How fast do you guys think Cayman will end-up (please vote)

[Shtal]

http://www.techpowerup.com/130393/Picture_of_AMD__Cayman__Prototype_Surfaces.html

 

[chavvdarrr]

as fast as 480 in tests that matter for DX11 hardware - tesselation.
For GPGPU it will be slower - not because lack of hardware power, but because lack of software support&tools.

 

[Shtal]

The colleagues over at VR Zone found some benchmark results of what is said to be the AMD Radeon HD 6870 graphics card, the info is really dodgy though. The numbers were published at the Asian PCinlife Forums, it includes 3DMark Vantage Extreme & Performance, Crysis Very High and Unigine Heaven scores.

The card delivers a score of X12092 in the 3DMark Vantage Extreme benchmark, and P24499 in the Performance benchmark. It's about 20 percent faster than the GeForce GTX 480, and closing in on the dual-GPU Radeon HD 5970. Here are the Crysis and Unigine scores:

But, can it play Crysis? At Very High, 1920x1200, 4x AA, the HD 6870 comfortably plays along at 43.55 fps. The minimum fps is 29.57. Crysis has been well and truly conquered at HD resolutions with high IQ settings. At these kind of settings, previous single GPU cards barely broke 30 fps average, which means the HD 6870's minimum framerate is the same as the average of HD 5870 / GTX 480. Remarkable!

Unigine's Heaven tessellation has been and one of the biggest weaknesses of the HD 5000 series, possibly signifying weak tessellation performance for future games. On the other hand, it was GTX 400's strongest benchmark, with the GTX 480 proving to be an incredible 75% faster than the HD 5870 for the (unrealistic) Extreme preset. Typical scores for 1920x1200 4xAA Extreme are 900+ for the GTX 480 and 500+ for the HD 5870. Amazingly, the HD 6870 scores a whopping 922, which is a massive increase over the HD 5870 and puts it on par with the GTX 480. Do note, however, that the previous results were on Heaven 2.0, whereas the HD 6870 benchmarks use Heaven 2.1. The relevance of the Extreme tessellation benchmark has been heavily questioned as well. Still, it is clear that AMD have taken tessellation very seriously with the HD 6000 series.http://www.google.com/imgres?imgurl...rev=/images?q=AMD+6870&hl=en&gbv=2&tbs=isch:1

This is kinda old info but it could be true

 

[psolord]

Between GTX 480 and 5970 is my guess. Actually its more of a wishful thinking than a guess.

In games that scale well, the 5970 would be quite faster. In games that don't, the Cayman XT should have the upper hand.

A price point of 400$ would be quite nice.

 

[Shtal]

Between GTX 480 and 5970 is my guess. Actually its more of a wishful thinking than a guess.

In games that scale well, the 5970 would be quite faster. In games that don't, the Cayman XT should have the upper hand.

A price point of 400$ would be quite nice.

Assuming Cayman XT is 480(X4) = 1920 Unified shaders.

 

[Lightman]

I'm the brave one who voted faster than 5970 :smile:

But this only in DX11 titles which uses new extensions like DC or Tesselation.
In older titles I'm betting on in between GTX480 and HD5970.

 

[mczak]

Well that's an easy question imho. GTX480 isn't all that much faster than HD5870, so a new, presumably larger chip than Cypress should beat both (that doesn't mean it will beat GTX480 in Unigine Heaven but you get the idea). OTOH, reaching HD5970 sounds a bit unlikely since noone really expects a monster sized chip and it is hard to believe it is that much faster if it's only somewhat larger than Cypress.

 

[psolord]

Assuming Cayman XT is 480(X4) = 1920 Unified shaders.

Yes, that's what I have in mind.

Still, so far, the most intriguing leak/rumor/fake I've seen, is that black slide that was mentioning the RPEs.

Spoiler

I don't know if it holds any ground, but the idea of an RPE, that can be glued together with its brethren and create an ever stronger chip, is just great.

According to the above slide, the Cayman could be a 3 RPE part, with 480X4 shaders, 96TMUS and 48 ROPS. Now if I recall correctly, that 48ROPS does not stick well with the B3D Gurus, but I will go past that, because it gets too technical for me! Such a part could easily be 30-50% faster than the Cypress, which is more than enough for a next gen product.

So I will stick with this theory and hope that it's somehow true. A three RPE part, could make the chip kinda rectangularish, like the GF104 though. :S

One more question that rises from all this, is what Antilles will be consisted of? Will it be a 2X 3 RPE card (2x Cayman) or 2X 2 RPE card (2x Barts)? 2x 2 RPE would make the 6990 a 4 RPE card while the Cayman a 3 RPE card, which seems only fair. The problem lies in scaling though and some games could be faster on the Cayman XT, than the Antilles XT. This could happen now as well though, when comparing the 5870 with the 5970. The difference would be greater on the 2X 2 RPE 6990 scenario though.

One last thing that has been bugging me, is the possibility of a 4 RPE part for the 6990. If the per RPE scaling is great, why the hell not? Yes the chip would be huge, but so what? It's not like they would be making millions of them, plus this part would go to an uber expensive card anyway.

The only thing I cannot fit in all that, is the memory bus. Yes I know, that's important!

All in all, this crazy secrecy on AMD's part, means one thing. They have a big ace up their sleeve and they are going for the kill. Even more so than they did with rv770 and Cypress. So even the craziest scenario, is not crazy at all.

 

[Harison]

Safe bet is between GF480GTX and HD5970. Still its very much possible fastest single-GPU can trade blows or even be faster than previous gen X2. Case and point: 3870 X2 vs 4870 or 4870 X2 vs 5870, same can happen again.

 

[LordEC911]

Safe bet is between GF480GTX and HD5970. Still its very much possible fastest single-GPU can trade blows or even be faster than previous gen X2. Case and point: 3870 X2 vs 4870 or 4870 X2 vs 5870, same can happen again.

I agree... in some cases, all cases being gaming scenarios, Cayman XT should at least be a bit faster than GTX480 except the outliers, in most cases hopefully performing around a 5970 and then the other outlier, due to scaling, might be a bit faster than a 5970.

On average I am guessing around a 5970 ±5%, based on the 5970 being ~30-40% faster than a GTX480 in most cases, only looked at anandtech's GTX480 launch benches so my %s might be a bit outdated.

 

[TKK]

One more question that rises from all this, is what Antilles will be consisted of?

According to AMD's own official drivers, Antilles uses Cayman chips.

 

[Davros]

10 million Bungholio marks

 

[Pressure]

10 million Bungholio marks

 

[digitalwanderer]

I think it'll be able to beat a '68 GTO.

 

[Shtal]

Looking at this slide chart how barts and cayman been compare (performance to cypress GPU), makes me wonder if caymanPRO will consume as much power as cypressXT; if so, I would buy caymanPRO for $300 US. dollars.

 

[psolord]

Interesting slide. Thanks.

It seems that each bubble has a percentage size of 20%. Yes that does not sound right, but I cannot put in words.

To make some examples of it, the 5850 seems to be overlapping 1/4 of the 5870. So leaving 3/4 of space. We all know that the 5870 is around 15% faster than the 5850.

Also the 5970 seems to be two full bubbles higher which would indicate a 40% performance lead over the 5870, which again sounds right.

The 6870 seems to reach around 2/3 of the total 5870 bubble, so they show a 6.7% performance advantage of the 5870 which also sounds right. About the same goes for the 6850/5850 comparison.

So, I could dare say, that the 6950 could be a full 16% faster than the 5870, and the 6970 a full 32% faster than the 5870. That sounds about right also in the respect of how slower the 6970 would be compared to the 5970.

The 6950 also seems like around 30ish % faster than the 5850. Not bad, but not too great either. It will all come down to prices I guess. With a 30% over the 5850, the 6850 could be breathing down GTX 480's neck. Slash 100$ off the 480's price, with better thermals, power draw, noise and you have a winner!

 

[Shtal]

Interesting slide. Thanks.

It seems that each bubble has a percentage size of 20%. Yes that does not sound right, but I cannot put in words.

To make some examples of it, the 5850 seems to be overlapping 1/4 of the 5870. So leaving 3/4 of space. We all know that the 5870 is around 15% faster than the 5850.

Also the 5970 seems to be two full bubbles higher which would indicate a 40% performance lead over the 5870, which again sounds right.

The 6870 seems to reach around 2/3 of the total 5870 bubble, so they show a 6.7% performance advantage of the 5870 which also sounds right. About the same goes for the 6850/5850 comparison.

So, I could dare say, that the 6950 could be a full 16% faster than the 5870, and the 6970 a full 32% faster than the 5870. That sounds about right also in the respect of how slower the 6970 would be compared to the 5970.

The 6950 also seems like around 30ish % faster than the 5850. Not bad, but not too great either. It will all come down to prices I guess. With a 30% over the 5850, the 6850 could be breathing down GTX 480's neck. Slash 100$ off the 480's price, with better thermals, power draw, noise and you have a winner!

What is your speculation "count" stream processors for Cayman GPU ??

 

[Shtal]

Card ---------------------- 3DMark Vantage ----- Unigine Heaven
Radeon HD 6970 ------- P24499 --------------- 36.6 FPS
GeForce GTX 480 ------ P21106 --------------- 29.5 FPS
Radeon HD 5870 ------- P19337 --------------- 17.3 FPS

http://translate.google.com/transla...l=en&u=http://vga.zol.com.cn/195/1953715.html


EDIT:

However, the card is still within the sample card, similar to plate 5870, the core at 800MHz-900MHz frequency range, power supply interface is 6 +8, I believe should be 255W TDP or more.

 

[thatdude90210]

I think that 36.6 number has been around for over a month. From a pic that many felt were Photoshopped.

 

[Shtal]

EDIT: This is what I think now: EDIT: BartsXT is about 0% - 5% percent faster than CypressPRO and I speculation CaymanXT would be about 10% percent slower than (Hemlock) HD5970.