I wonder if it's a 1GB or a 768MB GTX 460.
AMD: R9xx Speculation
- Thread starter Lukfi
- Start date
I wonder if it's a 1GB or a 768MB GTX 460.
you really shouldn't, at least not in this thread.
UniversalTruth
Veteran
Very interesting scores, indeed. I am really wondering when and what type of configuration will reach 10K, or 100K. 
And the framerate is interesting. Are those scores artificially limited to so low numbers or we will see slideshow on very powerful configurations?
And the framerate is interesting. Are those scores artificially limited to so low numbers or we will see slideshow on very powerful configurations?
Last edited by a moderator:
HD 5870 shows some pretty great CF-scaling there - I wonder whether HD 6970 will end up nearer to X2000 or X2500 (and whether Antilles will be able to break the X4*** barrier)
HD 5870 shows some pretty great CF-scaling there - I wonder whether HD 6970 will end up nearer to X2000 or X2500 (and whether Antilles will be able to break the X4*** barrier)
same thead page 3
HD6970: X1807
As I don't understand Chinese, I would rather say, 50% faster than HD6870, and a score of X1680:
按迪兰的说法，6970比6870强50%，简单粗暴的套用到这个跑分上只是X1680
from:
http://www.chiphell.com/thread-141469-2-1.html
按迪兰的说法，6970比6870强50%，简单粗暴的套用到这个跑分上只是X1680
from:
http://www.chiphell.com/thread-141469-2-1.html
Or a GTX 460 SE 1GB ...
Never mind, they are within 10% of each other:
http://www.hardwarecanucks.com/foru...178-nvidia-geforce-gtx-460-se-1gb-review.html
Anyway, GF104 will be a sparring partner to Cayman, despite having only 10-20% smaller die size, but that's life.
Unlikely, it's having it's hands full with Barts
Huh?
Bouncing Zabaglione Bros.
Legend
Still no one out there who can throw us a small bone and say when the Cayman NDA expires?
It'll be released on week 50, so most likely NDA ends on week 50 too
chomp chomp chomp... burp! need some more
It might launch (really) close to the 570?
When did you say the 570 would launch, again?
I believe there are 2 possibilities:
a. 30 SIMD (1920 ALUs) divided into 2 blocks, 120 TMUs, and 32 ROPs.
b. 24 SIMD (1536 ALUs) divided into 2 blocks, 96 TMUs, and 32 ROPs.
Now, which one is the most likely?
The fastest 2Gb GDDR5 memory on the market runs at 6 GHz:
· Elpida 50nm @ 1.5V (EDW2032BABG60-F)
· Hynix 44nm @ 1.5V (H5GQ2H24MFR-R0C)
· Samsung 46nm @ 1.5V (K4G20325FC-HC03)
Using this new GDDR5 Cayman can have 20% more memory bandwidth than Cypress.
With 20% more memory bandwidth Cayman can have 20% more ALUs: 1600 ALUs + 20% = 1920 ALUs.
One would thus be tempted to believe that Cayman really have 1920 ALUs, like the fake (?) Antilles's slide said.
But lets dig further shall we?
If Cayman really have 1920 ALUs, it would have 30 SIMD and 120 TMUs, which is 50% more than Cypress.
Would it be judicious to have 50% more TMUs if there is only 20% more memory bandwidth?
I'm inclined to believe that it's not.
That reinforces the idea that Cayman would only have 1536 ALUs.
1536 ALUs = 24 SIMD, which is 20% more than Cypress.
These SIMD being more effective, that would require approximately 20% more memory bandwidth.
24 SIMD x 4 TMUs = 96 TMUs, which is also 20% more than Cypress.
20% more TMUs, 20% more memory bandwidth.
The magic number seems to be "20%".
With 24 SIMD (1536 ALUs) Cayman would have...
· 20% more memory bandwidth, thanks to the GDDR5 @ 6 GHz;
· 20% more SIMD than the Cypress;
· 20% more TMUs than the Cypress;
· 50% more powerful ROPs, at least;
· 100% more geometrical power (tesselation);
· 100% more memory (2 GB).
With these specs, I think that Cayman could be ~30% more powerful than Cypress.
The die size will be smaller than 400mm², which makes it possible to reduce the production costs, and thus the selling price.
Logic wants that Cayman has 1536 ALUs... but perhaps that AMD engineers have lost their minds.
a. 30 SIMD (1920 ALUs) divided into 2 blocks, 120 TMUs, and 32 ROPs.
b. 24 SIMD (1536 ALUs) divided into 2 blocks, 96 TMUs, and 32 ROPs.
Now, which one is the most likely?
The fastest 2Gb GDDR5 memory on the market runs at 6 GHz:
· Elpida 50nm @ 1.5V (EDW2032BABG60-F)
· Hynix 44nm @ 1.5V (H5GQ2H24MFR-R0C)
· Samsung 46nm @ 1.5V (K4G20325FC-HC03)
Using this new GDDR5 Cayman can have 20% more memory bandwidth than Cypress.
With 20% more memory bandwidth Cayman can have 20% more ALUs: 1600 ALUs + 20% = 1920 ALUs.
One would thus be tempted to believe that Cayman really have 1920 ALUs, like the fake (?) Antilles's slide said.
But lets dig further shall we?
If Cayman really have 1920 ALUs, it would have 30 SIMD and 120 TMUs, which is 50% more than Cypress.
Would it be judicious to have 50% more TMUs if there is only 20% more memory bandwidth?
I'm inclined to believe that it's not.
That reinforces the idea that Cayman would only have 1536 ALUs.
1536 ALUs = 24 SIMD, which is 20% more than Cypress.
These SIMD being more effective, that would require approximately 20% more memory bandwidth.
24 SIMD x 4 TMUs = 96 TMUs, which is also 20% more than Cypress.
20% more TMUs, 20% more memory bandwidth.
The magic number seems to be "20%".
With 24 SIMD (1536 ALUs) Cayman would have...
· 20% more memory bandwidth, thanks to the GDDR5 @ 6 GHz;
· 20% more SIMD than the Cypress;
· 20% more TMUs than the Cypress;
· 50% more powerful ROPs, at least;
· 100% more geometrical power (tesselation);
· 100% more memory (2 GB).
With these specs, I think that Cayman could be ~30% more powerful than Cypress.
The die size will be smaller than 400mm², which makes it possible to reduce the production costs, and thus the selling price.
Logic wants that Cayman has 1536 ALUs... but perhaps that AMD engineers have lost their minds.
Those numbers are great, but if those leaked 3dMark11 numbers are true, then Cayman is much more than 30% more powerful than Cypress since it matches more or less the GTX 580.
And the other issue is assuming that Cayman even needs that same boost in memory bandwidth. Cypress apparently wasn't very bottlenecked by memory. And, furthermore, we don't know if Cayman is actually going to be running at 6GHz - AMD often uses chips rated higher than they actually get run at for final clocks. I wouldn't be surprised if they used these 6 GHz rated chips and clocked them at only 5.2-5.4GHz
There's too much of correlation and not enough causation with these numbers.
Actually those numbers look awfully good for a Cayman PRO though
Last edited by a moderator:
In December (Yes I know it's a circular reference)
Similar threads
