Can we assume that the 69XX is going to be on par with the 580?

Page 5 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Tempered81

Diamond Member
Jan 29, 2007
6,374
1
81
@the OP:

I would spend money on a fast single gpu. I had sli of mid-range gpus (8800gs 384 vs. gtx260 216). Overall I was happier with larger single gpu, even though the two smaller gpus in SLI were throwing up bigger numbers

460 sli probably puts up bigger numbers than a 480, but I'd probably go with the 480.

If its 350 for those 6850's in CF vs. 500 for a gtx580 or 6970, and you're looking at saving the money then go with the CF setup. The single cards are a bit more hassle free. There will be some games where the two weaker single cards will perform like just that: "a weaker card" despite showing a higher fps. However the beauty of it all is that you won't be able to tell a difference until you have a 580/Cayman sitting there to compare them to; so in the end you should be impressed and have smooth sailing all things considered.

TBH i'd rather have a 470 over 6850CF or a 5870 over 460SLI.

Theres a lot of people here that will argue in favor of fast single gpu vs. CF/SLI of weaker cards to get comparable frames. That number isn't everything, and I found from experience. Theres always some little tiny minute annoyance that will bug the sh#t out of you. Poor scaling, microstutter, lack of framebuffer, driver issues, heat, noise, etc.

There will be a lot of options though, I usually kept myself busier messing with my SLI cards as a hobby than doing actual PC gaming... LOL.

6850CF will not disappoint, however my vote is for 580/6970.
 

Daedalus685

Golden Member
Nov 12, 2009
1,386
1
0
Take a look at how the 6870 compares to the 5870

There are all being built under 40nm.

But you have to compare die sizes too, lets keep in mind the 6870 is much smaller than the 5870. Lets conservatively say that (worst case) the 6900 is exactly the same shader set up as the 6800 (V5 and we do not end up seeing v4).

We have a 255mm^2 6800 compared to a rumored 360mm^2 6900.

41% larger area. If they can only get 20% out of 41% more space something is terribly wrong. I'd expect at least 35% to make the increase is cost worth it. If they do in fact change the shader layout to a more dense V4 we could be looking at super linear die scaling in the 50% plus range.

20% over the 6870 would be well below the lowest guess I'd give. I'd put that at 35% or so for the 'worst case' and 80% faster if they really do pull off something grand.
 

Seero

Golden Member
Nov 4, 2009
1,456
0
0
But you have to compare die sizes too. Lets conservatively say that (worst case) the 6900 is exactly the same shader set up as the 6800 (V5 and we do not end up seeing v4).

We have a 255mm^2 6800 compared to a rumored 360mm^2 6900.

41% larger area. If they can only get 20% out of 41% more space something is terribly wrong. I'd expect at least 35% to make the increase is cost worth it. If they do in fact change the shader layout to a more dense V4 we could be looking at super linear die scaling in the 50% plus range.

20% over the 6870 would be well below the lowest guess I'd give. I'd put that at 35% or so for the 'worst case' and 80% faster if they really do pull off something grand.
That is a rumored die size. If it is really that big, then of course they can fit more SPs in it. However, you can't simply increase the size, full up the space and expect everything to work (Nvidia tried that and failed quite hard.) I'm sure AMD will use a size that is optimal for their product (It isn't a guess game that you and I play here.)
 

Daedalus685

Golden Member
Nov 12, 2009
1,386
1
0
That is a rumored die size. If it is really that big, then of course they can fit more SPs in it. However, you can't simply increase the size, full up the space and expect everything to work (Nvidia tried that and failed quite hard.) I'm sure AMD will use a size that is optimal for their product (It isn't a guess game that you and I play here.)

I know we can't guess at how large it will be.. but even if it is 340mm (identical to the 5800) only 20% improvement over the 6870 with some 35% larger die would be abysmal. But 360mm seems rather accurate, and I'd be shocked if what we get is smaller than cypress. can't see them failing on such a large scale as to release a chip with almost no scaling over the lower end, so a bit shy of the die size increase would be my bet as a low end.

It is certainly possible that the sizes are wrong and the 6970 will shock us all with a die in the ballpark of the 4870.. but I would not assume that to be the case.

The 5770 to 5870 should be in the ballpark of the 6850 to the 6970, with the important point that I use 6850 instead of 6870 as they stated the gap would be smaller this time around.
 

Seero

Golden Member
Nov 4, 2009
1,456
0
0
I know we can't guess at how large it will be.. but even if it is 340mm (identical to the 5800) only 20% improvement over the 6870 with some 35% larger die would be abysmal. But 360mm seems rather accurate, and I'd be shocked if what we get is smaller than cypress. can't see them failing on such a large scale as to release a chip with almost no scaling over the lower end, so a bit shy of the die size increase would be my bet as a low end.

It is certainly possible that the sizes are wrong and the 6970 will shock us all with a die in the ballpark of the 4870.. but I would not assume that to be the case.

The 5770 to 5870 should be in the ballpark of the 6850 to the 6970, with the important point that I use 6850 instead of 6870 as they stated the gap would be smaller this time around.
I personally don't think the die size is the controlling factor on this, so I won't be surprised if the performance doesn't proportionally increased by with size. Having more space in the die may actually decreases electrical noises. However, 41% up in size along with 20-30% increase looks reasonable as long as transistor don't bleed.(up from 20 bro, up from 20!)
 

Arkadrel

Diamond Member
Oct 19, 2010
3,681
2
0
But you have to compare die sizes too, lets keep in mind the 6870 is much smaller than the 5870. Lets conservatively say that (worst case) the 6900 is exactly the same shader set up as the 6800 (V5 and we do not end up seeing v4).

We have a 255mm^2 6800 compared to a rumored 360mm^2 6900.

41% larger area. If they can only get 20% out of 41% more space something is terribly wrong. I'd expect at least 35% to make the increase is cost worth it. If they do in fact change the shader layout to a more dense V4 we could be looking at super linear die scaling in the 50% plus range.

20% over the 6870 would be well below the lowest guess I'd give. I'd put that at 35% or so for the 'worst case' and 80% faster if they really do pull off something grand.


kinda hard to argue against this logic.

Yes youd expect that with 41% bigger size, theyd get like 41% or more performance (if it was linar in growth rate proportional to size increase). HOWEVER its not like that... usually you get more performance than chip size increase.

Why?

With bigger die size, its easier generally to get better performance. Because of certain things at a certrain size need to be included reguarless of how big the die is, so usually the bigger their made, the better they scale (size increase to performance gain).

It wouldnt be that far out there, to assume a 41% bigger chip means more than 41% more performance.

They have supposedly not just used the old architecture, but newer smaller/more effecient, eks the 4d shaders.
 
Last edited:

GaiaHunter

Diamond Member
Jul 13, 2008
3,650
218
106
That is a rumored die size. If it is really that big, then of course they can fit more SPs in it. However, you can't simply increase the size, full up the space and expect everything to work (Nvidia tried that and failed quite hard.) I'm sure AMD will use a size that is optimal for their product (It isn't a guess game that you and I play here.)

The debate so far has been "is just a bit larger than the 5870 or will it be a die size over 400mm^2".

newstrategy.jpg


If you look here you will see it will just be 2mm^2 bigger than a GF104.



If the 6970 ties or is faster than the GTX580 it will be the first time since the X1900XT that AMD is same speed/faster than NVIDIA in the same node.
 
Last edited:

chewietobbacca

Senior member
Jun 10, 2007
291
0
0
nApoleon at chiphell says:

6970 is faster than 580, now stop spreading posts

They also say the final specs appear to be 1920 4D shaders, 900MHz clock... if true, 580 isn't going to live long as the king
 

Will Robinson

Golden Member
Dec 19, 2009
1,408
0
0
nApoleon at chiphell says:

6970 is faster than 580, now stop spreading posts

They also say the final specs appear to be 1920 4D shaders, 900MHz clock... if true, 580 isn't going to live long as the king

Viva la France!:awe:

AMD fans:"Lets' get ready to rumble!"

The Other Guys:"Let's get ready to grumble"

():):biggrin:
 

happy medium

Lifer
Jun 8, 2003
14,387
480
126
nApoleon at chiphell says:

6970 is faster than 580, now stop spreading posts

They also say the final specs appear to be 1920 4D shaders, 900MHz clock... if true, 580 isn't going to live long as the king

Well president Bush said the gtx580 is 10% faster then the 6970.

Who should we believe?
 

ShadowOfMyself

Diamond Member
Jun 22, 2006
4,227
2
0
Why do you refuse to believe it? Just because its an AMD card right?

Isnt chiphell one of those chinese sites who always get their hands on cards way before anyone else? Why would he lie about it then?
 
Feb 19, 2009
10,457
10
76
Explains a lot why NV decides to rush the release.. early bird sets the MSRP. Now they're hoping AMD won't pricewar and price Cayman high.
 

Will Robinson

Golden Member
Dec 19, 2009
1,408
0
0
OMG its so funny watching the mood and tone of each camp as each new rumor hits the street.
A few hours ago it was "Cayman" is doomed,delayed to hell,gonna have to be cheap etc.:$
No wonder VC&G forum smashes all others for viewership...it's such a dynamic situation around new card launches.
GTX580 is a badass card and a damn good effort from NVDA which has been taking it on the chin since the HD48** debut.^_^
I sure hope AMD can do it again and surprise us all with a real,proper,high end chip to compete.:cool:
 

GaiaHunter

Diamond Member
Jul 13, 2008
3,650
218
106
If those 1920 shaders are true that will be indeed at least twice the 6850 shader power even ignoring the 5Dvs4D.

I guess if this is true than it will confirm that Barts does have 1280 shaders on die:

http://www.anandtech.com/show/3987/...renewing-competition-in-the-midrange-market/2

However it’s worth noting that internally AMD was throwing around 2 designs for Barts: a 16 SIMD (1280 SP) 16 ROP design, and a 14 SIMD (1120 SP) 32 ROP design that they ultimately went with. The 14/32 design was faster, but only by 2%. This along with the ease of porting the design from Cypress made it the right choice for AMD, but it also means that Cypress/Barts is not exclusively bound on the shader/texture side or the ROP/raster side.

What Cayman doesn't seem to have is twice the bandwidth of the 6850/6870.
 

tannat

Member
Jun 5, 2010
111
0
0
What Cayman doesn't seem to have is twice the bandwidth of the 6850/6870.

I assume you're thinking of memory bandwidth here?

It looks like it will have similar bandwidht as GTX4-580. But the recent cards have not really been limited by memory bandwidth anyway.
 
Feb 19, 2009
10,457
10
76
If those 1920 shaders are true that will be indeed at least twice the 6850 shader power even ignoring the 5Dvs4D.

I guess if this is true than it will confirm that Barts does have 1280 shaders on die:

http://www.anandtech.com/show/3987/...renewing-competition-in-the-midrange-market/2

What Cayman doesn't seem to have is twice the bandwidth of the 6850/6870.

Nobody but AMD knows the SP count. AIBs have cards and they can run benchmarks, but that's about it. However, because this is a recent Napoleon leak, i suspect he just got out of a pre-launch AIB briefing. :) If its 1920 sp 4D, it's already got 50% shader perf over cypress, 30 vs 20 simds. Addition to uncore tweaks including efficient memory usage and scalable tessellation.. oh, add insane CF scaling, talk about beastly.

I will be extremely surprised if AMD prices it below $499.
 

AtenRa

Lifer
Feb 2, 2009
14,003
3,361
136
If its 1920 sp 4D, it's already got 50% shader perf over cypress, 30 vs 20 simds.

Assuming Cayman do have 1920 shaders and 4-D architecture

Cypress 1600 shaders
Cayman 1920 shaders

that’s 20% difference

One more thing, I will assume that 4-D shaders don’t have the same performance as 5-D shaders, some people say they are at 90% the performance of 5-D. If that’s the case then Cayman needs 10% more shaders in order to have the same performance Cypress/Barts had.

Now if we will like to make a rough guess, I would say that Cayman needs 10% more shaders to get Barts (1120) performance and that translates to 1232 shaders(1120 +10%). Now if Cayman has 1920 4-D shaders it will have the same performance as ~1728 shaders Barts would have. Assuming AMD will keep 256-bit memory, 32 ROPs and clock it at 900MHz, we can say that performance will be at roughly in the +30-35% of Barts (6870).

I could be wrong
 

blastingcap

Diamond Member
Sep 16, 2010
6,654
5
76
Assuming Cayman do have 1920 shaders and 4-D architecture

Cypress 1600 shaders
Cayman 1920 shaders

that’s 20% difference

One more thing, I will assume that 4-D shaders don’t have the same performance as 5-D shaders, some people say they are at 90% the performance of 5-D. If that’s the case then Cayman needs 10% more shaders in order to have the same performance Cypress/Barts had.

Now if we will like to make a rough guess, I would say that Cayman needs 10% more shaders to get Barts (1120) performance and that translates to 1232 shaders(1120 +10%). Now if Cayman has 1920 4-D shaders it will have the same performance as ~1728 shaders Barts would have. Assuming AMD will keep 256-bit memory, 32 ROPs and clock it at 900MHz, we can say that performance will be at roughly in the +30-35% of Barts (6870).

I could be wrong

Your math is a bit off.

1920/4 = 480
1600/5 = 320

480/320 = 1.5

1.5*.90 = 1.35

So even if we assume a 4D cluster is 90% as effective as a 5D cluster, this implies 35% more shader power than Cypress XT.
 

cusideabelincoln

Diamond Member
Aug 3, 2008
3,274
41
91
Assuming Cayman do have 1920 shaders and 4-D architecture

Cypress 1600 shaders
Cayman 1920 shaders

that’s 20% difference

One more thing, I will assume that 4-D shaders don’t have the same performance as 5-D shaders, some people say they are at 90% the performance of 5-D. If that’s the case then Cayman needs 10% more shaders in order to have the same performance Cypress/Barts had.

Now if we will like to make a rough guess, I would say that Cayman needs 10% more shaders to get Barts (1120) performance and that translates to 1232 shaders(1120 +10%). Now if Cayman has 1920 4-D shaders it will have the same performance as ~1728 shaders Barts would have. Assuming AMD will keep 256-bit memory, 32 ROPs and clock it at 900MHz, we can say that performance will be at roughly in the +30-35% of Barts (6870).

I could be wrong

I'm pretty sure you're looking at it the wrong way.

Cypress has 1600 stream processors. 5 stream processors make up a "full" shader (SPU), so Cypress has 320 of them.

Those 320 shaders are our baseline. They work at 100% (relatively speaking).

If we assume Cayman SPUs only work 90% as well as the ones in Cypress, then it would take 355 Cayman SPUs to equal 320 Cypress SPUs.

Cayman has a rumored 1920 stream processors. 4 stream processors is rumored to be an SPU. So Cayman has 480 SPUs.

That makes Cayman have a theoretical, estimated, rumored, speculative 35% more shader performance.


And then we're under the impression Cayman will have more changes than just the shaders. Probably a tweaked memory controller. And also clockspeeds will be different.
 

dangerman1337

Senior member
Sep 16, 2010
333
5
81
Aren't we all forgetting that cayman is likely to partially at least have new architecture in it, so assuming 41% increase in die size = same or less isn't completely true?
 

tannat

Member
Jun 5, 2010
111
0
0
Aren't we all forgetting that cayman is likely to partially at least have new architecture in it, so assuming 41% increase in die size = same or less isn't completely true?

Yes, the above number only takes shader/SP count into consideration.

That's why I in another thread suggested that Barts is propably a better comparision. The architectural improvements in Barts should be towards Cayman while keeping the 4d shaders of Cypress. (Does anyone else than me remember the early info that NI (At that point SI) would have architectural changes of the planned 32nm chip while the 4d shaders would be implemented later?

Now it looks like this is what we are seeing, all in the same generation.
 
Last edited: