 to the top of the 5770, but we all know that the 5870 and 5770's gap is usually larger than the 5970 to 5870, hence its silly - it's a PR slide meant to exaggerate placement of cards

And that was my point on the 6850, its positioned higher

Of course, if the same people who want to argue that this chart is indicative of performance, then the fact that the 5770's top is 20 pixels higher than the 6850's bottom, and the 5870's top is 20 pixels higher than the 6950's bottom... well if the 6850 is 25% faster than the 5770, are we suggesting that the 6950 is 25% faster than the 5870? If so, then the 6970 will not be slower than the 5970 in gaming, hence the entire argument of using this chart for performance is silly unless one is trying to prove their agenda

[hurrdurr]

ly in a layout like that?

I mean, reading into pixels? If so, then why is the 6850 touching the 5770 when its obvious by now the 6850 is faster?

Er, what? 5770's TOP is touching 6850's BOTTOM. What kind of comparison did you just make

And for that matter, why is the 5970 so much higher than the 5870 when the 5870 should be quite big compared to the 5770?

In the graph, the 5970-5870 difference is slightly larger than the 5870-5770 difference which is actually the case in certain res/AA settings, but not all.

Also, on the fliip side, look at the 6950... it's placed higher than the 5870 so if the 6970 is 17% faster than the 6950 (like 5870 is to 5850), and we say 6950 is even a conservative 10% faster than the 5870, the 6970 is going to be close to the 580 and probably be ~ the 5970 depending on the scenario.

According to the graph, the difference between 6950 and 5850 is about double that of 5850 and 5870. So, if 5850 is 100 performance points, 5870 is 117 and 6950 is 135, and 6970 is around 150 which makes slightly lower than GTX 580 and lower than HD5970.

It's pointless to argue this - the drivers already put Antilles as 2xCayman, so using Barts is a non-sequitur

Didn't know that.

Furthermore, the 5970 was already theoretically crippled - it was 2 x Cypress XT cores clocked at Cypress PRO.

Not this much. Full Cypress was 180W. Full Cayman is probably between 230 and 250W.

AMD can take it a step further and make it 2 x Cayman PRO cores @ Cayman PRO clocks, with Cayman PRO supposed to be < 225W (my guess is in the 175-200W envelope) which would fit with a X2 configuration.

Would it? You think saving one PCB would save 50 to 100 watts? I don't think so.

That would actually leave room for AMD to keep the higher binned XT cores to reach higher clocks and thus make the 6970 stand out even further while the ones that couldnt reach such clocks get clocked lower / have units disabled and put in the PRO and X2 config

This doesn't make sense. "Ones not reaching such clocks" means "ones not reaching such clocks under a certain voltage". If you want to save power, which is definitely what you want to be doing in Antilles, you have to use the best binned cards possible, so that they will reach the same clocks using lower voltages, ending in lower power usage. Using lower binned GPUs would merely mean you having to use a higher voltage to reach the same clocks as the higher binned GPUs.

[tajoh111]

That slide makes me want to buy a gtx 580. I don't think the disclaimer about performance would be their if this chart didn't at least somewhat show performance differences between cards.

Not only that the performance between everything released so far has a pretty accurate distance between them. E.g the difference between a 5770, 5750, a 6870 a 6850, and a 5870, 5850 are all about 14 % according to tech report and the distance in games between a 5870 and a 5970 is similar to the difference between a 5870 and a 5770.

With a poor extrapolation from that chart, I would imagine the 6950 10-12% faster than a 5870 and a 6970 12-15% faster than a a 6950. Pretty believable numbers although somewhat disappointing.

I think this looks alot more reliable than the few leaks which have proven not very accurate so far and/or much harder to judge for accuracy.

You and me are thinking the same way hurdurr. 150/117 = 1.28% performance difference between the 5870 and 5970. This is about 5-10% slower that a gtx 580, I think this is he most realistic scenario for the 6970, because I don't see AMD going mammoth size with the 6970 and they are still on the same node.

[Dimitriman]

you really cant wait 12 days to see the performance? plus there will likely be price cuts on gtx 580.
btw if amd doest give actual performance figures they will not suddently reveal how fast their products are in an overall chart like that.
take the 6850 as an example its placed just above the 5770 on the graph but performs as fast and sometimes faster than 5850.

[tajoh111]

you really cant wait 12 days to see the performance? plus there will likely be price cuts on gtx 580.
btw if amd doest give actual performance figures they will not suddently reveal how fast their products are in an overall chart like that.
take the 6850 as an example its placed just above the 5770 on the graph but performs as fast and sometimes faster than 5850.

I would like to be somewhat optimistic but man there is simply no leaks and I for one would leak something at this point because it looks like there are real quantities of the gtx 580, not paper quantities like original imagined. This is the best indication of performance directly from AMD, everything else has been hype or stories from Charlie or his sources.

And I think your point about the 5850 and the 6850 more or less proves the accuracy of this chart. The 5850 is faster overall compared to a 6850, this is fact and the chart indicated the 5850 performing between a 6850 and 6870. This is completely true when taken from an average of games and even in percentage terms, this chart is not so far off base.

[zerazax]

Since we're arguing performance numbers, Xbitlabs has (arguably) the best performance summary roundup for all our arguing needs

http://www.xbitlabs.com/articles/vid..._12.html#sect2

Er, what? 5770's TOP is touching 6850's BOTTOM. What kind of comparison did you just make

You claim you did pixel analysis didn't you? You can clearly see the 5770's TOP is 20 pixels above the 6850's bottom.. the same difference as in 5870 vs. 6950

That's the point I was using to post the rest, but it ended up getting carried to my 2nd response. (And my overall point is that it's silly to use pixel analysis, read on to see more)

In the graph, the 5970-5870 difference is slightly larger than the 5870-5770 difference which is actually the case in certain res/AA settings, but not all.

Everything is relative based on certain res/AA of course, but the overall gap is not in the 5970's favor unless you're talking about absurd scenarios where the 5770 should never have even been put in like 2560x1600 4xAA or something, which skews the average.

And again, the key word is overall - esp. when you factor in CF scaling/driver issues, the overall gap most people agree with is that 5970 gets an average 40% more

According to the graph, the difference between 6950 and 5850 is about double that of 5850 and 5870. So, if 5850 is 100 performance points, 5870 is 117 and 6950 is 135, and 6970 is around 150 which makes slightly lower than GTX 580 and lower than HD5970.

Again I think that's reading too much into it. The 6950 and 5870's position is exactly the same as the 6850 and the 5770.

Now, unless you think the card performance is EXACTLY the same, what's more likely: this is actual performance, or pure aesthetics? I really highly doubt the AMD guys put the cards position on this PR slide to be exact, esp. since the disclaimer says these are just estimates. SO most likely, it's all aesthetics

Didn't know that.



Not this much. Full Cypress was 180W. Full Cayman is probably between 230 and 250W.


Would it? You think saving one PCB would save 50 to 100 watts? I don't think so.

You do realize prior to release, the 5870 was listed as < 225W on the slide and ended up significantly lower?

Cayman is still unknown, but a lot of that may be up to how much they want to tweak power in anticipation of the 580 or whatever. Saying <225W and <300W only means the cards will be somewhere between 150 and 225, and 225 and 300, which tells us next to nothing other than that one is 6+6 pin and the other is 6+8 pin.

They did after all put Barts XT at > 150W and some people were thinking it was ~170W at release, and it ended up being at exactly 151W so go figure

And NONE of this is factoring in the fact that ATI's TDP rating is different from Nvidia's - I forgot the specifics, but IIRC AMD is using the total board power, so it's not a simple 2x conversion as a single board with different VRMS and what not will definitely change total board power. You said it yourself: 188W rated full Cypress could be made into 294W rated Hemlock with some downclocking. There's no reason to think that Cayman PRO, which might actually eat less juice than Cypress (depending on clocks) can't make it either, though it'll be awfully close. You'll have to dig back in this thread to find the definition of the TDP ratings tho

This doesn't make sense. "Ones not reaching such clocks" means "ones not reaching such clocks under a certain voltage". If you want to save power, which is definitely what you want to be doing in Antilles, you have to use the best binned cards possible, so that they will reach the same clocks using lower voltages, ending in lower power usage. Using lower binned GPUs would merely mean you having to use a higher voltage to reach the same clocks as the higher binned GPUs.

No, my point is that they might want the cards that can clock best to be put in XT, with the ones that can't clock as high get some units removed and downclocked to save power and put into PRO and Antilles configurations.

Whether AMD wants Antilles to be as great an overclocker as Hemlock is a different story - for all we know, AMD doesn't care if Antilles is a great overclocker this round so they're just going to put 2xCaymans that fit in the envelope and forget about whether they're binned or not.

And the whole issue of AMD crippling the 6990... that's been the trend, it's nothing new.

3870X2 was a higher clocked 2X RV670XT
4870X2 was exactly 2X RV770XT
5970 was two lower clocked 2X Cypress XT

Going off that trend, 6990 might well be 2X even lower clocked Cayman XT's or 2Xs lower clocked PROs

I would like to be somewhat optimistic but man there is simply no leaks and I for one would leak something at this point because it looks like there are real quantities of the gtx 580, not paper quantities like original imagined. This is the best indication of performance directly from AMD, everything else has been hype or stories from Charlie or his sources.

And I think your point about the 5850 and the 6850 more or less proves the accuracy of this chart. The 5850 is faster overall compared to a 6850, this is fact and the chart indicated the 5850 performing between a 6850 and 6870. This is completely true when taken from an average of games and even in percentage terms, this chart is not so far off base.

No leaks has been the trend for surprising cards usually. See: G80, RV770 vs. R600 & GF100. Hell, for that matter, AMD didn't announce the actaul release date of Barts until literally the fact that October 19th was the actual date, and then people realized 22nd it was...

And if you follow history, remember what happened with GT200 and the 480SP RV770? I mean, feel free to buy the GTX 580, but I think your just doing what you've been wanting to do, since you said it yourself: there appears to be real quantities... well.. worst case, prices stay the same in 12 days. Best case, you save yourself a major headache and a lot of money. But like I said, do what you really want to do

And again for the last point, you too are reading WAYYY too much into the chart. They're just performance placement, but in all likelihood set to be AESTHETICALLY pleasing! It's a PR slide for a reason - why would AMD cryptically release actual performance through a slide on product placement?

And for that matter, who remembered when RV770's PR slides had the 4870 competing against the 9800GTX and the 4850 against the 9800GT? That turned out to be a riot. Or when those same product slides put 4870 barely faster than R680 (the 3870X2)... that was a hoot

Or when the 4870X2 was placed as a higher product than the 5870, but in real world performance the 5870 is a better performer...

Or when Barts PRO and XT were compared to the GTX 768 and GTX 1GB, but the 6850 was a better match for the GTX 1GB and the 6870 the GTX 470...

Basically, don't read too much into how some power point designer decided to line up graphs

Or when the

[kuroikenshi]

According to the graph, the difference between 6950 and 5850 is about double that of 5850 and 5870. So, if 5850 is 100 performance points, 5870 is 117 and 6950 is 135, and 6970 is around 150 which makes slightly lower than GTX 580 and lower than HD5970.

150% more means according to your logic 6970 is only 1/2 times better than a 5850.....

[highoctane]

I think the placement is more relative to market segment as a visual representation.

Besides:

Could mean more or less, only the creator of the map really knows.

[kuroikenshi]

That's right, its just a drawing on a ppt file, it depicts progress that's all, if we took the slide too serious then we would expect an OC'ed 5970 to be just as good as Antilles. And we all know that would never be the case...

[hurrdurr]

150% more means according to your logic 6970 is only 1/2 times better than a 5850.....

Hmm, 50% faster than HD5850 would make a card on par with a GTX 480, thus 15% slower than GTX 580. You're right, I don't think it sounds right for HD6970.

[kuroikenshi]

Hmm, 50% faster than HD5850 would make a card on par with a GTX 480, thus 15% slower than GTX 580. You're right, I don't think it sounds right for HD6970.

It's ok, I was gonna make a harsh replay, but I gathered that you haven't used % systems before.

I think what you meant could've been somewhere in the 250% ( 1 and a half times better if you make 5850 100%)

However to be more precise I would use a more complex number analysis, where every feature is tested and ranked and then contrasted, it takes many many hours and also require one to have both items being compared.

[spursindonesia]

I'll grab a popcorn basket and happily watching over the side for all of these silliness.

[kuroikenshi]

I'll grab a popcorn basket and happily watching over the side for all of these silliness.

LOL OMFG guess what I have right now in the microwave.....an actual bag of popcorn, wow.

On another factual serious note the presentation was made not long ago, and you should expect all sorts of comments to be thrown around, there is been quite a lot of anticipation for these cards to come out.

[zerazax]

Well the good news I guess is that the cards are still set for Q4 2010 so those rumors the cards wouldn't be out this year should be thrown out the door

Can't wait to see what happens. Would be sweet to see cards battle again at all segments... it's been quite some time

[Chickenfeed]

I think the placement is more relative to market segment as a visual representation.

Besides:

^^^^

This

[Zloyd]

If amd was confident that 6970 was faster than 580 they would be shouting it out from the rooftops for sure, we might still see that in a couple of days, leaked graphs like we saw for 580. I think its going to end up how previous generations have ended up on the single chip front, its going to be about 10-15% slower and its going to be 20% cheaper.

We heard a few noises about 6970 being hot and powerhungry, I think that might be whats delaying the launch, it cant justify its powerconsumption and heat atm ?

[Nintendork]

The "official" launch is the 22. Are we in november 22?

[zerazax]

If amd was confident that 6970 was faster than 580 they would be shouting it out from the rooftops for sure, we might still see that in a couple of days, leaked graphs like we saw for 580. I think its going to end up how previous generations have ended up on the single chip front, its going to be about 10-15% slower and its going to be 20% cheaper.

We heard a few noises about 6970 being hot and powerhungry, I think that might be whats delaying the launch, it cant justify its powerconsumption and heat atm ?

That makes no sense. AMD's been on top of their game in terms of efficiency and heat for the past two generations, why would they suddenly hit major problems on a process that they're on their 3rd generation for?

And AMD hasn't released info on their card because it's a HUGE business risk to advertise their product before its release. It's been covered repeatedly throughout this thread, but people repeatedly refuse to understand the business of GPUs.

So I'll put it this way: name the GPUs that have been extremely hyped up that have been successful, and name the GPUs that have been wildly successful that were silent til release.

You'll see a pretty obvious trend

[tbone8ty]

lets just hope AMD put a little more effort into their heatsink/fan setup for the 6900 series. 6870 heasink/fan is fine for its price point but cayman better have a quality setup like Nvidias 480/580 series cards.

pretty sure delays r driver related so well get cayman a yr from now...haha i keed i keed

nah but seriously some new info better come out soon or im gonna buy an overpriced 6870 from newegg so i can kill some zombies in black ops

[snoid_zero]

Another article from fudzilla. not Fuad this time.

AMD delays Antilles dual-GPU board
http://www.fudzilla.com/graphics/ite...dual-gpu-board

[tajoh111]

Radeon HD 5870 is still AMD's fastest single-GPU card - yeah, we know what's coming - and it falls comfortably behind both the GTX 580 and 480 GPUs in the performance stakes.

http://www.hexus.net/content/item.ph...=27307&page=18

Did anyone catch this in the hexus review? it seems to concern cayman I think.

[zalbard]

Radeon HD 5870 is still AMD's fastest single-GPU card - yeah, we know what's coming - and it falls comfortably behind both the GTX 580 and 480 GPUs in the performance stakes.

http://www.hexus.net/content/item.ph...=27307&page=18

Did anyone catch this in the hexus review? it seems to concern cayman I think.

No, lol. Not really. Beating 5870 for Cayman will be ridiculously easy. Hell, even 6870 comes close at times.

[rekleif]

Radeon HD 5870 is still AMD's fastest single-GPU card - yeah, we know what's coming - and it falls comfortably behind both the GTX 580 and 480 GPUs in the performance stakes.

http://www.hexus.net/content/item.ph...=27307&page=18

Did anyone catch this in the hexus review? it seems to concern cayman I think.

It just means they know what's coming. But for now Ati's 5870 falls comfortably behind the 580 and 480, which it does.

[neliz]

Cayman is not delayed.

[Kuroimaho]

Oops seems the 6990 got bulldozed, aka delayed. Not that after a pair of 5870 I would get another CF config, but at least cayman can still make it this year.