• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

ATI 6000 series, possibly by end of year!

Caporegime
Joined
18 Oct 2002
Posts
33,188
Performance takes transistors, architecture changes account for VERY little performance difference, its all about more transistors, Fermi is HUGE, theres exactly no where at all for them to go in terms of a continuation of Fermi, which would require more shaders, more transistors and more size. It can't make the current one at a profit, one half as big again would basically be unmakeable, remember the current Fermi can't release a single fully enabled 512shader part because yields are so bad, no way in hell they could make a 768shader one.

There will not be a new "fermi" until 28nm, there "could" be a new Nvidia architecture on 40nm, they need something FAR more compact with much more performance per transistor used, and really no one can see that happening.

As for a refresh, where does it say refresh exactly, 6xxx series, not a 5970/5990, its likely to be bigger, but not massively better.
 
Associate
Joined
28 Nov 2009
Posts
881
Location
Manchester
I m going out on a limb and saying this is the gpu that will really break nvidia. All they have to do is scale up the 5000 series, no pressure, its not like they need to respond to nvidia.

If they do manage to make the design more efficient by a significant amount. Green team will be in massive trouble.

They have a lot of room to play with. Add £100 to die cost, 20% to die size, 20C to peak thermals, higher power draw, better cooler, tack on the features nvidia has i.e. 3d, cuda and physx, and what are you left with? 5870 isnt far off a 480, what would a chip that really pushed the limits ati so far have only touched do. And any complaints, they can just point at nvidia and say 'they did it first'.

It boggles me how anyone can still say fermi is a great architecture- if they are going to claim anything they should say ATI were not ambitious enough, had they gone all out and released a 5890, with bigger die, hotter and higher cost...the 480 would have been utterly slain.
 
Last edited:
Associate
Joined
9 Mar 2008
Posts
1,559
I wouldnt be suprised, we went from rumours of a total new AMD arch, to, a hybrid to a sort of refinement. Fermis performance no doubt relaxed the ATi development schedule I would imagen, no sense straining a new chip that isnt nescesary. Though as above, new arch doesnt = second coming of christ, still I think AMD are taking there foot of the gas so to speak.
 
Caporegime
Joined
18 Sep 2009
Posts
30,097
Location
Dormanstown.
I wouldnt be suprised, we went from rumours of a total new AMD arch, to, a hybrid to a sort of refinement. Fermis performance no doubt relaxed the ATi development schedule I would imagen, no sense straining a new chip that isnt nescesary. Though as above, new arch doesnt = second coming of christ, still I think AMD are taking there foot of the gas so to speak.

No, 32nm failing is what changed AMD's plans.
 
Man of Honour
Joined
13 Oct 2006
Posts
90,816
I m going out on a limb and saying this is the gpu that will really break nvidia. All they have to do is scale up the 5000 series, no pressure, its not like they need to respond to nvidia.

If they do manage to make the design more efficient by a significant amount. Green team will be in massive trouble.

They have a lot of room to play with. Add £100 to die cost, 20% to die size, 20C to peak thermals, higher power draw, better cooler, tack on the features nvidia has i.e. 3d, cuda and physx, and what are you left with? 5870 isnt far off a 480, what would a chip that really pushed the limits ati so far have only touched do. And any complaints, they can just point at nvidia and say 'they did it first'.

It boggles me how anyone can still say fermi is a great architecture- if they are going to claim anything they should say ATI were not ambitious enough, had they gone all out and released a 5890, with bigger die, hotter and higher cost...the 480 would have been utterly slain.

GF100 is a great architecture - it was just never designed for 40nm and too much got butchered shoe horning it in. They can actually double the performance in many key areas without a huge increase in complexity or die size but unfortunatly thats about as possible on 40nm as doubling the number of SPs would be.

Theres a reason why they said this when it comes to the ATI architecture also:

thinq.co.uk said:
Rather than offering a revolutionary new GPU architecture, the site describes the new line of GPUs as a "refresh" of the Radeon HD 5000-series GPUs, and says that the new GPUs will focus more on efficiency improvements than cramming in more stream processors.
 
Soldato
Joined
7 Nov 2007
Posts
6,814
Location
Required
GF100 is a great architecture - it was just never designed for 40nm and too much got butchered shoe horning it in. They can actually double the performance in many key areas without a huge increase in complexity or die size but unfortunatly thats about as possible on 40nm as doubling the number of SPs would be.

Theres a reason why they said this when it comes to the ATI architecture also:

Do you know this for fact and can provide evidence or are you working on assumption?
 
Caporegime
Joined
18 Oct 2002
Posts
33,188
GF100 is a great architecture - it was just never designed for 40nm and too much got butchered shoe horning it in. They can actually double the performance in many key areas without a huge increase in complexity or die size but unfortunatly thats about as possible on 40nm as doubling the number of SPs would be.

Theres a reason why they said this when it comes to the ATI architecture also:

Excuses excuses, the fact is, everyone else in the world has adapted the manufacturing available, making large monolithic designs is EASY, using a similar/basic/incredibly easy shader that takes up a huge amount of space is easy. Nvidia's problem for over 3 years has been failure to design for the production available to them, Intel can design for the production they can make, Nvidia aren't a manufacturer, they have no control on the process they use, they need to design for the processes they will use, they don't and thats why Fermi is a flop.

Also no GF100 is a horrible design, we're in an age of high leakage very small transistor where the leakage will only get worse, its a design that would only work at incredibly high speeds that WERE NEVER available to it. Every generation for 3 years their monolithic core has been proving to be less and less effective due to manufacturing. GF100 is a truly awful design, a humoungous part of a succesful design is the ability to actually be able to make it, therefore, Fermi failed, because not a single 512shader part is out in the wild.

GF100 is a great architecture in the same way the architecture I made was brilliant, its basically a GF100 but scaled up with 4000 shaders at 2Ghz, fantastic architecture, blows the competition away....... I just can't make it, no where can?

I think what Rroff means is the design is a nice idea, of course old basic easy shaders run at breakneck speeds isn't a good design, its an OLD design.

AMD's design isn't perfect, but space/transistor to performance efficiency is out of this world and obliterates Nvidia, and when we're talking about manufacturability being half the design battle, Fermi is a huge failure and AMD's is a huge win.
 
Man of Honour
Joined
13 Oct 2006
Posts
90,816
As designs go its not bad, just not very well suited... and theres a real chance of seeing it on a suitable process so personally I'm holding off calling it bad til I've seen how well or not it works on its "home territory". Despite the core being same old same old nVidia theres some nice features too which have a lot of potential.
 
Associate
Joined
23 May 2010
Posts
25
I have heard rumblings of a minor architecture tweak in g104 460 could make a mid-range card people aren't scolded as fanboys for saying they own. It is apparently not the much rumored fermi 2 arch more like fermi 1.5. Usually card makers bend over-backwards to pretend their mid and low range cards have some architecture from their high-end. That it is now opposite for Nvidia does not bode well.

As for ATI 6k series ill quote here what I put in another thread.
Here is how I see the future of high end.

I had strongly debated waiting for the "southern Islands" 6k series of cards which comes out late this fall. With no price pressure from Nvidia I expect 6k prices will be high probably 350$- 6850 and 450$- for 6870. Current gen probably drop to 5850-200$ and 5870-250$'ish.

6k series are hybrid of 5800 and the future architecture 7800 but still at 40nm. For the 6k cards I expect at most 30-40% performance increase over 58xx but with plus 20% power. They are meant as a stopgap because of the cancellation of TSMC and GF 32nm process.

Global foundries 28nm is doing well and Northern Islands might be another September launch in 2011. I expect low and mid range cards at 28nm in early-mid 2011.

My plan is to use my cheap 5870 to skip 6k series and hold on until the true next gen 28nm architecture 7k series comes in late 2011. If 7k gets delayed I will probably add a second 5870 for then about 150$ and hold on through the extra fall-winter-spring months when crossfire is doable for me.

In short If you care about noise and power like I do then Nvidia is in a ditch for awhile like Intel was with "press-hott" 65nm P4. The 6k wont blow away the 5800 on price or performance and the next die shrink and new arch is almost a year and half away. So unless you are a super serious gamer a good deal on a 58xx series will hold you fine for sometime.

-that no 6k cards were seen at computex means either a very late fall launch for 6k or that ATI is locked down regarding 6k and really trying to leave Nvidia in the dark in an attempt to catch them off-guard with launch numbers.
 
Last edited:
Soldato
Joined
28 May 2007
Posts
18,199
wonder how much improvement these will have over the current top-crop of 5000 series, performance wise.

ATI will be looking to put a clear gap between them and the three Fermi cards while increasing on the yeilds over the 5000 cards.

I think the Southern Islands hybrid core might be a lot better than people are expecting it to be.
 
Man of Honour
Joined
13 Oct 2006
Posts
90,816
ATI will be looking to put a clear gap between them and the three Fermi cards while increasing on the yeilds over the 5000 cards.

I think the Southern Islands hybrid core might be a lot better than people are expecting it to be.

The only thing I would expect to be significantly faster on the SI hybrids is tessellation performance. Shader wise they are likely to be just slightly better optimized, higher clocked but otherwise similiar to the 5800 series.
 
Soldato
Joined
28 May 2007
Posts
18,199
Southern Island will be a bigger jump then what we seen from the 4870 - 4890 move and that was quite a decent performance increase.

Tessellation performance will increase by default.
 
Soldato
OP
Joined
18 May 2010
Posts
22,297
Location
London
Personally Im hoping they improve the tesselation performance, of the new cards.

Thats why Im holding off purchasing a 5870.... becuase, compared to Nvidia, its tesselation can be improved.
 
Back
Top Bottom