AMD Cayman info (or rumor)

[zalbard]

If it takes 1536 shaders to perform better than the GTX 580 then I'm down...

Yeah...
GTX480 -> GTX580: +32 shaders
5870 -> 6970: -64 shaders
They might be -somewhat- more efficient, but come on, WTF!

[Aten-Ra]

Don’t count the SPs 320 vs 384
Count the shaders or the SIMDs 20 vs 24
And yes ROPs (render Back Ends) will be 2x the performance of Cypress

[zerazax]

Not only is 384 more than 320, but hasn't there been rumors about improved ROPs aswell?

Yes, the slides said they doubled the functional ability of each ROP

Don’t count the SPs 320 vs 384
Count the shaders or the SIMDs 20 vs 24

That's the exact same ratio..

[Dimitriman]

If 1536 shaders is true, then...

http://www.youtube.com/watch?v=xHS-fOaO87w

[Dami3n]

If last rumors are true, Nvidia will sell a lot of GTX570

[WintersSC]

If last rumors are true, Nvidia will sell a lot of GTX570

Maybe those SMs are super SMs xDD

[Oztopher]

Had someone over on OCUK forums saying 6900 is fantastic, for the price. So looks like AMD are still keeping their strategy. I don't mind too much tbh; i'd rather have the low prices, than the 'new r300' but expensive.

Looks like it's going to be a 6950 for me then, GTX 570 a bit too expensive atm.

And i can easily see how 1920 SP's could get thrown in, as generics_user said.

[Calmatory]

~20 % increase in shader power(minus the small inefficiency in rare cases of high ILP), "double the functionality" of ROPs. Doesn't sound that bad. However 50 % increase in shader power would've been way better. But guess AMD knows better.

[eric66]

i hear that 6970 is close to 580 .....

[kaktus1907]

if 1536 sp is true, then a cheap 5870 will be better choice than a 6970 for boinc users like me.. mw, collatz and dnetc is already a lot faster with vliw5 sp mArch..

[Mechanical Man]

I have to say, no idea anymore what to expect from Cayman. In otherhand it is clear fact that Cypress was highly inefficient design, remember double shaders but only ~60% more performance compared to HD4870 (if memory serves, feel free to correct if wrong]. Same happened on HD4870 vs HD3870, performance increase was not straight up comparable to shader increase. Basically that leaves load of room to improve. I would say +40% to Cypress if they can find all bottlenecks and figure how to remove them. I believe it can then "trade blows" with GTX580. That would leave room for Cayman Pro at +10-20% over Cypress. But there is lots of if's in that.

[Particle]

Regarding complaints about die size, power, and shader counts: Remember that this ASIC was never supposed to exist on 40nm. They're just doing the best they can to deal with the target process having become unavailable.

[SpuTnicK]

Unless HD6970 performs quite close to GTX 580 for that amount of money (450-450$), it is better (performance/ price) to get 2xHD6850 and enjoy performance level of the last

BTW here is a short teaser
http://www.youtube.com/watch?v=tuYck...layer_embedded

[jeanjean15]

BTW here is a short teaser
http://www.youtube.com/watch?v=tuYck...layer_embedded

This looks like a striptease but we see nothing .

[paulhamm]

Just a bit of fun with numbers. Caymen uses 90% of the die area for 4sps that Cypress uses for 5sps. Check the slides if you doubt. I used barts as a reference.
Cypress 334mm2 1600sp 256bit mem
Barts 255mm2 1120sp 256bit mem
Caymen ???? 1920sp 256bit mem
GF100/GF110 550mm2 512cuda 384bit mem
GF104/(GF114) 368mm2 384cuda 256bit mem

Die area estimates just for fun. Yep they have long WAGgy tails.
Caymen 334mm2/(1600/5)*(1920/4)*.9=450.9mm2
Caymen 334mm2/1600*1920=400.8mm2
Barts 334mm2/1600*1120=233.8mm2
Better Barts 334mm2/1600*1280=267.2mm2 What if HD6870 is a cut down Barts?
Barts reference accuracy 233.8mm2/255mm2=0.9168
Better Barts reference accuracy 267.2mm2/255mm2=1.0478

Hmm 1536sp

Die area estimates just for fun. Yep they have long WAGgy tails.
Caymen 334mm2/(1600/5)*(1536/4)*.9=360.72mm2

Caymen die the size of a gf104 with the performance of the gf110? Ouch!

[kuroikenshi]

This looks like a striptease but we see nothing .

welcome to PG15+ content

[FischOderAal]

if 1536 sp is true, then a cheap 5870 will be better choice than a 6970 for boinc users like me.. mw, collatz and dnetc is already a lot faster with vliw5 sp mArch..

I'm sorry to burst your bubble, but I doubt AMD/ATI developed Cayman (mainly) with GPGPU in mind

[kaktus1907]

I'm sorry to burst your bubble, but I doubt AMD/ATI developed Cayman (mainly) with GPGPU in mind

Yeah you are right. i totally forgot Cayman is able to do double precision as 1:4.. It will be great card for MilkyWay.. But for single precision i dont expect much difference than cypress. We will see soon

[jvc]

Can we end the "slower than nvidia" speculation please? I have no reason not to trust this guy, and no I'm not trying advertise his site.

http://forums.overclockers.co.uk/sho...&postcount=707

Price is about the same as nvidia counterparts

[TheBreezyBB]

5 freaking long days and this would be put to rest.

[zerazax]

Can we end the "slower than nvidia" speculation please? I have no reason not to trust this guy, and no I'm not trying advertise his site.

http://forums.overclockers.co.uk/sho...&postcount=707

Price is the same or higher than nvidia counterparts

From that:

Hi there

All I can say is they are bloody fast.

For the money I am amazed.

I can see this launch being hugely sucessul for ATI, easily on par with 58xx series but better because I literally have enough stock to prevent running out which will also mean our pricing will be excellent and I should be able to keep the price low low, no price hikes.

Can't wait! 5 freakin days

[zalbard]

It can't be right! He said ATI, not AMD!

Can't wait! 5 freakin days

+1
Even though I can not afford those atm, damn, it's bone thrilling!

[Death^Dread]

I can see this launch being hugely sucessul for ATI, easily on par with 58xx series but better because I literally have enough stock to prevent running out which will also mean our pricing will be excellent and I should be able to keep the price low low, no price hikes.

I hope so. It would be great to see a well done launch (without shortages/price gouging), they are not as common as they should be.

[Loque]

the US government should learn a trick or two from AMD on how to keep stuff secret..

maybe breaking their nda means a fate worse than death and listening to Justin Bieber ?

[Nintendork]

I think AMD listened a bit from Nintendo. No one in hell can leak any Nintendo future consoles data (remember revolution specs or the "what kind of controller it will have").