• This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn more.

ATi 5.6 overview

#2
Finally...a release that focuses on performance. Not that bugfixes are bad, but WAAAAY back in the day (2000), driver updates were all about performance gains. Nice to see some of that back.
 

mlakrid

OSNN BASSMASTER
Political User
#4
I was wondering the same thing for my powercolor 9100

Last update listed for it was nov 2003....

Ugggh and it wont run World of Warcraft without locking up... and my friend is using a TNT2 64MB version and it can run his.... what gives...
 

Sazar

F@H - Is it in you?
Staff member
Political User
#7
Noticeable improvements will always be primarily limited to the newer tech because by now we have squeezed most of the performance out of the older cards.

The r3xx might see more improvements because dx9 shader speed increases will still affect it.
 

Son Goku

No lover of dogma
#8
Well, my gut instinct would be in concurance with what saz is saying here... The cards, after all have been out for awhile... There is one note worth mentioning however from this article:

ATI did this by using some Z optimisations, removing unseen pixels before running through the pixel shaders. I guess that this is an allowed method and Halo might be running even fifteen per cent faster.
Depending on the number of unseen pixels that get removed, there is a possibility for some performance increase. Anyone familiar with the differences between a tiler (though the card obviously won't become one) and the traditional gfx architecture will know what I'm getting at here. The Prophet 4500 was able to get some performance increase (being a tiler) but not rendering unseen pixels...

But what exactly they will remove, and how this will impact performance. I guess time will tell, as no doubt after the drivers are out (and with these claims), various review sites will start benching them...
 
#9
Son Goku said:
But what exactly they will remove, and how this will impact performance....
Uhm, hate to be a smart ass but they will remove unseen pixels. You will get a speed increase because the shader engine then doenst waste time processing them. :p


Kinda obvious from the article really :D
 

Son Goku

No lover of dogma
#10
LordOfLA said:
Uhm, hate to be a smart ass but they will remove unseen pixels. You will get a speed increase because the shader engine then doenst waste time processing them. :p


Kinda obvious from the article really :D
Umm, :lol: I already said that in my second paragraph, where I said...

Depending on the number of unseen pixels that get removed, there is a possibility for some performance increase. Anyone familiar with the differences between a tiler (though the card obviously won't become one) and the traditional gfx architecture will know what I'm getting at here. The Prophet 4500 was able to get some performance increase (being a tiler) but not rendering unseen pixels...
What I was getting at, in the part of my post that you quoted was, what will get removed (in the context of, to what extent). Yes they'll be removing pixels, but how many will get removed, and to what extent... Depending on how much we see in the way of hidden pixel removal, will depend on just how much of a load (in rendering pixels which aren't seen) is taken off the gfx chip... Perhaps I could have phrased it a little better...

Besides, as someone who formerly owned a Prophet 4500, which was actually based on a tile based architecture, the performance benefits one saw varied from game to game, and this was with a card which was based upon a tile based architecture... Some games (with a lot of unseen pixels) benefited greately, compared to the competition; whereas other games didn't perform quite so well (in comparison) due to the slower chip and as I remember lower clocked memory... This isn't unlikely where the suggested 10-50% performance improvement is comming from. But as ATI's card isn't actually based upon a tile based architecture, I'm suspecting that the degree of hidden pixel removal might be a bit less...

Actually, in general principle I'm expecting the same as saz, as the arch has been around for awhile. But in theory, what they suggested seems plausible (hence my above consideration) and I'm willing to take a wait and see attitude. The suggestion certainly isn't unreasonable (if that's what they're in fact doing)...
 
Last edited:
#11
Heheh I think you're over ananlysing what will happen. THe drivers will remove unseen pixels. THe exact count will depend on the game and where you got the camera pointing.. there is no context or extent to worry about. If you aint gonna see it, the card wont process them. Therefore it wont wste clock cyles on useless calculations so will give you a speed increase.

It may not be a stellar increase but there will be an increase.

Edit: unfortunate typo in "count" :p
 

Son Goku

No lover of dogma
#12
If you aint gonna see it, the card wont process them.
You see, I don't think this is necessarily going to be the case. ATI's card isn't really a tiler, and I don't think these drivers will make it a full fledged tiler. Therefore, no I don't expect it to remove 100% of all unseen pixels from processing, aka the perf benefit could be less then one could expect if it was running like a full fledged tiler...

But to what degree, especially if it removes some of the work load from the GPU, but not necessarily all... Course, time will tell, and when these puppys are out, we can expect many websites to put them through their paces, and get back to everyone with the bench results...

And actually, looking at the article again

ATI did this by using some Z optimisations, removing unseen pixels before running through the pixel shaders.
A tiler like the Kyro 2 (which the 10-50% performance improvement seems close to something I remember from discussions back then) didn't just remove hidden pixels from the shader (actually I don't think they had shaders back on that chip, but I could probably look it up again)...

Edit:

I keep comparing what they're doing (hidden pixel/surface removal) to what was done on tilers such as Power VR's cards, the Kyro 2 (which was the gfx chip found on the Prophet 4500), etc.. Perhaps pulling up an old article (from time past) will help clarify, what exactly was done there, and how what is being suggested with ATI's drivers isn't necessarily as encompassing as what was done there...

BTW, and in all do fairness, there is mention given in the article to 3dfx and nVidia having said

Some competitors (both 3Dfx and Nvidea) of PowerVR keep saying that the technique doesn't work or will run into trouble when scenes get more complex.
Indication was however, that had 3dfx stayed in business (and following Rampage, which is what would have come out following Voodoo5), the card after Rampage would have been using a tile based architecture... They're stocks plummeted, nVidia bought much of their assets, and the rest was history...

http://www.beyond3d.com/articles/tilebasedrendering/index1.php

As can be seen in the pages, with statements such as this:

The last step is obvious, when the tile is completely rendered it can be transferred to the big frame buffer in graphics memory. In this last move dithering can be done (so only 1 time instead of several times !) or you can do anti-aliasing by down filtering this on chip tile using bicubic filters and storing the result in a lower resolution in the big frame buffer. All these actions are part of the pipeline.

How is this different from the traditional architectures?

Traditional architectures, like 3Dfx Voodoo2 - Riva TNT and others, work on a per polygon basis. This means that their pipeline will take a triangle render it, take the following triangle and render it, and take again the following triangle and render it,... this means that they do not know what is still to come. PowerVR uses an overview of the scene to decide what to render, traditional renderers just rush into it and do a lot of unnecessary work. The following figure shows this:
what was done here seems to go beyond what is being suggested for these drivers (including differences on how things are transfered into the frame buffer itself) to achieve the level of hidden surface removal and performance benefits that tilers saw... I'm not really convinced all the benefits to a tile based arch could be achieved without revising the hardware... In fact, meniton of Z optomizations sort of rule it out:

We have all heard that PowerVR doesn't have a Z-Buffer and instead uses a Hidden Surface Algorithm ( performed by the ISP ) . Well it is for that Hidden Surface Algorithm that PowerVR needs to use Infinite Planes. The Hidden Surface Algorithm is based on Ray Tracing Principles...
I'm sure some benefits could be seen if they find a way to prevent some of the hidden pixels from being processed, and though a tiler is the closest thing I can think of (which demonstrates the ability to use hidden surface removal to boost performance), the architectures are fundamentally different.

I do seem to remember some mention of Z-optomizations in the past wrt one hardware company (nVidia perhaps?), for perhaps similar reason... It was so long ago, the specifics aren't fresh in my mind however, so plz don't quote me on this last paragraph...
 
Last edited:

Members online

No members online now.

Latest posts

Latest profile posts

Hello, is there anybody in there? Just nod if you can hear me ...
Xie
What a long strange trip it's been. =)

Forum statistics

Threads
61,961
Messages
673,239
Members
89,014
Latest member
sanoravies