R520 has 16pipelines

[Gamingphreek]

Originally posted by: crazySOB297
I'm honestly hoping they're carying a little of the R500 tech over, maybe a lil bit of integrated memory for free AA?

I assume you are referring to equipping it with some cache (or are you referring to texture memory)

-Kevin

 

[monster64]

OK shader performance: even if the r520 has 6 vertex shaders, 6x700= 4200, while with the 7800gtx 430x8= 3440. Still, the r520 wins.

 

[kylebisme]

Originally posted by: xtknight
We don't know if unified shaders are going to affect performance for the better or worse either, but lets hope the former.

But we do know that the r520 doesn't use a unified shader design like the r500, so your point is moot.

 

[ZobarStyl]

Originally posted by: monster64
You people don't get it. The r520 can actually have a BETTER fillrate than the 7800gtx. Think:
16pp x lets say 700mhz will be 11200. The gtx has 24pp x 430=10320. Which has the better fillrate? As for memory, pp don't matter, and ATI's memory will be faster, so naturally it will be better. I'm don't really take sides when it comes to video cards (I've owned just about as many AIT cards as Nvidia ones, both have been great.) Just think about it. Everyone whose saying "OMG 16 pp sucks" obviously don't know how fillrate works.

I don't know where you geniuses get off always thinking that one company or another has better memory, when they all get it from the same companies. Unless ATi pays through the nose(and passes it on to the consumer) for better memory, it'll be the same stuff. Make all the guesses about cores, but you aren't going to see a big change in memory.
Plus, fillrate doesn't mean as much as you guys would like to make it out to be with your number crunching. Architectural differences can mean a lot...for instance an x800XTPE has a 35% higher core clock than a 6800U with the same number of pipes, and while it's definitely faster than an Ultra, it's not 35% faster, more like 10-15%. Without some serious redesign changes on the arch, the R520 is going to need one hell of a speed advantage to get far. And if that's the case, there's little to refresh afterwards; whereas nV can drop to 90nm or add a dual slot cooler to the GTX, the R520 has no more room to scale in speed if they are already milking it for every mhz it's got. Tack on the 3-4 month advantage nV will have on the launch of R520, and I worry for ATi. They need to start staying with the game if they intend to pull a profit on anything other than XBox360.

 

[EightySix Four]

Originally posted by: Gamingphreek

Originally posted by: crazySOB297
I'm honestly hoping they're carying a little of the R500 tech over, maybe a lil bit of integrated memory for free AA?

I assume you are referring to equipping it with some cache (or are you referring to texture memory)

-Kevin

The xbox 360 GPU has like 6mb's (don't quote me on that, I can't remember the actual size of it) built in memory, I guess it would be graphics cache, but allows for more or less free anti aliasing, as in AA minus all the performance hits.

 

[xtknight]

Originally posted by: TheSnowman

Originally posted by: xtknight
We don't know if unified shaders are going to affect performance for the better or worse either, but lets hope the former.

But we do know that the r520 doesn't use a unified shader design like the r500, so your point is moot.

No we don't know.

 

[kylebisme]

Originally posted by: ZobarStyl

Originally posted by: monster64
You people don't get it. The r520 can actually have a BETTER fillrate than the 7800gtx. Think:
16pp x lets say 700mhz will be 11200. The gtx has 24pp x 430=10320. Which has the better fillrate? As for memory, pp don't matter, and ATI's memory will be faster, so naturally it will be better. I'm don't really take sides when it comes to video cards (I've owned just about as many AIT cards as Nvidia ones, both have been great.) Just think about it. Everyone whose saying "OMG 16 pp sucks" obviously don't know how fillrate works.

I don't know where you geniuses get off always thinking that one company or another has better memory, when they all get it from the same companies. Unless ATi pays through the nose(and passes it on to the consumer) for better memory, it'll be the same stuff. Make all the guesses about cores, but you aren't going to see a big change in memory.
Plus, fillrate doesn't mean as much as you guys would like to make it out to be with your number crunching. Architectural differences can mean a lot...for instance an x800XTPE has a 35% higher core clock than a 6800U with the same number of pipes, and while it's definitely faster than an Ultra, it's not 35% faster, more like 10-15%. Without some serious redesign changes on the arch, the R520 is going to need one hell of a speed advantage to get far. And if that's the case, there's little to refresh afterwards; whereas nV can drop to 90nm or add a dual slot cooler to the GTX, the R520 has no more room to scale in speed if they are already milking it for every mhz it's got. Tack on the 3-4 month advantage nV will have on the launch of R520, and I worry for ATi. They need to start staying with the game if they intend to pull a profit on anything other than XBox360.

They are going to have better memory simply because there is better memory available now. Just like the fx5800 had better memory than the 9700pros. It doesn't take a genius to figure that out.

As for filtrate, it matters but you are right that those theoretical maximum and it comes down to how efficient the design is. Regardless, for all we know ATI has come up with such an efficient design that even with 8 less pipes they could top the 7800gtx at the same clockspeed. I highly doubt that is the case, but I simply bring up the possibility to prove a point; simply saying "OMGWTF only 16 pipes?!?" is overlooking the fact that number of pipelines is only one of many factors that contribute to overall performance. I don't think anyone here is claiming the r520 will be faster, but rather simply trying to share some logic with the people who are jumping to conclusions.

As for nVidia being able to drop to a lower process and add a bigger cooler, sure but then those aren't the only ways to increase performance. For instance ATI recently licensed this tech:

http://www.intrinsity.com/technology/technology.htm

 

[kylebisme]

Originally posted by: xtknight

Originally posted by: TheSnowman

Originally posted by: xtknight
We don't know if unified shaders are going to affect performance for the better or worse either, but lets hope the former.

But we do know that the r520 doesn't use a unified shader design like the r500, so your point is moot.

No we don't know.

Well not if you live in a closet maybe, but the rest of us who have been paying attention know that ATI has made it clear that they will not introduce a unified shader design to desktops until the r600.

 

[shabby]

Moo?

 

[monster64]

I meant faster memory in terms of clockspeed. The r520s will most likely be in the 1400mhz range. Sorry it took so much effort to figure out what I meant.

 

[allies]

Originally posted by: monster64
You people don't get it. The r520 can actually have a BETTER fillrate than the 7800gtx. Think:
16pp x lets say 700mhz will be 11200. The gtx has 24pp x 430=10320. Which has the better fillrate? As for memory, pp don't matter, and ATI's memory will be faster, so naturally it will be better. I'm don't really take sides when it comes to video cards (I've owned just about as many AIT cards as Nvidia ones, both have been great.) Just think about it. Everyone whose saying "OMG 16 pp sucks" obviously don't know how fillrate works.

That is true, but having more pipes and then overclocking has a markedly larger percentage jump than having less pipes and overclocking. I've been seeing a bunch of manufacturers selling OC versions of card that run at 490 which boost the fillrate quite a bit. Also, this allows nVidia the opportunity to capitalize off of an Ultra running at 550 mhz or so... just my thoughts.

 

[imported_Ged]

Originally posted by: monster64
You people don't get it. The r520 can actually have a BETTER fillrate than the 7800gtx. Think:
16pp x lets say 700mhz will be 11200. The gtx has 24pp x 430=10320. Which has the better fillrate? As for memory, pp don't matter, and ATI's memory will be faster, so naturally it will be better. I'm don't really take sides when it comes to video cards (I've owned just about as many AIT cards as Nvidia ones, both have been great.) Just think about it. Everyone whose saying "OMG 16 pp sucks" obviously don't know how fillrate works.

Heh.

ATI X850XT-PE has a better fill rate now.

The 7800GTX has 16 ROPs. So, 24x whatever for fill rate doesn't work.

16x(clock speed) does work though.

The 7800GTX has a peak fill rate of 6880 Mp/s @430Mhz and 7360 Mp/s @460Mhz.

The ATI X850XT-PE has a peak fill rate of 8640 Mp/s @540Mhz.

These comparison of "OMGz! Look at the fill rates differencez!1!" doesn't work becaue there is a LOT more involved here.

We all know how the 7800GTX performs compared to the ATI X850XT-PE. Observe.

 

[Polish3d]

Originally posted by: monster64
You people don't get it. The r520 can actually have a BETTER fillrate than the 7800gtx. Think:
16pp x lets say 700mhz will be 11200. The gtx has 24pp x 430=10320. Which has the better fillrate? As for memory, pp don't matter, and ATI's memory will be faster, so naturally it will be better. I'm don't really take sides when it comes to video cards (I've owned just about as many AIT cards as Nvidia ones, both have been great.) Just think about it. Everyone whose saying "OMG 16 pp sucks" obviously don't know how fillrate works.

Remember that 430 is just one of the 7800gtx's clocks. It likely clocks up to 500mhz (12000) when running at full speed.

Which incidentally is what is reported as my GPU's clock speed in benchmarking programs like PC Mark 05

 

[Biatche1488]

Better Memory new memory controller new architecture certainly the same fillrate R520 will be better anyway.
Ati spend so much time on this new achitecture so i think they dont failed.They made it to prepare the next card that will have a unified shader.

 

[imported_Ged]

Originally posted by: Frackal

Originally posted by: monster64
You people don't get it. The r520 can actually have a BETTER fillrate than the 7800gtx. Think:
16pp x lets say 700mhz will be 11200. The gtx has 24pp x 430=10320. Which has the better fillrate? As for memory, pp don't matter, and ATI's memory will be faster, so naturally it will be better. I'm don't really take sides when it comes to video cards (I've owned just about as many AIT cards as Nvidia ones, both have been great.) Just think about it. Everyone whose saying "OMG 16 pp sucks" obviously don't know how fillrate works.

Remember that 430 is just one of the 7800gtx's clocks. It likely clocks up to 500mhz (12000) when running at full speed.

Which incidentally is what is reported as my GPU's clock speed in benchmarking programs like PC Mark 05

Even at 500Mhz it's only 8000 Mp/s because the GTX has 16 ROPs.

 

[imported_Ged]

Originally posted by: monster64
OK shader performance: even if the r520 has 6 vertex shaders, 6x700= 4200, while with the 7800gtx 430x8= 3440. Still, the r520 wins.

ATI X850XT-PE has a peak Vertex Rate of 810 MTriangles/s @540Mhz.

810/6 = 135 MTris/s for each Vertex Engine in the X850XT-PE @540 Mhz.

0.25 Tris/clock per Vertex Engine

NVIDIA 7800GTX has a peak Vertex Rate of 920 MTriangles/s @460Mhz and 860 MTris/s @430Mhz.

920/8 = 115 MTris/s for each Vertex Engine in the GTX @460Mhz.

0.25 Tris/clock per Vertex Engine

Our Rumored 6 Vertex shader 700Mhz R520 would do 175 MTris/s for each Vertex Engine at 700Mhz, which is 1050 MTri/s for 6.

Compare the 7800 GTX (@460Mhz) to the X850XT-PE (@540Mhz):

X850 XT PE 52.6 48.0 48.6 48.4 NA NA
7800 GTX 86.3 70.9 71.4 44.5 56.0 56.2

If we scale the X850XT-PE from 540Mhz to 700Mhz, we get:

R520 68.2 62.2 63.0 62.7 NA NA

(Bold for the winners)

Again, we see how the GTX performs compared to the X850XT-PE even though the X850XT-PE has the better peak numbers it doesn't mean that the X850XT-PE is always better.

The Rumored 700Mhz R520 is a big improvement, but this still isn't a good comparison because the R520 could have much higher efficiency than the X850XT-PE.

As for 'Shader Performance' though, I think Kevin was talking about Pixel Shader Performance. Since ATI doesn't have a SM3.0 part right now, we don't have much to go off for PS3.0.

ATI X850XT-PE has 4 quads, 16 fragment pipes, @540Mhz
NVIDIA 7800 GTX has 6 quads, 24 fragment pipes, @460Mhz

ATI X850XT-PE first, NVIDIA 7800GTX OC second (winners bolded):

PS1.1 Procedural 1166.0 720.1
PS1.4 Procedural 460.7 608.1
PS2.0 Procedural 566.7 389.8
PS2.0 1 Light (FP) 513.5 360.3
PS2.0 1 Light (PP) 511.8 402.3
PS2.0 3 Lights (FP) 288.1 208.4
PS2.0 3 Lights (PP) 288.1 234.3
PS2.x 3 Lights (FP) 119.1 72.5
PS2.x 3 Lights (PP) 119.2 96.9

Can see how the X850XT-PE outperforms the NVIDIA 7800 GTX OC here in all but one benchmark.

The NVIDIA GeForce 7800 GTX still outperforms the ATI X850XT-PE in lots of games and outperforms it by larger margins in games that require more shading power. This might be because NVIDIA has better early rejection, because NVIDIA's pipelines are more flexible, or for a number of other reasons. It's very hard to judge any graphics card just by the raw numbers. Sustained rates and real world efficiency are what matters rather than these theoretical peaks that everyone likes to throw around.

For those who want to check out my numbers:
The Numbers I used were from Beyond3D's BFG GeForce GTX OC Review and Beyond3D's X850XT-PE Review.

 

[coldpower27]

Originally posted by: Ged

Originally posted by: monster64
You people don't get it. The r520 can actually have a BETTER fillrate than the 7800gtx. Think:
16pp x lets say 700mhz will be 11200. The gtx has 24pp x 430=10320. Which has the better fillrate? As for memory, pp don't matter, and ATI's memory will be faster, so naturally it will be better. I'm don't really take sides when it comes to video cards (I've owned just about as many AIT cards as Nvidia ones, both have been great.) Just think about it. Everyone whose saying "OMG 16 pp sucks" obviously don't know how fillrate works.

Heh.

ATI X850XT-PE has a better fill rate now.

The 7800GTX has 16 ROPs. So, 24x whatever for fill rate doesn't work.

16x(clock speed) does work though.

The 7800GTX has a peak fill rate of 6880 Mp/s @430Mhz and 7360 Mp/s @460Mhz.

The ATI X850XT-PE has a peak fill rate of 8640 Mp/s @540Mhz.

These comparison of "OMGz! Look at the fill rates differencez!1!" doesn't work becaue there is a LOT more involved here.

We all know how the 7800GTX performs compared to the ATI X850XT-PE. Observe.

Hehe pixel shader fillrate is usually the most important indicator of performance, just as long as your comparing across same architectures. Though these fillrate comparison seems to be useless without knowing the underlying architectueal efficiency of R520.

The G70 has 2 fillrates. 10.32 GPixels from Bilinear Texel Fillrate, derived from the no of pipes it has. 6.88 GPixels for Peak Fillrate derived from the number of Rasterizer Operators.

Which one is more important? From G70 performance vs 6800 Ultra/X850 XT PE, the Bilinear Texel Fillrate is more important in determining performance, what this show is that the amount of pixels a card can output 16 ROP isn't as important as the amount of pixels it can work on with 24 Pipelines.

So far from we have seen though, the trend is greater pixel pipelines = greater performance. I think the FX 5900 vs 9800 Pro created this stigma :frown:

To me these are the leading indicators of performance: (rough)

graphics architecture, no. of pixel pipes, core clock of pixel pipes, memory bandwidth, no. of vertex pipes, core clock of vertex pipes.

Graphics architecture may not be no. one, it might be exchangeable with no of pixel pipes.

I just don't know if I believe the 700MHZ rumors though with the RV530 rumors putting it at 600MHZ only

 

[nRollo]

Originally posted by: bunnyfubbles
Why are people still making a big deal of dual slot coolers? Just about every enthusiast motherboard I know of allows for dual slot video cards without sacficing another slot. The only major downside worth concern that I can see is that its there to cool a very hot chip ala 5800U. Hot isn't so bad, but hot usually means loud. Unless they've gone with a dual slot cooler to reduce noise vs. a similar single slot cooler.

If performance isn't in clear favor of the R520, then that will definately be a big factor against ATI, otherwise this "inefficient design" could be worthy if it is indeed clearly faster than the competition, it would just be a sacrifice you'd have to make. Heat/noise for performance. People would be so against Prescotts/PDs if they were actually faster than the competition.

What what what?!?!?

Oh Bunnyfubbles.

Must I link you to 8,396 threads wherein ATI fans tell me ATI cards are better because they use less power, are quieter, are cooler, are smaller?

Or where they say SLI is bad because it's bigger, uses more power, generates more heat, and has more fan noise.

ATI fans do not make "sacrifices"!

LOL I think I'm going to be a very jolly Rollo if all this proves true. I've never cared about any of this stuff, but it will be hilarious to see all the ATI fans have to eat their past words or diss their new card to aspire to.

Anyway, it will be interesting.

 

[malG]

Originally posted by: Ged

Originally posted by: monster64
OK shader performance: even if the r520 has 6 vertex shaders, 6x700= 4200, while with the 7800gtx 430x8= 3440. Still, the r520 wins.

ATI X850XT-PE has a peak Vertex Rate of 810 MTriangles/s @540Mhz.

810/6 = 135 MTris/s for each Vertex Engine in the X850XT-PE @540 Mhz.

0.25 Tris/clock per Vertex Engine

NVIDIA 7800GTX has a peak Vertex Rate of 920 MTriangles/s @460Mhz and 860 MTris/s @430Mhz.

920/8 = 115 MTris/s for each Vertex Engine in the GTX @460Mhz.

0.25 Tris/clock per Vertex Engine

Our Rumored 6 Vertex shader 700Mhz R520 would do 175 MTris/s for each Vertex Engine at 700Mhz, which is 1050 MTri/s for 6.

Compare the 7800 GTX (@460Mhz) to the X850XT-PE (@540Mhz):

X850 XT PE 52.6 48.0 48.6 48.4 NA NA
7800 GTX 86.3 70.9 71.4 44.5 56.0 56.2

If we scale the X850XT-PE from 540Mhz to 700Mhz, we get:

R520 68.2 62.2 63.0 62.7 NA NA

(Bold for the winners)

Again, we see how the GTX performs compared to the X850XT-PE even though the X850XT-PE has the better peak numbers it doesn't mean that the X850XT-PE is always better.

The Rumored 700Mhz R520 is a big improvement, but this still isn't a good comparison because the R520 could have much higher efficiency than the X850XT-PE.

As for 'Shader Performance' though, I think Kevin was talking about Pixel Shader Performance. Since ATI doesn't have a SM3.0 part right now, we don't have much to go off for PS3.0.

ATI X850XT-PE has 4 quads, 16 fragment pipes, @540Mhz
NVIDIA 7800 GTX has 6 quads, 24 fragment pipes, @460Mhz

ATI X850XT-PE first, NVIDIA 7800GTX OC second (winners bolded):

PS1.1 Procedural 1166.0 720.1
PS1.4 Procedural 460.7 608.1
PS2.0 Procedural 566.7 389.8
PS2.0 1 Light (FP) 513.5 360.3
PS2.0 1 Light (PP) 511.8 402.3
PS2.0 3 Lights (FP) 288.1 208.4
PS2.0 3 Lights (PP) 288.1 234.3
PS2.x 3 Lights (FP) 119.1 72.5
PS2.x 3 Lights (PP) 119.2 96.9

Can see how the X850XT-PE outperforms the NVIDIA 7800 GTX OC here in all but one benchmark.

The NVIDIA GeForce 7800 GTX still outperforms the ATI X850XT-PE in lots of games and outperforms it by larger margins in games that require more shading power. This might be because NVIDIA has better early rejection, because NVIDIA's pipelines are more flexible, or for a number of other reasons. It's very hard to judge any graphics card just by the raw numbers. Sustained rates and real world efficiency are what matters rather than these theoretical peaks that everyone likes to throw around.

For those who want to check out my numbers:
The Numbers I used were from Beyond3D's BFG GeForce GTX OC Review and Beyond3D's X850XT-PE Review.

Nice numbers, thanks for confirming the fact 7800GTX is far more efficient than X850XT-PE in REAL games.

 

[Biatche1488]

THe 7800 ultra will have 24pipe just higher clock and will have dual slot design(Look at the quadro4500 its the pro version of the incoming 7800 ultra).So stop saying that R520 will be soo hot compare to 7800gtx the R520xl will be one slot design like the gtx.Forget my last post(lol saying that ati will be better was dumb).
I'm just wating for bench because now it's REEEEALY unclear.

 

[imported_Ged]

Nice numbers, thanks for confirming the fact 7800GTX is far more efficient than X850XT-PE in REAL games.

No Problem! I'm here all week and whatnot.

 

[Gamingphreek]

Originally posted by: Biatche1488
THe 7800 ultra will have 24pipe just higher clock and will have dual slot design(Look at the quadro4500 its the pro version of the incoming 7800 ultra).So stop saying that R520 will be soo hot compare to 7800gtx the R520xl will be one slot design like the gtx.Forget my last post(lol saying that ati will be better was dumb).
I'm just wating for bench because now it's REEEEALY unclear.

Links?? Also could you clear up your post, no offense, but i cannot really understand it towards the end.

-Kevin

 

[imported_Ged]

Originally posted by: coldpower27


Hehe pixel shader fillrate is usually the most important indicator of performance, just as long as your comparing across same architectures. Though these fillrate comparison seems to be useless without knowing the underlying architectueal efficiency of R520.

The G70 has 2 fillrates. 10.32 GPixels from Bilinear Texel Fillrate, derived from the no of pipes it has. 6.88 GPixels for Peak Fillrate derived from the number of Rasterizer Operators.

Which one is more important? From G70 performance vs 6800 Ultra/X850 XT PE, the Bilinear Texel Fillrate is more important in determining performance, what this show is that the amount of pixels a card can output 16 ROP isn't as important as the amount of pixels it can work on with 24 Pipelines.

Ya, I wasn't sure which "fill rate" they were talking about. More than not people refer to ROP fill rate, so I assumed that. I stand corrected if it wasn't in referrence to the ROPs.

I agree with the bolded statements.

 

[yhelothar]

Pipelines doesn't matter if it has the clockspeed to makeup for the loss in fillrate. As long as the fillrate of the 16pipe card is the same as the 24pipe card, the performance should be similar.
The fillrate of the card is calculated by mutiplying the clockspeed by the number of the pipelines, then by the number of TMUs(for multitexturing).

 

[ddogg]

Originally posted by: Rollo

Originally posted by: bunnyfubbles
Why are people still making a big deal of dual slot coolers? Just about every enthusiast motherboard I know of allows for dual slot video cards without sacficing another slot. The only major downside worth concern that I can see is that its there to cool a very hot chip ala 5800U. Hot isn't so bad, but hot usually means loud. Unless they've gone with a dual slot cooler to reduce noise vs. a similar single slot cooler.

If performance isn't in clear favor of the R520, then that will definately be a big factor against ATI, otherwise this "inefficient design" could be worthy if it is indeed clearly faster than the competition, it would just be a sacrifice you'd have to make. Heat/noise for performance. People would be so against Prescotts/PDs if they were actually faster than the competition.

What what what?!?!?

Oh Bunnyfubbles.

Must I link you to 8,396 threads wherein ATI fans tell me ATI cards are better because they use less power, are quieter, are cooler, are smaller?

Or where they say SLI is bad because it's bigger, uses more power, generates more heat, and has more fan noise.

ATI fans do not make "sacrifices"!

LOL I think I'm going to be a very jolly Rollo if all this proves true. I've never cared about any of this stuff, but it will be hilarious to see all the ATI fans have to eat their past words or diss their new card to aspire to.

Anyway, it will be interesting.

lol.....