So RV770 has 800SP's after all?

ZerazaX

[H]ard|Gawd
Joined
May 18, 2007
Messages
1,582
Just saw this post at Nordic Hardware here

Remember how we told you that Eastern sources claimed that RV770 has 800 shader processors, not just 480 shader processors? When we look at the slide presented to us by a reliable source, we can only conclude that they were right. AMD evidently planted the stories of 480 shader processors (it's not uncommon for AMD/ATI to plant stuff). The slide tells us three things about both Radeon HD 4850 and 4870. First of all that they both sport a whopping 800 shader processors, but also the clock frequencies and they're individual calculation capacity measured in FLOPS. We've already told you that Radeon HD 4850 will launch at 625MHz and that AMD will clock the 4870 at at least 100MHz more, the slide says 750MHz.

rv770slideuj5.jpg

Read the rest at the link

People @ folding should be quite happy... 1TFp for $199!

Lets see how these work though.. wow 250% increase in ALU's (64->160) in that small die size...

Obviously flops dont mean anything to actual performance in game, but the GPU computing person in me is quite happy. If ATI gets on board with PhysX (which Nvidia wants the API to be open for license at the very least), that could be quite the card to have on the side to a GT200 for physics at the least. My folding@home computer will be quite happy...
 

ZerazaX

[H]ard|Gawd
Joined
May 18, 2007
Messages
1,582
I believe so...

Well this means they went from 64 5D shaders to 160 5D shaders... 250% hmm..
 

gs29

Weaksauce
Joined
May 13, 2008
Messages
118
Nvidia: $650 for 930 GFlops.

ATi: ~$200-$230 for 1 TFlops

It's too bad this means little in terms of performance but it's funny nonetheless.
 

ZerazaX

[H]ard|Gawd
Joined
May 18, 2007
Messages
1,582
It makes for funny marketing schemes :p

I will admit though that it must feel like a kick in the balls for Tesla/CUDA if your archrival offers more Flop performance for less than half the price for GPGPU purposes
 

gs29

Weaksauce
Joined
May 13, 2008
Messages
118
You might be thinking of TMUs (of which there are 32 to RV670's 16). The # of ROPs hasn't really mattered in quite a while. 16 is as good as 32. The reason GT200 has 32 is because the # of ROPs is tied to bus width.
 

berzerker

Gawd
Joined
Jun 16, 2004
Messages
820
QFT. Hasn't this been the problem since the 2900 series?
One of the threads had a link that says that they changed the way AA is done in the 48xx series, so it doesn't cause such a big performance hit like the older generation of cards.
 

ZerazaX

[H]ard|Gawd
Joined
May 18, 2007
Messages
1,582

LOL his evidence its a fake is based on how someone higlights a PPT?

For god's sake it could've been a highlight to show how the mainstream card (considering the 4850 is approaching 2600XT prices at launch atm based on rumors) is easily many times more powerful than a quad core at FLOP performance.

After all, PR slides are often accompanied with people *presenting* these slides and certainly aesthetics is only part of what a powerpoint presentation is about...

pxc said:
Fake. The die size only grew by 56mm^2 vs the RV670 on the same process, a size increase of ~25%.

Doesn't matter - die size is not completely made up of ALU, TMU, etc. There's a lot more in there, some of which doesn't scale as well with smaller process size.

Think about how huge R600 was then shrunk to RV670 after cutting down the internal ring bus, memory bus, and adding UVD. It was tiny compared to R600 and in all likelihood the 1024-bit internal/512-bit external bus took far more transistors and area than 512/256. Given that they're using the same memory bus, likely same ROPS (though double clocked from before apparently), it's not out of the question to see 64 * 5D SP's -> 160 * 5D.

And last I heard, R600 architecture ALU's have low transistor cost.


This is the "reliable source" from 3 months ago: http://bbs.chiphell.com/viewthread.p...extra=page=1

There was a huge debate between Eastern and Western sources over SP count. Sources in Asia said 800SP, sources in the west said 480SP.

That hardware infos released that supposed release from AMD on reference card specs and yet its been proven fake since those clocks just aren't true at all.

I posted this elsewhere already:

I'll offer one more piece... and this was posted in this very own thread but dismissed as unlikely it went from 64 * 5D to 160 * 5D...

Now that perkam mentioned it... remember that GPU-Z shot that showed RV770 at 256mm^2 and people were saying it might be fake since how can a database program know the area size unless its programmed within the BIOS? Well we all know that w1z gets a lot of this knowledge before we often do for GPU-Z (and the GPU-Z shots of GTX280 all have info there as well that might not be in the card).

Anyways, on the techpowerup forums almost an entire month ago, w1z posted this:

http://forums.techpowerup.com/showpost.php?p=790539&postcount=36

see the yellow thingies inside the red shader blocks? the big question is how many of these are in the rv770 per red block.

if it's 5 then 800 is correct (160 * 5 = 800)
if it's 3 then 480 is correct (160 * 3 = 480)

He already hinted at 160 with that post since at the time everyone was saying it can only be 96 * 5D for 480... now all of a sudden 160 * 5D is very possible

Note that he didn't suggest 96*5D as the alternative but immediately went to 160 * 3 which would've been a step back in how SP's work, which seems unlikely for AMD
 

pxc

Extremely [H]
Joined
Oct 22, 2000
Messages
33,064
I posted this elsewhere already:

Note that he didn't suggest 96*5D as the alternative but immediately went to 160 * 3 which would've been a step back in how SP's work, which seems unlikely for AMD
Interesting post by W1zzard. It's possible that AMD decreased the ALUs per SP block. That was one of the complaints about the R(V)6xx architecture, not getting enough parallelism from the 5-way superscalar blocks. If he knows the SP block count is 160, he pretty much covered the possibilities:

5 ALUs/SP block: 480 processors = 96 SP blocks (16 x 6, doesn't match the SP block count)
4 ALUs/SP block: 480 processors = 120 SP blocks (unlikely)
3 ALUs/SP block: 480 processors = 160 SP blocks (16 x 10, disco)
5 ALUs/SP block: 800 processors = 160 SP blocks (16 x 10, ditto)


The problem that you forgot to quote though is that it's very unlikely AMD crammed in an extra 480 SPs in only 25% more space that the die size increased (both RV670/RV770 at 55nm process), excluding the other enhancements made to the chip.

In any case, we're around 2 weeks away from official reviews. I'd bet on 3 ALUs/SP block if W1zzard's 160 SP block count is correct.
 

Hypernova

[H]ard|Gawd
Joined
Jan 16, 2007
Messages
2,027
LOL his evidence its a fake is based on how someone higlights a PPT?

I follow that B3D thread too. And I haven't seem much to suggest 800 besides that slide and W1Z's post.

I want to believe, but I'm going to stay pessimistic until the real thing comes.
 

5150Joker

Supreme [H]ardness
Joined
Aug 1, 2005
Messages
4,569
So uh, where's all the games requiring all the power these cards bring to the table? I don't have any trouble playing current games with my 8800 GTX. Seems a lot of the new PC games are also simultaneously released for the 360/PS3 so they're going to be limited in how far they push the PC aside from AA/resolution. PC gaming sucks right now, there's very little incentive to purchase an expensive new card unless you just enjoy measuring your e-penis.
 

cannondale06

[H]F Junkie
Joined
Nov 27, 2007
Messages
16,180
So uh, where's all the games requiring all the power these cards bring to the table? I don't have any trouble playing current games with my 8800 GTX. Seems a lot of the new PC games are also simultaneously released for the 360/PS3 so they're going to be limited in how far they push the PC aside from AA/resolution. PC gaming sucks right now, there's very little incentive to purchase an expensive new card unless you just enjoy measuring your e-penis.
perhaps thats YOUR view on it so dont waste your money then. there are others that would like to play with more aa/af at 1920 and 2560 resolutions.
 

sethk

[H]ard|Gawd
Joined
May 3, 2005
Messages
2,042
A lot of console games are struggling to push out even 720p@30fps. COD4, Halo3 and GTA IV are all examples of console games that couldn't even do 720p. And that's with fixed HW. Even though these games are just ported to the pc and ports are inefficient, games like GoW, DMC4 and COD4 look better on the PC and can be played at higher res with better AA and better AF.

If you have a 8800GTX and can play all your games at 1920x1200 (or 1080) with settings cranked then there are many modern games you haven't been playing.

As someone else said, for your use a 8800GTX may be great, many others will be happy to upgrade.
 

baddog121390

Limp Gawd
Joined
Jun 18, 2006
Messages
277
QFT. Hasn't this been the problem since the 2900 series?

r6xx doesn't take a big hit from AA, it takes a big hit from AF due to its low texturing capabilities. People just see reviewers putting on 4xAA/16xAF and assume the big hit comes from AA, since on most architectures (ie g80) AF is nearly "free".
 

cannondale06

[H]F Junkie
Joined
Nov 27, 2007
Messages
16,180
r6xx doesn't take a big hit from AA, it takes a big hit from AF due to its low texturing capabilities. People just see reviewers putting on 4xAA/16xAF and assume the big hit comes from AA, since on most architectures (ie g80) AF is nearly "free".
sorry but you are very wrong as the R6xx cards plumment with AA on not AF.
 

kllrnohj

Supreme [H]ardness
Joined
Apr 1, 2003
Messages
6,845
So uh, where's all the games requiring all the power these cards bring to the table? I don't have any trouble playing current games with my 8800 GTX. Seems a lot of the new PC games are also simultaneously released for the 360/PS3 so they're going to be limited in how far they push the PC aside from AA/resolution. PC gaming sucks right now, there's very little incentive to purchase an expensive new card unless you just enjoy measuring your e-penis.

There are also those of us who like to skip a generation or two to save money. For example, I bought a 7900GT when it came out even though an 7600GT would have been plenty for my resolution (1280x1024), but the 7900GT is still going strong for this resolution several years later (though it is starting to show its age in a couple of games). So I like to buy a higher end card to avoid upgrading for several years at a time. So if this new info is true, a 4850 is a likely upgrade for me in the near future (as ~$300 is about the most I can spend), which will then hopefully last me 3-4 years.
 

ZerazaX

[H]ard|Gawd
Joined
May 18, 2007
Messages
1,582
sorry but you are very wrong as the R6xx cards plumment with AA on not AF.

Have any benches to show that?

Almost everyone puts AF on when they test AA since AF is one of the first things on, so its hard to tell that AA kills and not AFconsidering both are often on.

And I've often seen game benches where the drop from 4xAA to 8xAA isn't very noticeable

I know 3dMark06 doesn't mean much when comparing between games, but when done on the same rig with the same card between settings, it can certainly give you a good idea of what settings hurt more than others. So hoom at b3d did these tests on the same rig with the same card in one sitting at different AA and AF levels:

Just for kicks & since I'm currently reading a good book, I have just done a quick bunch of 3Dmark06 runs on my 3870 with various AA/AF configs.
E6600 @stock, 3670 @stock, Cat 8.5

no AF/AA 9883
16* AF/8*AA 8098
8* AF/no AA 8552
16* AF/no AA 8164
no AF/4*AA 9456
no AF/8*AA 9449
no AF/24*AA 8903

This was just one run of each config & I have stuff open in the background so not by any means scientific but there seems to be a pretty clear pattern.

That would certainly give credence to the idea that AF is what hurts the most, and since AF is almost always turned on since it affects IQ the most, it certainly would be a good reason why real world tests which almost always have AA on show significant problems for the R600-derivative cards
 

kllrnohj

Supreme [H]ardness
Joined
Apr 1, 2003
Messages
6,845
Have any benches to show that?

Almost everyone puts AF on when they test AA since AF is one of the first things on, so its hard to tell that AA kills and not AFconsidering both are often on.

And I've often seen game benches where the drop from 4xAA to 8xAA isn't very noticeable

I know 3dMark06 doesn't mean much when comparing between games, but when done on the same rig with the same card between settings, it can certainly give you a good idea of what settings hurt more than others. So hoom at b3d did these tests on the same rig with the same card in one sitting at different AA and AF levels:



That would certainly give credence to the idea that AF is what hurts the most, and since AF is almost always turned on since it affects IQ the most, it certainly would be a good reason why real world tests which almost always have AA on show significant problems for the R600-derivative cards

[H]ard has shown that the r6xx cards run just fine with AF on. Indeed, many of their benches have just AF for the r6xx cards. Here: http://hardocp.com/article.html?art=MTM5OSw0LCwxNDA= 1600x1200 "Very High" no AA/16xAF vs. 4xAA/4xAF Despite lowering the AF (which you claim hurts more), the 2900xt's FPS plumets with 4xAA compared to the 8800GTS 640mb. Therefore, AA is the weak spot, not AF.

Or if you'd rather: http://www.hardwarecanucks.com/foru...0-review-crossfire-performance-preview-5.html has many 0xAA/16xAF vs. 4xAA/16xAF graphs, AA kills the r6xx over and over again.
 

HAF72

1337 :)
Joined
Nov 2, 2006
Messages
7,459
darn!! im going to switch out my crossfire mobo for a sli mobo so in the future i can sli now i see this :confused: i really want to get back into folding... ofcourse for the [H] so i may get a 4870 :/ sighs....
 

Simpson5774

2[H]4U
Joined
Feb 1, 2005
Messages
2,937
Isn't the RV670 technically faster in terms of Gflops than G92 as it is (not including MADD+MUL operations since nothing makes use of its benefit yet), and yet being the more efficient design, the G92 is faster at what the board is being sold for, games. Or am I wrong?
 

HAF72

1337 :)
Joined
Nov 2, 2006
Messages
7,459
Isn't the RV670 technically faster in terms of Gflops than G92 as it is (not including MADD+MUL operations since nothing makes use of its benefit yet), and yet being the more efficient design, the G92 is faster at what the board is being sold for, games. Or am I wrong?

yea i think your right where g92 = for gaming while the rv670 = Gflops not really sure though :rolleyes: :p
 

ZerazaX

[H]ard|Gawd
Joined
May 18, 2007
Messages
1,582
[H]ard has shown that the r6xx cards run just fine with AF on. Indeed, many of their benches have just AF for the r6xx cards. Here: http://hardocp.com/article.html?art=MTM5OSw0LCwxNDA= 1600x1200 "Very High" no AA/16xAF vs. 4xAA/4xAF Despite lowering the AF (which you claim hurts more), the 2900xt's FPS plumets with 4xAA compared to the 8800GTS 640mb. Therefore, AA is the weak spot, not AF.

Or if you'd rather: http://www.hardwarecanucks.com/foru...0-review-crossfire-performance-preview-5.html has many 0xAA/16xAF vs. 4xAA/16xAF graphs, AA kills the r6xx over and over again.

Sure it kills it but how much compared to *no* AF at all. AA will always hurt performance, as it does to Nvidia cards as well. Basically are there any reviews where they have tested AA only (no AF) at different levels vs. AF only (at different levels) vs. both on vs. neither on? That's really the only way to test it because having those other variables there (AA) with no base runs (neither) doesn't tell you which causes bigger performance issues.

That's the thing, I don't think we've seen any conclusive tests out there on the web that actually investigates which it is
 

MrWizard6600

Supreme [H]ardness
Joined
Jan 15, 2006
Messages
5,779
I follow that B3D thread too. And I haven't seem much to suggest 800 besides that slide and W1Z's post.

I want to believe, but I'm going to stay pessimistic until the real thing comes.

you know the worst thing about that is that we know kyles got one/will have one very shortly. NDA's annoy the hell out of me.

just like the 2900xt's "free AA" :)

ATI made that statement based on the fact they had the semi-working memory infrastructure of a ground breaking 140GB/s. The problem of course was that any AA call to the ROP was ran back over the cards internal bus to the top (which was not a design choice, it was a "get a working product or your all fired" choice) where it ran through a shader in an extrordinary demonstration of poor efficiency. So, when you tried to cache anything for an AA call, it would get put in memory at warp speed, but then you'd queue for the cards internal bus to take you back up to the top so it could be executed on a general purpose shader. So, yes, the RV770 does use a different AA setup then the RV670 and R600... the RV770 uses the right non-fail way to do AA, on the ROP.

darn!! im going to switch out my crossfire mobo for a sli mobo so in the future i can sli now i see this :confused: i really want to get back into folding... ofcourse for the [H] so i may get a 4870 :/ sighs....

As I just said in another thread, "upgrading to an Nvidia chipset" is an oxymoron. I would wait to see just how 790i plays out. Give it another couple months to let any failures really sink in.
 

evolucion8

Gawd
Joined
Mar 15, 2006
Messages
917
Even though in normal resolutions (1280x1024 up to 1920x1080) the RV6XX architecture plummets when Anti Aliasing is on, in ultra high resolutions, it often outperforms the 8800GT even though both are quite unplayable at that level of eye candy anyways (Above 1920x1080). Is not a bad feat considering that the Anti Aliasing is resolved on the pixel shaders which offers a greater flexibility in image quality at the expense of performance. Just hope that ATi hasn't sacrificed their Anti Aliasing flexibility when they changed the way of how anti aliasing works in the new architecture.
 
Joined
Nov 3, 2006
Messages
973
Even though in normal resolutions (1280x1024 up to 1920x1080) the RV6XX architecture plummets when Anti Aliasing is on, in ultra high resolutions, it often outperforms the 8800GT even though both are quite unplayable at that level of eye candy anyways (Above 1920x1080).

Can you provide some links to back that up?, because I think your talking BS (I seem to remember there are a few games but its rare).

I am happy to eat my words if i turn out to be wrong.
 

pxc

Extremely [H]
Joined
Oct 22, 2000
Messages
33,064
All benchmarks at 1920x1200 4xAA/16xAF:

BF2142 http://www.tomshardware.com/charts/desktop-vga-charts/battlefield-2142,552.html?p=1590,1616,,
8800GT OC 57.20fps
HD 3870 33fps

Dark Messiah of Might and Magic http://www.tomshardware.com/charts/...iah-of-might-and-magic,557.html?p=1590,1616,,
8800GT OC 40.40fps
HD 3870 39.50fps

MS Flight Simulator X http://www.tomshardware.com/charts/...oft-flight-simulator-x,566.html?p=1590,1616,,
8800GT OC 21.90fps
HD 3870 18.60fps

Oblivion http://www.tomshardware.com/charts/...lder-scrolls-4-outdoor,571.html?p=1590,1616,,
8800GT OC 30.60fps
HD 3870 24.20fps

Warhammer http://www.tomshardware.com/charts/...mer-mark-of-chaos-v1-6,578.html?p=1590,1616,,
8800GT OC 36fps
HD 3870 34.3fps

Stupid THG only has the 8800GT OC version, the core 60MHz faster than stock and the memory is 100MHz over stock. The OC doesn't explain away the performance differences because scaling is not perfect.

Hot Hardware has one test where the HD 3870 beats the 8800GT at 1920x1200 4xAA/16xAF: CoH. The 8800GT wins the other 3 tests at that resolution. http://www.hothardware.com/Articles/ATI_Radeon_HD_3870_and_3850_55nm_RV670/?page=7

I can't imagine many other reviews showing much different. The results are not a surprise. edit: let me try: http://www.driverheaven.net/reviews/3870-XXX/COD4.php :p
 

ZerazaX

[H]ard|Gawd
Joined
May 18, 2007
Messages
1,582
Ironically the 4xAA -> 8xAA drop in the 3800's is a lot lot less than the 4xAA to 8xAA drop in the G8x/G9x cards but usually the damage is done by then.
 

evolucion8

Gawd
Joined
Mar 15, 2006
Messages
917
Yeah it is strange, I couldn't say it any better. Probably the anemic Anti Aliasing performance of the R6XX can't get any much worse as the samples increases. Probably that's the only advantage of using pixel shaders to resolve the anti aliasing, less performance drop at higher resolutions, but still too weak anyways, it will have more use in future architectures, RV7XX?? We shall see.

http://www.ngohq.com/home.php?page=articles&go=page10&arc_id=137 <<Check those which uses 8x MSAA

http://www.hardwarecanucks.com/foru...0-review-crossfire-performance-preview-5.html <<Check the 8x AA results

https://techspot.com/review/94-visiontek-radeon-3870-x2/page4.html <<HD 3870X2 Against the 9800GX2

That's only to prove that also the GeForce 8/9 architecture plummets once you crank the Anti Aliasing up to 8x. So now eat your BS, I mean words :)
 

bfellow

[H]ard|Gawd
Joined
Dec 20, 2007
Messages
1,082
Yeah it is strange, I couldn't say it any better. Probably the anemic Anti Aliasing performance of the R6XX can't get any much worse as the samples increases. Probably that's the only advantage of using pixel shaders to resolve the anti aliasing, less performance drop at higher resolutions, but still too weak anyways, it will have more use in future architectures, RV7XX?? We shall see.

http://www.ngohq.com/home.php?page=articles&go=page10&arc_id=137 <<Check those which uses 8x MSAA

http://www.hardwarecanucks.com/foru...0-review-crossfire-performance-preview-5.html <<Check the 8x AA results

https://techspot.com/review/94-visiontek-radeon-3870-x2/page4.html <<HD 3870X2 Against the 9800GX2

That's only to prove that also the GeForce 8/9 architecture plummets once you crank the Anti Aliasing up to 8x. So now eat your BS, I mean words :)

Those 1st 2 makes me cry b/c my 3870 is still weaker than 8800GT
 

evolucion8

Gawd
Joined
Mar 15, 2006
Messages
917
But it is weaker than the 8800GT, but for an average of 15% to up to 20% in the best case scenarios, so don't feel that bad. Even at 8x FSAA, both cards are equally too slow to play many games at those ubber high resolutions.
 
Top