It's 6~10% at Hardware.fr: http://www.hardware.fr/articles/894-23/recapitulatif-performances.html
I guess the average across all reviews is probably somewhere around 10~15%. That seems like the bare minimum for a new generation.
AMD: Volcanic Islands R1100/1200 (8***/9*** series) Speculation/ Rumour Thread
It's 6~10% at Hardware.fr: http://www.hardware.fr/articles/894-23/recapitulatif-performances.html
I guess the average across all reviews is probably somewhere around 10~15%. That seems like the bare minimum for a new generation.
Actually, if you take a look at latest review at TPU, you will see for instance the MSI GTX 780 TwinFrozr Gaming just 10$ above normal ones, while offering ~30% higher performance...
In the other case, it is 20%, not 15-18%
http://www.techpowerup.com/reviews/MSI/GTX_780_TF_Gaming/26.html
I don't expect anything from them, especially good. I mean, if I have high expectations, I am always disappointed by them.
So, yes, maybe they will be able to release something like 15% faster
Ok.... you take an AIB OC card for compare it with an one old year gpu... In addtion TPU have a strong list of games from high gain on the Nvidia side, this push allways exagerate result.
But anyway not really important.. i dont see AMD ditch Sea island ( as it seems ) for end with 15-18% more performance of the old generation.. i got this with a 7970 at 1150mhz.
Now.. the problem is maybe more on the 28 vs 20nm .. and so whatever is the next generation, there will be a die shrink if the cards are still on 28nm ..
Last edited by a moderator:
And the MSI card has a <5% core clock advantage but is generally seeing higher than a 5% increase in performance... something is wrong there.
Not to mention a review that doesn't monitor actual clockspeeds, thanks to both side's GPU boost, and makes no mention of how they actual test their cards.
55 dx11 cards , 18 games + 4 gpgpu test suites: http://ht4u.net/reviews/2013/55_directx11_grafikkarten_im_test/
1080P
1440P
GPGPU
Plus a searchable database with all of their data points: http://ht4u.net/reviews/2013/55_directx11_grafikkarten_im_test/index32.php
1080P
1440P
GPGPU
Plus a searchable database with all of their data points: http://ht4u.net/reviews/2013/55_directx11_grafikkarten_im_test/index32.php
UniversalTruth
Veteran
For now, the most decent rumour I have read so far, is that the new top card will compete with GTX 780, and not with Titan
I don't know how to define what will be more decent than this
http://www.sweclockers.com/nyhet/17...force-gtx-titan-siktar-istallet-pa-700-serien
I don't know how to define what will be more decent than this
http://www.sweclockers.com/nyhet/17...force-gtx-titan-siktar-istallet-pa-700-serien
GTX780 like performance appears to be reasonable, as long as AMD can keep the die size below 450, have a launch price at 550 and make good use of the 4Gb gddr5 chips then they should be onto a winner.
The difference between 450mm2 and 500mm2 is only ~10 die per wafer. Beating Titan at $600 would be better.
Here is why they'll beat Titan (imo).
28nm should be as good as it'll ever be, there are few excuses for not going with a big die.
They've had over 18 months to figure out the flaws in GCN/Tahiti.
They've known or had a good estimate of Titan's performance for a long time. It's nothing special.
Expecting a proper turbo that gives them the free 5%-10% Nvidia gets in reviews.
Given their position at the high end they need to start winning instead of being second with much smaller dies.
If they don't try this time, they'll probably never try again.
Here is why they'll beat Titan (imo).
28nm should be as good as it'll ever be, there are few excuses for not going with a big die.
They've had over 18 months to figure out the flaws in GCN/Tahiti.
They've known or had a good estimate of Titan's performance for a long time. It's nothing special.
Expecting a proper turbo that gives them the free 5%-10% Nvidia gets in reviews.
Given their position at the high end they need to start winning instead of being second with much smaller dies.
If they don't try this time, they'll probably never try again.
boxleitnerb
Regular
The turbo isn't free, jimbo75. It comes at the expense of increase power and noise 
The turbo isn't free, jimbo75. It comes at the expense of increase power and noise
That depends on when the power and noise measurements are taken though. If taken immediately after the end of a benchmark suite they'll probably look quite good, or of course the driver can detect Furmark or something else and throttle down. I dislike the whole thing tbh but AMD has no alternative but to follow suit.
boxleitnerb
Regular
Computerbase measured the default state and all sliders@max. The difference was about 70W, so the Titan used about 265W as compared to around 200. In the German/French reviews that test with pre-heating and/or fixed boost clocks, Titan uses significantly less power than the 7970 GHz. In all other reviews it uses about 20W more. That's your difference right there
Yeah I'm not disagreeing but some other sites might not take so much care, or some might just be downright devious about it.
Anyway, I doubt AMD will be caring too much about power numbers this time around. If they go top on performance now they'll be there for well over a year.
Anyway, I doubt AMD will be caring too much about power numbers this time around. If they go top on performance now they'll be there for well over a year.
What makes you think so? Assuming a normal refresh cycle, Maxwell should be released some time around late Q1/early Q2 2014, likely on 20nm.
Boosts/Turbo's don't go beyond the specifications (electrical / thermal) of the platform the ASIC is on, they are designed to increase the performance for applications that are not utilizing that specification.The turbo isn't free, jimbo75. It comes at the expense of increase power and noise
boxleitnerb
Regular
I never claimed as such. Obviously you will always stay within the specs, but due to Boost 2.0 at different operating points depending on the current conditions (power, temp). I only meant there is a drawback to the higher benchmark numbers in non-German and non-French reviews, and that is (compared to more "realistic" reviews) increased power draw and noise.
HWinfo64 4.22 released today, new codename appears for Volcanic and Pirates Island.
http://www.xtremesystems.org/forums/showthread.php?286978-HWiNFO32-and-HWiNFO64-v4.22-released
http://www.xtremesystems.org/forums/showthread.php?286978-HWiNFO32-and-HWiNFO64-v4.22-released
- Added AMD Volcanic Islands/Crystal: Maui, Iceland/Topaz, Tonga/Amethyst.
- Added AMD Pirate Islands: Bermuda, Fiji, Treasure.
- Added AMD Carrizo, Amur, Bhavani.
Sweclockers claim that Hawaii will be released first on a FirePro, possibly as early as September: http://www.sweclockers.com/nyhet/17...ration-grafikkort-med-firepro-satsar-pa-gpgpu
I suppose this might make sense if AMD's priority is the pro market, or putting the emphasis on HSA with Kaveri following up shortly after. But as professional cards take longer to validate, this suggests the chip taped-out a long time ago, which matches what Charlie reported a while back. It also means we're most likely talking about a 28nm chip.
Could also be complete nonsense, of course.
I suppose this might make sense if AMD's priority is the pro market, or putting the emphasis on HSA with Kaveri following up shortly after. But as professional cards take longer to validate, this suggests the chip taped-out a long time ago, which matches what Charlie reported a while back. It also means we're most likely talking about a 28nm chip.
Could also be complete nonsense, of course.
Last edited by a moderator:
It can make sense.. AMD will need to counter Nvidia before the end of the year on the pro market.
- The first Hawaii cards could come in 28nm, or on 20nm with early process for professional parts ( lower volume, the risk is possible to take and the cards are sold at high price, so yields are not important ). if they are in 28nm there will be a die shrink to 20nm some times after it.. And maybe this is what they will wait for launch the high end consumer product.
But i cant tell where is the true.
- The first Hawaii cards could come in 28nm, or on 20nm with early process for professional parts ( lower volume, the risk is possible to take and the cards are sold at high price, so yields are not important ). if they are in 28nm there will be a die shrink to 20nm some times after it.. And maybe this is what they will wait for launch the high end consumer product.
But i cant tell where is the true.