Will R580 have 48 shader 'pipes?'

[Supreet Virdi]

Ok, so did I got it right now?, I created a picture..see it please, and tell me.
http://img482.imageshack.us/img482/8767/5809ao.jpg

Its just rough picture, don't laugh...lol.

As far as 7800GTX, its 16:1:1.5:1

 

[Ailuros]

Don't forget that the Xenos also lacks a geometry processor, and has no framebuffer compression logic.

And all of that would add another 120M transistors or so?

 

[wireframe]

As far as 7800GTX, its 16:1:1.5:1

I think the 7800 GTX in ATi nomenclature would be:

16-1.5-3-2

Clarification:
16 ROPs
24 texture units (these go with the shader units and not ROPs)
2 ALUs per Shader Unit with 24 shader units in total (different from ROP count, again)
2 stencil ops per clock.

 

[Ailuros]

But NV? Heckifino. If my sig turns out right, that could throw a monkey in the wrench in a lot of ways.

My signature is betta dan yours

 

[KimB]

I think the 7800 GTX in ATi nomenclature would be:

I don't think you can use the same nomenclature on them. If you're going to bother limiting their descriptions to their respective number of pipelines, texture units, and execution units, it's always going to be misleading unless you also talk about the architectural differences.

And even then, there are going to be too many factors to patently say which is going to have the better performance, so you're better off deferring to real-game benchmarks.

 

[wireframe]

I don't think you can use the same nomenclature on them. If you're going to bother limiting their descriptions to their respective number of pipelines, texture units, and execution units, it's always going to be misleading unless you also talk about the architectural differences.

And even then, there are going to be too many factors to patently say which is going to have the better performance, so you're better off deferring to real-game benchmarks.

I was simply assuming that this was what the OP was doing (Supreet) and making some suggestions. Perhaps I read it wrong, but otherwise I think maybe your comment is best aimed at him (unless you want to correct my count).

 

[Geo]

My signature is betta dan yours

I definitely like yours. Tho I would suggest replacing "easier" with "safer".

 

[Supreet Virdi]

I think the 7800 GTX in ATi nomenclature would be:

16-1.5-3-2

Clarification:
16 ROPs
24 texture units (these go with the shader units and not ROPs)
2 ALUs per Shader Unit with 24 shader units in total (different from ROP count, again)
2 stencil ops per clock.

Yep..figured it out now..thanks

 

[no-X]

a bit OT: This question is answered differently in many reviews: RV530 and ROPs. Does it have 4 ROPs with doubled Z/Stencil ops. (like e.g. NV43) or 8 "classical" ROPs (like RV410)? Practical results would be probably very similar (doubled Z ops. to text. ops), but only one version can be right...

 

[Dave Baumann]

4 ROPS, double Z/Stencil.

 

[Rys]

4 ROPS, double Z/Stencil.

Yes, 3 quads of pixels, 4 ROPS, twice the Z rate. Not bad for a drunk man eh?

 

[Mintmaster]

So after all this dribble talk of pipes and ALUs, how much faster will the R580 be compared to the R520?

Let's assume a clock speed of 650Mhz core/800Mhz mem for the R580 just for the sake of argument..and leave the R520(X1800XT) at it's stock 625Mhz/750Mhz.

If I was a betting man (and I am), I would guess the average performance of the R580 to be about 150% that of the R520(X1800XT). Anyone have other guesses on this?

I think your guess will be pretty accurate in some cases.

Before the X1K series was released, I thought putting only 4 texture units into RV530 would be a disaster. But I am amazed at how well it performs compared to the 6600GT and X800, which have twice the texture units. I guess pixel processing is a lot more math intensive than it used to be. I suppose its bandwidth advantage over the 6600GT helps too, but with only 4 ROPs I wouldn't expect that to be an issue with either.

What's amazing is that the X1600 really does achieve a huge pixel shader speed increase over the X1300Pro (often the full 3x), and the X1800XT performs almost exactly 4 times the speed of the X1300Pro in both theory and practice. This is not like NVidia's second ALU, which doesn't seem to do a whole lot - its only around 30% faster on average than R3xx per pipe per clock.

So if indeed, with respect to pixel shader, R580 is to R520 as RV530 is to RV515, then
It'll be hard to judge the game impact without seeing results from a X1600 Pro, but it looks very good indeed.

Now for release date - if Anand is right, we're talking about the middle of next year? Who knows what NVidia is cooking...

 

[wireframe]

So if indeed, with respect to pixel shader, R580 is to R520 as RV530 is to RV515, then
It'll be hard to judge the game impact without seeing results from a X1600 Pro, but it looks very good indeed.

I turned the crank on the ol' Excel with some numbers and came up with "roughly double the shading performance in real world applications." I would say between 2 and 2.5 times the shading performance to be safe.

This was done by using lower end models to project performance using current data sets. It's possible that current drivers and applicatons make poor use of the single shading processor and even worse the triple, so who knows. I came up with something like 2.5x the performance in specific shading benchmarks and rounded that down to 2 for "the real world mess." I think twice the power (per clock) sounds like a reasonable ballpark figure, but it may well prove to be the full three once applications are demanding in a fashion that suits such an architecture.

What's amazing is that the X1600 really does achieve a huge pixel shader speed increase over the X1300Pro (often the full 3x)...

I don't think I saw that happening anywhere. I think the best numbers I used showed an improvement in the 2.8 region, but most were in the 2.5 area.

 

[KimB]

What kinds of benchmarks were you considering, wireframe?

 

[Geo]

Now for release date - if Anand is right, we're talking about the middle of next year? Who knows what NVidia is cooking...

Not quite sure how you got that. He suggested not earlier than Cebit, which is the second week of March.

Here's a little sumtin' sumtin' that supports that:

http://www.rage3d.com/board/showpost.php?p=1333975090&postcount=1

Don't know this fellow, but it sniffs right to me on a couple different levels --tho I can't foreclose it is some guy who is just up on the current thinking making plausible lies.

 

[Mintmaster]

Not quite sure how you got that. He suggested not earlier than Cebit, which is the second week of March.

Whoops, I thought it meant don't expect to hear anything between now and Cebit. I figured that meant it would be a while after Cebit.

Regarding that new info you posted, I hope he isn't correct about R580 not being much different from R520.