Anisotropic Options with 9700

[Joe DeFuria]

I believe I've seen different sites report different things WRT 9700's anisotropic options.

All have said you can enable trilinear with aniso, and all have said ATI have improved their algorithm copared to 8500. Some have said that the base aniso option is the same as 8500's. This is what I would like to know:

What is the actual difference between the "Performance" and "Quality" modes. Some give the impression that Quality is the same as performance, but with forced Trilinear. Some say that "Performance" mode = 8500, and "Quality" mode = improved algorithm. (In that case, is Trilinear forced elsewhere in the drivers, or is it forced by the app?)

 

[Dio]

R9700 can do both bilinear and trilinear aniso fully in hardware.

It has an enhanced sample-selection algorithm over R8500 and no longer suffers from a lack of anisotropy at 45 degrees. This is (I believe) independent of the quality/performance selection.

 

[Doomtrooper]

There is only one guy here that can answer this, and since it was a hot topic how about a screen shot on a 45 degree plane Dave

 

[Joe DeFuria]

Dio,

So what you are saying then is that the difference between Quality and Performance is that Qaulity forces trilinear.

Wavey...can you clarify?

This is what Anand says:

In our technology overview we explained that the R300 core had a slightly different anisotropic filtering engine when compared to the Radeon 8500; the two aren't drastically different, but a few bugs have been fixed. Part of the result of this is a new driver option for performance/quality anisotropic filtering....It's important to note that in most cases (such as the one above), you won't be able to tell any difference between ATI's performance and quality anisotropic filtering settings.

Not clear at all in the technical difference between performance and quality settings. The implication is that performance = 8500, and quality = new anisotropic engine.

Here's what B3D said:

Bear in mind that the quality mode of anisotropic filtering is able to produce trilinear sampling when told, as is the case here.

So B3D appears to say that Performance and Quality have different ansio engines, and that Quality mode CAN support trilinear (when directed by the app or another control panel setting?)

Tom's hardware: Just talks a bit about nVidia's method being "honest" (?), though ATI's method is pratically speaking indistiguishable. However, zero explanation on performance vs. quality modes of Radeon.

Firing Squad: Huh? Don't even know what setting (performance or Quality) they used in their benchmarks...

Extreme Tech: Same as firing squad....

 

[Dio]

So what you are saying then is that the difference between Quality and Performance is that Quality forces trilinear.

I don't think I said anything like that

Not clear at all in the technical difference between performance and quality settings. The implication is that performance = 8500, and quality = new anisotropic engine.

That's not really the right way to view it, there's no 'R8500 mode'.

 

[Joe DeFuria]

OK Dio, then how would you describe (technically) the difference between ATI's performance and Quality modes, and how does Trilinear support factor into the mix?

 

[Dio]

Don't shoot me, I'm just the piano player

What I posted originally is the limited extent of what I know as publicised facts, and therefore is certainly the most I could say even if I knew more, which I don't.

Everybody confused now?

 

[Joe DeFuria]

Heh...yeah, we're all just trying to find out the facts here. While you're at the piano, know any show-tunes?

 

[martrox]

Billy Dio?

 

[KimB]

I saw nowhere in those quotes that the performance mode was akin to the 8500's aniso. If you ask me, this would be entirely counterproductive, as it would require additional hardware for an inferior implementation to be enabled, which would be absolutely silly. The 9700 has enough texture filtering power to do good anisotropic with little performance hit.

The B3D quote looked to me like the performance mode forced bilinear, while the quality mode allowed trilinear. This could easily be verified if, upon the cards' release, we saw some benchmarks at high resolution, without FSAA, with both modes of anisotropic, in bilinear and trilinear each (i.e. four benches...). The more fillrate-limited the game, the better.

If quality + bilinear = performance + bilinear = performance + trilinear, then it should be obvious that bilinear is forced in performance mode (One thing to watch out for is that these benches may be very close in framerate...multiple runs would have to be done to ensure that any differences weren't just due to the variation between tests).

If none of the modes line up, then it should be obvious that different algorithms are used, but benchmarks should easily give at least an idea of how the modes differ.

 

[Joe DeFuria]

I saw nowhere in those quotes that the performance mode was akin to the 8500's aniso.

Well, nowhere in those quotes does it say much of anything. That's why I satarted this thread. However, Anand's quote, paraphrased, reads to me like: "ATI has addressed some of the shortcomings of the R200 implementation of Aniso. As a result, there is now a performance and quality setting." From which it could be implied that the difference between performance and quality settings is the difference between the old and new methods.

If you ask me, this would be entirely counterproductive, as it would require additional hardware for an inferior implementation to be enabled, which would be absolutely silly.

It's not silly if the "performance" mode is significantly faster than the quality mode. That would simply mean that the additional hadware gives us a choice. Based on Anand's tests, there is a significant difference on UT 2003.

The B3D quote looked to me like the performance mode forced bilinear, while the quality mode allowed trilinear.

That may be the case...although the driver control panel has a setting for "application preference" anyway. Why would there be a setting for application preference, AND a "quality" mode then?

 

[andypski]

From which it could be implied that the difference between performance and quality settings is the difference between the old and new methods.

Yup, that could be implied, but the implication is wrong.

The only difference between the quality and performance modes is that the quality mode takes trilinear samples and the performance mode takes bilinear samples. They both use the same calculations in all other respects.

That may be the case...although the driver control panel has a setting for "application preference" anyway. Why would there be a setting for application preference, AND a "quality" mode then?

In application preference mode the application decides all aspects of texture filtering (nearest, bilinear, trilinear, anisotropic ratio, whatever...)
The other anisotropic controls in the panel are disabled and greyed out when in application preference mode.

Performance and quality modes both force anisotropic filtering on at all times regardless of what filtering mode is specified by the application.

Hope this clears it up.

 

[Joe DeFuria]

Yes, it does, thanks!

That does make the most sense: Performance forces bilinear samples, and Quality forces trilinear. It would've been nice if at least one site could have said that.

Now, all that remains to be seen wrt Aniso, is some screenshot comparisons betwen GeForce, Radeon 8500 and Radeon 9700 when viewing textures "on an angle." Then we can get an idea of the actual improvement in 9700's implementation. (Someone already asked this in another thread.)

 

[Dave Baumann]

Performance and quality modes both force anisotropic filtering on at all times regardless of what filtering mode is specified by the application.

Thats where it gets a little confusing though, since if you 'force' Quality the you can still choose bilinear or trilinear withn the app and it would do the appropriate filtering. In essence you may as well always use Quality and select bi/tri withing the app.

I think I did conclude in the review that, given the scores 'Quality' with Bilinear in the app could wee just be the same forcing 'Performance'.

Just to further reiterate, I;ve just had this sent to me:

Quality vs. performance on anisotropic: In both cases, the R300 uses a much improved version of anisotropic filtering, compared to 8500; it should not show Z axis problems. When in quality setting, it enables trilinear, instead of just bilinear for performance (8500 only had bilinear, I believe).

I'll update the review accoringly when I can.

 

[Dave Baumann]

That does make the most sense: Performance forces bilinear samples, and Quality forces trilinear. It would've been nice if at least one site could have said that

Thats the problem, as far as I could see - it doesn't 'force' trilinear, that could still be selectable under SS:SE according to my tests (and I checked it with coloured mip levels).

 

[Doomtrooper]

What about the 45 Degree plane issue Dave ?? Does the 9700 still have problem filtering walls etc on that angle ??

 

[Joe DeFuria]

Thanks for the follow-up Dave....

Perhaps there is just a driver or registry setting bug with the "Quality" setting? It doesn't make sense to me for the Quality setting to allow for bilinear if the app asks for it. It should "force" trilinear IMO.

Did ATI tell you that Quality mode will accpept bi or tri samples, or is that the behavior that you observed in your testing?

EDIT: Or perhaps this behavior has something to do with the "reboot" issue. Most site are reporting that with the current drivers, reboots are sometimes needed to put the aniso / AA settings into effect. Maybe with a reboot trilinear is indeed forced?

 

[Reverend]

Joe, I think andypski explained rather clearly.

App determines bilin or trilin. The 9700's drivers will take either filtering options as specified by the app when/if you enable the "Quality" option. "Performance" will not use trilin even if you select it within the app. "Quality" will use trilin + additional-samples/"hw aniso" if trilin is selected in the app and bilin + additional-samples/"hw aniso" if bilin is selected within the app.

If bilin or trilin is selected within the app, "Quality" still gets you clear mip boundaries. "Performance" will still show mipmap boundaries if bilin is selected within the app but won't if trilin is selected.

That's the way I read it anyway.

 

[Doomtrooper]

Yes the Reboot option is sometimes needed when using FSAA on 3Dmark with the 8500..setting it in the panel then running 3Dmark would not detect FSAA which forced a reboot...don't know about Catalyst drivers anymore as I don't have 3Dmark installed anymore.

 

[Dave Baumann]

Did ATI tell you that Quality mode will accpept bi or tri samples, or is that the behavior that you observed in your testing?

EDIT: Or perhaps this behavior has something to do with the "reboot" issue. Most site are reporting that with the current drivers, reboots are sometimes needed to put the aniso / AA settings into effect. Maybe with a reboot trilinear is indeed forced?

No, my obseravations are from testing withing the game. i.e. (if I remember correctly, I'll double check later) I would enabled 'Quality' in the driver panel, load SS:SE enable to Extreme addon (puts triliear on by default), check that trilinear is on (with colour mip levels), go to the options switch to biliear and check again.