AMD Cayman info (or rumor)

[ColonelCain]

Mmmmm.... Juicy juicy info.

[Frananta]

4gb antiles is really great, finally a true eyefinity card

[zerazax]

The power draw thing seems interesting... the ability to set a thermal threshold directly rather than manipulating clocks/voltages. If they let us set it really high, we could see some interesting results... perhaps in OC'ability too

[Dimitriman]

120 TMUs will be simply bonkers... imagine the texture rendering power of Antilles!

[Heinz68]

price wars incoming !!!!! i'm assuming it'll also have UVD3.0,EyeSpeed etc not mentioned in the slides

AMD sure needs to make some money to pay the debts so I doubt there is going to be any price war.

As for Nvidia, I just checked Newegg and out of 11 GTX 580 listed only cards from one partner are in stock, with such limited supply there want be any need for price war from Nvidia.

[Katzenschleuder]

Hmmm I don't like what I see with the "Tesselation done right / adaptive tesselation", I'm afraid this is a whole new level of driver optimizations and it's going to be hard to see the different level of IQ but the performance increase or decrease could be very great indeed. I see some more PR fighting going on from nVidia and AMD, telling reviewers their level of tesselation is best.

?? That's just an guideline for developers. Nvidia is saying exactly the same thing because too small triangles destroy the point of rasterization!

The tessellation shown in demos is absolutely meaningles for AAA game titles anyway. Tessellation can only be used with accaptable quality for meshes which can be defined completely procedurally (tires, etc.) or for absolutely organic looking objects (which is virtually non-existent in games).
And no one is going to see a difference between a tessellated cobblestone and a parallax-mapped one in a game scene.

There are many quality issues with tesselation in reality like uv-mapping.

So overall, tessellation is not and will not be important for games. There are much more useful things to do with geometry shaders.

[Mad Pistol]

AMD's Turbo Core reached the GPU line.

It makes sense. Since they're a single company now, why not borrow technologies to make their products better overall?

[SimBy]

Power containment is an excellent feature especially for dual-GPU cards. You don't have to factory lower clocks and voltage to stay below 300W. Just slap on two full blown Caymans and limit them to 300W the rest is taken care of automagically. You get max perf out of the card in all but borderline cases.

But it's useful in single GPU cards too of course, especially mobile.

[tbone8ty]

but we wont be able to run gpu viruses at max! (furmark) jkjk

this is alot of info to take in...

what will the heatsink design and temps look like, thats what im curious about

phew the waiting was killin me!! now were in the home stretch

[zerazax]

Power containment is an excellent feature especially for dual-GPU cards. You don't have to factory lower clocks and voltage to stay below 300W. Just slap on two full blown Caymans and limit them to 300W the rest is taken care of automagically. You get max perf out of the card in all but borderline cases.

But it's useful in single GPU cards too of course, especially mobile.

That's a really good point... Antilles is going to be nuts

Really feels like Turbo Core for GPUs

[iMacmatician]

The power containment feature looks really nice.

Is it just me or is it odd that out of the HD 6970 specs that are not given a specific number in the slide, the only one that isn't explicitly stated as being greater than its HD 5870 equivalent is the SPU count?

Not that I'm suggesting the 6970 will have 1536 SPUs, I think 1920 is more likely (if it hasn't been confirmed), given things like previous rumors.

In any case, Cayman's shaping up to be one heck of a chip.

[SicKlown42012]

The power containment feature looks really nice.

Is it just me or is it odd that out of the HD 6970 specs that are not given a specific number in the slide, the only one that isn't explicitly stated as being greater than its HD 5870 equivalent is the SPU count?

Not that I'm suggesting the 6970 will have 1536 SPUs, I think 1920 is more likely (if it hasn't been confirmed), given things like previous rumors.

In any case, Cayman's shaping up to be one heck of a chip.

I think there's two reasons why they don't have finalized specs on that slide. The first is that the rumors about yield issues are true and AMD had to get enough chips back to see which shader count would give them enough chips to sell. The other is that they want to keep Nvidia on their toes and not give them a target. I believe the second is the more likely. But if we do get a card with 30 SIMDs, it's going to be a beast. Let's just hope that AMD doesn't bottleneck the card with not enough memory bandwidth. They haven't done that yet, so it's likely not going to be an issue.

[zerazax]

Keep in mind this was presented in October. Look at the bottom.. October 2010, and an NDA of 11/22.

Most likely the specs were hidden so Nvidia wouldn't get the leaks. Flash forward to the 6990 slide posted a few pages back, which had a bottom date of 11/18, and we see its clear the Cayman XT has 1920 SP's... it's most likely the specs were only revealed on the 18th's conference meant for the 6900's

[tbone8ty]

Keep in mind this was presented in October. Look at the bottom.. October 2010, and an NDA of 11/22.

Most likely the specs were hidden so Nvidia wouldn't get the leaks. Flash forward to the 6990 slide posted a few pages back, which had a bottom date of 11/18, and we see its clear the Cayman XT has 1920 SP's... it's most likely the specs were only revealed on the 18th's conference meant for the 6900's

nice catch

[tajoh111]

I hope this is faster than the gtx 580. I would rather pay 500 dollars for something 15% faster than a gtx 580, than 400-450 for something 5-10% percent slower.

But any word on quantities from anywhere?

[SimBy]

TBD means 'To Be Disclosed' and not 'To Be Decided'. Of course it's already decided internally, they just didn't want nV to know about it.

But like zerazax said, now we know it's most likely 1920 SP, 30 SIMDs and 120 TMUs cause we have the Antilles specs.

[kuroikenshi]

120 TMUs will be simply bonkers... imagine the texture rendering power of Antilles!

OMG this will simply not beat the gtx580!!! sorry NVFB

[Aten-Ra]

From the latest info, I estimate

CAYMAN XT

Dual Front End (Graphics Engines) with double Tessellator Units
30 SIMD 4-way VLIW
1920 Shaders (64 per SIMD) (900MHz = 3,45 Tflops)
120 TMUs (4 Tex Units per SIMD)
2GB 256-bit Memory
32 ROPS
Performance ~30% avg more than HD5870 (DX-9/10)
Performance ~2x DX-11 Tessellation more than HD5870
Price ~$450


CAYMAN Pro

Dual Front End (Graphics Engines) with double Tessellator Units
28 SIMD 4-way VLIW
1792 Shaders (64 per SIMD) (775MHz = 2,77 Tflops)
112 TMUs (4 Tex Units per SIMD)
1GB 256-bit Memory
32 ROPS
Performance ~15% avg more than HD5870 (DX-9/10)
Performance ~1.8x DX-11 Tessellation more than HD5870
Price ~$350

[Sn0wm@n]

Maybe, maybe not. But performance difference between cards is usually reasonably small, but with some smart adaptive tesselation, the two cards reviewed could have a much bigger difference, hence one will come out on top and could therefore have the highest sales. But like you said, which would be best?

I really don't see how any of this applies to fastfood by the way.

burger king mcdonald or anyother burger joints probably try to sell their burger as the ultimate burger .... and amd and nvidia are doing the same with their hardware .... its PR ...

[Dezmen]

From the latest info

CAYMAN XT

Dual Front End (Graphics Engines) with double Tessellator Units
30 SIMD 4-way VLIW
1920 Shaders (64 per SIMD) (900MHz = 3,45 Tflops)
120 TMUs (4 Tex Units per SIMD)
2GB 256-bit Memory
32 ROPS
Performance ~30% avg more than HD5870 (DX-9/10)
Performance ~2x DX-11 Tessellation more than HD5870
Price ~$450


CAYMAN Pro

Dual Front End (Graphics Engines) with double Tessellator Units
28 SIMD 4-way VLIW
1792 Shaders (64 per SIMD) (775MHz = 2,77 Tflops)
112 TMUs (4 Tex Units per SIMD)
1GB 256-bit Memory
32 ROPS
Performance ~15% avg more than HD5870 (DX-9/10)
Performance ~1.8x DX-11 Tessellation more than HD5870
Price ~$350

Fako? Cause 450$ for 6970 is too much... 399$ and 299$... but still not fit...

[safan80]

Fako? Cause 450$ for 6970 is too much... 399$ and 299$... but still not fit...

I'd pay $450... That is with the GTX 580 is selling for and it does not even have 2GB of memory which ends up being a major sticking point for me. You have to realize this is the cost of inflation as well... I might wait for Asus to make an ares version of the 6990 and get that later. $450 also suggests to me that ATI believes it's that good of a card. I'm more concerned that the 6990 might stick to the 300watt pcie power rule. So it looks like I'll get 2 6970s.

[SimBy]

From the latest info, I estimate

CAYMAN XT

Dual Front End (Graphics Engines) with double Tessellator Units
30 SIMD 4-way VLIW
1920 Shaders (64 per SIMD) (900MHz = 3,45 Tflops)
120 TMUs (4 Tex Units per SIMD)
2GB 256-bit Memory
32 ROPS
Performance ~30% avg more than HD5870 (DX-9/10)
Performance ~2x DX-11 Tessellation more than HD5870
Price ~$450


CAYMAN Pro

Dual Front End (Graphics Engines) with double Tessellator Units
28 SIMD 4-way VLIW
1792 Shaders (64 per SIMD) (775MHz = 2,77 Tflops)
112 TMUs (4 Tex Units per SIMD)
1GB 256-bit Memory
32 ROPS
Performance ~15% avg more than HD5870 (DX-9/10)
Performance ~1.8x DX-11 Tessellation more than HD5870
Price ~$350

I actually expect Cayman XT to be Cypress XT + 40%. Putting it slightly above GTX580 on average.

[zerazax]

How are you estimating the performance?

Cypress XT -> Cayman XT is apparently:
320 -> 480 SP's (50% increase)
80 TMU -> 120 TMU's (50% increase)
32 ROPs but they are doubled in power
2x tesselation across the board
1GB -> 2GB
850 MHz -> 900 MHz

Of course, if memory is actually a bottleneck now, they won't scale that well... but 30% seems far far too low given those other specs. Those are some seriously beefed up specs. 50% more TMUs, twice the ROP power, twice tesselation, twice the memory, 50% more SIMDs... that would require some serious bottlenecking to slow down even in DX9/10 games where teselation is a nonfactor.

Actually, looking at that... wow Cayman XT is one serious beast

[kuroikenshi]

lol..... are we ready???

[Therk]

Would anybody be so kind to give me a quick rundown on these cards? I wish I had the time to read the whole thread. Stuff like, rough performance, price, when released etc.

Also from what I've seen from the last couple of pages NDA is the 22nd Nov - so why is there no benchmarks? :\

Many thanks.