AMD Cayman info (or rumor)

[Vipeax]

So, my point its is HD6800 or HD6900 cards

Question was if it (Cayman) would have a hole there too --> heat.

[[XC] hipno650]

I realised that this thread is almost as epic as this Fermi-thread:

http://www.xtremesystems.org/forums/...ighlight=fermi

Last two intresting and hyped graphic card releases made about 2000 replies and 200 000 views. I wonder is there some pattern to it ?

My optimistic guess is that 6970 will be about as fast as 570. If its faster, thats even better.

edit. i noticed Fermi-thread continued to other threads so Cayman hype isnt even half of what Fermi was..

i remember before that there was the R600 threads, now those were epic, I still remember reading some crazy rumor that R600 scored 30K 3dmark06 single card on a stock processor. that has to be the most ridiculous rumor in GPU history. that card was more over-hyped then Fermi by far. because it was the first ATI card changing from pixel pipelines to Stream Processors most people thought that the shader units were directly comparable to the Nvidia ones. and considering how fast the 8800gtx was most people thought the 320SP's on the 2900xt were going to blow Nvidia out of the water... and what did we end up with, a hot, power consuming card (invention of the 8pin PCIE) that performed lower then a 8800GTS... at least Fermi was a good bunch faster then the 5870...

I don't think this will be a R600 repeat at all but I do think it's being overhyped by a long shot. Im thinking 6970 comes in just slower then the 580 and the 6950 just slower then the 570. Think about it the GTX 570 is just as fast as the GTX 480. Considering the difference between the 570/480 and the 5870 for the 6950 to beat the 570 with LESS SP's then the 5870 and a lower clock speed... that would be a feat and a half

[zerazax]

I don't think SP counts are really comparable with the move to 4VLIW

And 1536/4 = 384 > 1600/5 = 320, if AMD's "4-VLIW is roughly equal to 5-VLIW" is true

And besides, the bigger issue is how the cards perform relative to their other offerings. The 6870 trails the 5870 but not by a whole ton (it also has fewer SP's than the 5870 by a lot, but is close) - if the ridiculous expectations that 6970 is slower than 570 were true, there'd be no room for a 6950 at all.

edit: Oh yeah, LOL @ FUD being caught basically quoting the damn slides. Read his purported numbers and what the slide says.. lol

[Oztopher]

Take into account that the 6870 is only an inch away from the 5870, with only 1120 SP's. A 6950 with (supposedly) 1536 SP's is going to beat the 5870 by a bit, which puts it in 480 territory.

edit: zerazax pretty much beat me to it.

[Motiv]

well if tessellation is better on AMD side, then hopefully gaming is too.

to be better in a very specific synthetic benchmark but worse in all games, i would call that a failure, even as an AMD/ATI owner and fan

Well the only benchmark should be games. Future gaming performance, based on the tessellation engine, is plainly ridiculous.

The vast majority of games are console ports, they basically translate to dx 9/10 games. Once DX11 gaming takes hold for the PC, the 6850, 6870, 6950, 6970, 580 and 570 will be underpowered.

Synthetic benchmarks have always been the fallback ones for fanbois. If there company of choice isn't much cop, then there will surely be a benchmark that still looks good.

Just look at anything that involves tessellation or physx. You can count the games on one hand that use them (to a degree it's possible to see differences between AMD/Nvidia), but fanbois will search them out and point to the scores.

The same can be said with AMD too. Anyone remember all the threads telling us how having 27 monitors at home is the future.

[tbone8ty]

The vast majority of games are console ports, they basically translate to dx 9/10 games. Once DX11 gaming takes hold for the PC, the 6850, 6870, 6950, 6970, 580 and 570 will be underpowered.


.
. .

The 6970 and gtx 580 underpowered?

[Oztopher]

He said once DX11 properly takes hold they will be. Pretty much like it is every single year.

[SKYMTL]

Take into account that the 6870 is only an inch away from the 5870, with only 1120 SP's. A 6950 with (supposedly) 1536 SP's is going to beat the 5870 by a bit, which puts it in 480 territory.

edit: zerazax pretty much beat me to it.

Actually, the HD 5870 is ~ 30% slower than the GTX 480 on average at 1920 x 1200 / 4xAA. Meanwhile, the HD 6870 is usually ~15-20% slower (average again) than the HD 5870.

The needed 40-50% improvement over the HD 6870 to hit GTX 480 territory is a huge jump no matter which way you look at it.

[tbone8ty]

He said once DX11 properly takes hold they will be. Pretty much like it is every single year.

Are there more features on dx11 not being utilized yet?

Crysis 2 and rage will be a good test

[Vipeax]

Actually, the HD 5870 is ~ 30% slower than the GTX 480 on average at 1920 x 1200 / 4xAA. Meanwhile, the HD 6870 is usually ~15-20% slower (average again) than the HD 5870.

The needed 40-50% improvement over the HD 6870 to hit GTX 480 territory is a huge jump no matter which way you look at it.

??

[informal]

Actually, the HD 5870 is ~ 30% slower than the GTX 480 on average at 1920 x 1200 / 4xAA. Meanwhile, the HD 6870 is usually ~15-20% slower (average again) than the HD 5870.

The needed 40-50% improvement over the HD 6870 to hit GTX 480 territory is a huge jump no matter which way you look at it.

5870 1GB is approx ~7% slower in 25x16 resolution and ~8% in 19x12.
AMD needs ~26/30% improvement with 6970 at 19x12/25x16 resolutions to basically match 580GTX 1.5GB.
Also your numbers for 6870 are incorrect.

EDIT: I see Vipeax beat me to it .

[SKYMTL]

5870 1GB is approx ~7% slower in 25x16 resolution and ~8% in 19x12.
AMD needs ~26/30% improvement with 6970 at 19x12/25x16 resolutions to basically match 580GTX 1.5GB.
Also your numbers for 6870 are incorrect.

EDIT: I see Vipeax beat me to it .

Sorry. No. Too many ancient games and synthetics thrown into that result unforunately and I believe Mike uses a combination of AA AND non-AA numbers. In newer games (DX11 predominantly) with AA enabled, the gap is much more significant.

I should also say that I am referring to IN-GAME results here and not certain built in and stand-alone benchmarks that have been proven to completely screw up results.

[spursindonesia]

Let's see what ComputerBase.de has to say about it:

HD 5870 relatif performance vs others



Then onto HD 6870



EDIT:

Hey, what about DX 11 games only

   *  Battlefield: Bad Company 2
    *  Battleforge
    *  Colin McRae: Dirt 2
    *  Metro 2033
    *  Stalker – Call of Pripyat

:



Ofcourse, if you only bench "games" like Metro, HawX 2, Batman AA, Unigine Heaven, StoneGiant, .......

[[XC] hipno650]

you guys trust that site? i don't they have put out a few "interesting" reviews in the past.

http://www.hardwarecanucks.com/forum...review-22.html

GTX 570 to 5870 at 1920(4X AA) is about 32%

Factor on another 10% depending on the game to the 6870 and thats 40% at the most popular settings for PC gaming... to make up that much ground with only a little more Mem bandwidth and the same number of ROP's as a 6870... that will be a feat no matter what they do to the shaders. considering the clock speed will most likely be lower as well if anything the 6950 will be a touch slower then the 570 and the 580 and 6970 will be neck and neck i would guess

[spursindonesia]

So, HardwareCanucks is the only representative of the truth right now ?

[zerazax]

to make up that much ground with only a little more Mem bandwidth and the same number of ROP's as a 6870...

ROPs are doubled-up in functionality, according to slides

[h0bbes]

Different review sites tend to favour one company over the other and people will just pick the one that agrees with their views.

[Alex-Ro]

HardwareCanucks just suck,only true test that i am certain of are the one from my own PC.Everyone out there has an interest/in influenced/is given money/makes free advert of their favorites.Only things i am certain of are guys from lab501.So,u realize how pitifull you are when you debate 86 pages of forum for nothing sure?You are waisting resources for nothing,stop your useless discutions and wait to know for sure

[-=DVS=-]

No matter how you slice it , it is still alot of ground to cover on same 40nm node , 6970 will need to be a big chip , away from AMDs tiny chip philosophy.

[Oztopher]

Are there more features on dx11 not being utilized yet?

Crysis 2 and rage will be a good test

Yep that's true, but what was meant was when DX11 becomes the norm, then the likes of the GTX 580 will probably be obselete. In fact i don't think DX11 will ever become the norm, DX10 hasn't and that's because of dev's coding for consoles. By the time next gen consoles come out, DX12 or whatever will probably be here.

I mean, GTX 580 gets about P6000 on 3Dmark 11, it's only a matter of time before single GPU's will be hitting 20k easy, albeit 2-3 years from now.

I think i'm starting to state the obvious now.


SKYMTL; considering that there's every possibility you have the cards in your possession now, you're either gearing us up for something focused purely on price/perf. or simply playing them down for more surprise

I'm guessing that 32% over the 5870 includes tessellated games as well? Don't forget the fact that fermi rocks with tessellation. But also that cayman will apparently be 2x the tessellation power of 5870, there's another factor to consider when judging 6950 performance.

[generics_user]

you guys trust that site? i don't they have put out a few "interesting" reviews in the past.

http://www.hardwarecanucks.com/forum...review-22.html

GTX 570 to 5870 at 1920(4X AA) is about 32%

Factor on another 10% depending on the game to the 6870 and thats 40% at the most popular settings for PC gaming... to make up that much ground with only a little more Mem bandwidth and the same number of ROP's as a 6870... that will be a feat no matter what they do to the shaders. considering the clock speed will most likely be lower as well if anything the 6950 will be a touch slower then the 570 and the 580 and 6970 will be neck and neck i would guess

while i don't agree with their hardware choice in driver specific reviews (no dualgpu) and hate for SLI and Crossfire configs their single GPU reviews were pretty good in the last few years


additionally somethign is fishy with that hardware canucks chart, the difference between 4xAA and no AA is significantly higher than in every other review out there were 5xxx and 4xx/5xx AA scaling is much closer, their results are completely different to most user tests and professional reviews and completely overrated....

[herderien]

real answers will comme very soon and some of us will be athorized to release the famous :

"i told you so"

we're still not sure about what it is "they told us so"

my guess is :

*rolls a 100 dice"

50% faster then 5870

[Manicdan]

I'm guessing that 32% over the 5870 includes tessellated games as well? Don't forget the fact that fermi rocks with tessellation. But also that cayman will apparently be 2x the tessellation power of 5870, there's another factor to consider when judging 6950 performance.

i took a quick glance over his review, and i think its pretty clear that if the 6970 is right around 2x 6850 it will decimate. the shortcomings of the 5870 which helped lead to that 32% difference were in tessellation, which is where the 6800s were passing the 5800s.

[bl4ckdr4g00n]

[zerazax]

been proven fake beyond a doubt

god, this thread went to hell