R6XX Performance Problems

[Mintmaster]

It wasn't nearly half the framerate for the games I played back then; I had even written a minor write-up about it and the difference was slightly above the oc gain.

Okay, maybe I'm cherry picking, but I did find at least a couple games with this kind of disparity between computerbase.de and other sites.

There are way too many applications out there where any developer had the funky idea to use for textures X an idiotic negative LOD. You get shimmering there even without optimisations; enable optimisations and all hell can break lose.

Good point.

I don't think that much has changed in terms of AF related optimisations between from NV3x to G7x. IMHLO they merely saved on transistors on NV4x/G7x with the way larger angle-invariance, something that bounced back to NV3x levels with G80 again.

In terms of hardware, sure, but I was talking more about driver settings. There was terrible loss of detail in NV3x cards (e.g. lower mipmap often used) visible in comparison screenshots, and brilinear usage was really over the top with only a small transition region.

I'd rather have 4x good AF samples than 16x crappy AF samples; just as much as folks used to use extensively 2xRGMS instead of 4xOGMS around the GF3 timeframe.

Your AA analogy actually supports my viewpoint more than yours. 2xRGMS is usually equal in quality to 4xOGMS, but is much faster. You sacrifice a bit of quality for a lot of speed.

There's not much perf drop from 4xAF to 16xAF, and I think it would be much less than going from 4x crappy AF to 4x good AF. I doubt you'd have an equal performance comparison there.

If in a case where severe underfiltering is present, if I say that I'm using AA it would build a nice oxymoron since it's limited to polygon edges/intersections. Given the amount of data that polygon interior data captures, proper texture AA is far more important to me than any MSAA.

If severe, I agree (which is why AF is far more important than resolution IMO), but I really think more reviewers would mention the problem if it was ubiquitous. Disabling all optimizations for all games in a review isn't justified to me.

Imagine you could pick in driver panel X between a losless MSAA and a quite lossy MSAA algorithm; the latter ends up quite a bit faster. Now which one would you chose, given that the latter would fail in a healthy persentage to antialias poly edges for instance.

Now would the same apply for reviewers/users? Or else why wasn't stuff like Quincunx or ATI's new custom filters received generally with enthusiasm?

My decision in your scenario obviously depends on what the perf change is and what the artifact density is (along with the artifact behaviour in motion). Going back to your earlier GF3 example, 4xOGMS entailed a huge perf drop compared to 2xRGMS, but didn't improve the worst case and only marginally improved a few edges.

Quincunx and CFAA reduce the quality of everything, IMO.

Well you could have skipped the obvious exaggeration. Shader aliasing and/or crappy content (such as negative LOD on textures) could be catered for by developers themselves; the first needs way more resources and I could find a good excuse for the absence of it. For the latter though I cannot find a single viable excuse.

The reasoning for the latter is same as for everything. Higher performance with an IQ drop that is generally unnoticeable. If I was developing a game, I'd enable trilinear filtering to eliminate the visibility of mip-map boundaries. Brilinear achieves the same thing most of the time with a lower cost.

 

[Ailuros]

In terms of hardware, sure, but I was talking more about driver settings. There was terrible loss of detail in NV3x cards (e.g. lower mipmap often used) visible in comparison screenshots, and brilinear usage was really over the top with only a small transition region.

I doubt much had changed on NV4x/G7x (except of course the angle dependency). At best they might have replaced a few nasty optimisations with others.

Your AA analogy actually supports my viewpoint more than yours. 2xRGMS is usually equal in quality to 4xOGMS, but is much faster. You sacrifice a bit of quality for a lot of speed.

It wasn't a very good example, yet it speaks volumes about the efficiency of each algorithmic implementation. You don't sacrifice just a bit of quality IMHO when you enabled the AF optimisations.

There's not much perf drop from 4xAF to 16xAF, and I think it would be much less than going from 4x crappy AF to 4x good AF. I doubt you'd have an equal performance comparison there.

That's true yet my point is elsewhere. My main priority is a specific standard of IQ and I then adjust performance to that if possible and not the other way around.

If severe, I agree (which is why AF is far more important than resolution IMO), but I really think more reviewers would mention the problem if it was ubiquitous. Disabling all optimizations for all games in a review isn't justified to me.

I recall xbitlabs being another site that had used high quality tests in the past for that reason. Pardon me but the majority of reviewers concentrate sadly enough mostly on performance and way less on image quality.

The cases where negative LODs are being used aren't few and it's the exact spot where any underfiltering sticks so much out that one would have to be blind to not notice anything. Granted you hardly will see any of it in game timedemos; in order to find the problematic spots you'd actually have to play through a game. My first catch of the issue was as early as summer 2004 but I didn't know what was going on back then:

http://www.mitrax.de/?cont=artikel&aid=24&page=14

I afterwards played through more than just a few maps and it was apparent in many spots, albeit it wasn't as noticable as on this ground texture. And it's obviously not as visible as in motion.

My decision in your scenario obviously depends on what the perf change is and what the artifact density is (along with the artifact behaviour in motion). Going back to your earlier GF3 example, 4xOGMS entailed a huge perf drop compared to 2xRGMS, but didn't improve the worst case and only marginally improved a few edges.

A lossy AA algorithm would probably leave a certain degree of polygon edges untouched. If the degree would be in relative terms as high as texture shimmering used to occur with AF optimisations I for one wouldn't use it.

Quincunx and CFAA reduce the quality of everything, IMO.

With a more sensible implementation it might be worth investigating for the future. If you'd take such an idea even further: http://v3.espacenet.com/textdoc?DB=EPODOC&IDX=WO2007049049&F=0
...it might be worth thinking of custom filters. Downside being that I'm not so sure it's as easy to implement as on a TBDR.

My major disagreement with either/or above is that the filters get applied over the entire scene. If applied selectively it would be far more interesting.

The reasoning for the latter is same as for everything. Higher performance with an IQ drop that is generally unnoticeable. If I was developing a game, I'd enable trilinear filtering to eliminate the visibility of mip-map boundaries. Brilinear achieves the same thing most of the time with a lower cost.

It goes for the most part unnoticable when you test your result with just basic bi- or trilinear. Once you enable then AF with an aggressive set of optimisations it will lead to more than just a few nasty side-effects.

Yes it's true for brilinear, yet it also there depends how aggresive the brilinear itself is. IMHO that degree was lowered from NV3x over Nv4x/G7x and should be at it's lowest on G8x/9x. The better your brilinear, the closer it is to full trilinear, which also makes artifacts way less noticable.

 

[ERK]

Quincunx and CFAA reduce the quality of everything, IMO.

Really? Mintmaster=blurring[--------X-]aliasing

 

[ERK]

... the filters get applied over the entire scene. If applied selectively it would be far more interesting.

Yes for Quincunx. For CFAA, that's not the case, AFAIK. Have a link? I thought CFAA was just MSAA but where the samples could be outside the "pixel box." Or do you mean applied over the whole scene in the same way MSAA is applied over the whole scene?

Thanks, sorry to butt in.

ERK

 

[Ailuros]

Yes for Quincunx. For CFAA, that's not the case, AFAIK. Have a link? I thought CFAA was just MSAA but where the samples could be outside the "pixel box." Or do you mean applied over the whole scene in the same way MSAA is applied over the whole scene?

Thanks, sorry to butt in.

ERK

There's a difference between the tent and edge detect filters; the latter costs quite a bit more in performance.