3DMark05 and certain websites

[martrox]

My favorite game currently is City of Heroes, which the 6800s seem to do a lot better in.

Me too, but.... the 6800 has real texture shimmering problems in COH, the IQ of the X800 just blows the 6800 away - and I own both cards!

 

[Scali]

Seems I was almost right with 3).
The reality is that NVIDIA extensions are used, but that the alternative is not entirely equivalent in quality, in order to keep the performance hit low. It uses fixed weights for the 4 samples rather than a bilinear filter like NVIDIA uses.
So both quality and performance on ATi is less. Mostly the quality though, comparing between the 9600Pro and 6800LE I have here, it seems that the 6800LE has much softer shadows in some cases. ATi is quite grainy.

 

[madshi]

So both quality and performance on ATi is less.

Why do you say performance on ATI is less? The benches seem to indicate otherwise!

 

[digitalwanderer]

So both quality and performance on ATi is less.

Why do you say performance on ATI is less? The benches seem to indicate otherwise!

How do you say the quality is less? :|

 

[Reverend]

So both quality and performance on ATi is less.

Why do you say performance on ATI is less? The benches seem to indicate otherwise!

How do you say the quality is less? :|

I haven't been catching up on the 3DMark05 articles on the Net but in such articles, which I presume is the ususal shootouts amoingst vid cards involving ATI and NV cards, is DST disabled?

 

[Hanners]

I haven't been catching up on the 3DMark05 articles on the Net but in such articles, which I presume is the ususal shootouts amoingst vid cards involving ATI and NV cards, is DST disabled?

Apart from a section of Hexus' article, I don't think anyone else disabled it (Not even me, I totally spaced the implications of it and the hardware that supported it ).

 

[PatrickL]

Hardward.fr made tests with DST on and off but mostly to show the performance impact.

 

[digitalwanderer]

What is "DST"?

 

[Cat]

It's explained well in Hexus.net's FM article.

 

[Pete]

Yep, definitely read Rys' article. It's also explained in the whitepaper:

The depth maps DST or R32F are both sampled using Percentage Closest Filtering (PCF). If the
hardware supports DST and hardware accelerated PCF, a single bilinearly filtered sample is taken. The
non-DST rendering path uses four point samples. These two implementations produce a bit different
rendering, which can be seen in close inspection, by magnifying parts of frames with some shadow
artifacts, and comparing these side by side. In theory, the bilinear filtering is of higher quality than point
sampling, but in point sampling, the samples are taken from a larger area, and so in some cases point
sampling can produce a smoother looking rendering.

One could argue that the DST and hardware accelerated PCF implementation vs. the non-DST and point
sampling code paths do not produce comparable performance measurements, since the resulting
rendering shows slight differences. 3DMark05 was designed with the firm belief that those two are indeed
comparable, and in the fact that it is the right way to reflect future 3D game performance. Our study has
proved that over a dozen of the biggest game developers are using DST and hardware PCF for dynamic
shadow rending in their latest or upcoming titles. So if DST and hardware PCF are supported, they should
be used in depth shadow map implementations, because that is what is done also in the latest and future
games. However, if the benchmark user wishes to compare exactly identical rendering performance
across different architectures, DST can be disabled in the benchmark settings, and the dynamic shadows
are then always rendered using R32F depth maps and four point sample PCF.

 

[Reverend]

Yep, definitely read Rys' article. It's also explained in the whitepaper:

The depth maps DST or R32F are both sampled using Percentage Closest Filtering (PCF). If the
hardware supports DST and hardware accelerated PCF, a single bilinearly filtered sample is taken. The
non-DST rendering path uses four point samples. These two implementations produce a bit different
rendering, which can be seen in close inspection, by magnifying parts of frames with some shadow
artifacts, and comparing these side by side. In theory, the bilinear filtering is of higher quality than point
sampling, but in point sampling, the samples are taken from a larger area, and so in some cases point
sampling can produce a smoother looking rendering.

One could argue that the DST and hardware accelerated PCF implementation vs. the non-DST and point
sampling code paths do not produce comparable performance measurements, since the resulting
rendering shows slight differences. 3DMark05 was designed with the firm belief that those two are indeed
comparable, and in the fact that it is the right way to reflect future 3D game performance. Our study has
proved that over a dozen of the biggest game developers are using DST and hardware PCF for dynamic
shadow rending in their latest or upcoming titles. So if DST and hardware PCF are supported, they should
be used in depth shadow map implementations, because that is what is done also in the latest and future
games. However, if the benchmark user wishes to compare exactly identical rendering performance
across different architectures, DST can be disabled in the benchmark settings, and the dynamic shadows
are then always rendered using R32F depth maps and four point sample PCF.

Whitepaper explains PCF, not DST.

[edit]Actually, the whitepaper doesn't even explain what PCF is, just how 3DM05 uses its support.

 

[Scali]

DST stands for Depth/Stencil Texture.
And basically it is exactly that. You can use a depth/stencil surface as a texture. Combine this with render-to-texture, and you have very nice shadowmaps.
So in essence it's just another textureformat, which is very suitable for shadowmaps. But it is not a standard format of D3D.

But the most important thing here is that NV's DST textures allow bilinear filtering, but performed after the depth-test, which gives us the bilinear filtered Percentage Closer Filtering (PCF) for soft shadow edges (basically it calcs the percentage of samples from the shadowmap that are closer to the current lightsource than the surface of the current pixel).
On NV hardware this is free. Other hardware has to emulate the bilinear filter in the shader. Futuremark opted to use a simpler filter instead, basically just the average of 4 samples after the depth test.
This gives lower quality output, but the performance hit is a lot smaller.

You should be able to clearly see a difference in quality on the shadow edges. As I say, on my Radeon 9600Pro they look considerably more grainy than on the 6800LE.

 

[Pete]

You're right, Rev. You can guess PCF's function by its name and FM's description of its use, but no real explanation of either feature in the whitepaper.

 

[digitalwanderer]

Thanks Scali, I appreciate ya taking the time to explain it to me.

Damn it, I just bit the bullet and actually bought a 3dm2k5 license....there goes my rep.

 

[Reverend]

You're right, Rev. You can guess PCF's function by its name and FM's description of its use, but no real explanation of either feature in the whitepaper.

Well, FM can't possibly explain every 3D tech in a 3DMarkXX whitepaper. That should be in a separate document ("3D Dictionary"). A few days ago, after I read the whitepaper, I suggested how it can be improved for better clarity and understanding. Maybe we'll have an updated 3DM05 whitepaper in the future, shrug.

 

[trinibwoy]

You should be able to clearly see a difference in quality on the shadow edges. As I say, on my Radeon 9600Pro they look considerably more grainy than on the 6800LE.

So why is Hexus all in a huff and a puff about image quality being lower with DST enabled?

 

[madshi]

You should be able to clearly see a difference in quality on the shadow edges. As I say, on my Radeon 9600Pro they look considerably more grainy than on the 6800LE.

So why is Hexus all in a huff and a puff about image quality being lower with DST enabled?

FutureMark seems to claim that both methods sometimes look better and sometimes worse. Hexus shows a case where ATI looks better. Scali seems to have found situations where NVidia looks better, but I've not seen screenshots of that yet.

 

[ERP]

It looks from the Hexus shots that they take the samples from more disparate locations when DST is disabled.

So probably where the number of samples in the shadow map/pixel is higher NVidia's PCF will look better, and in the cases where the number of shadow Samples/pixels lower, the none DST version will look less bad.

 

[digitalwanderer]

I was looking at both and I honestly can't say which looks worse/better in motion, they just look a tad different. :?

 

[MasterBaiter]

Damn it, I just bit the bullet and actually bought a 3dm2k5 license....there goes my rep.

Hehe, self-0wnage.

I don't think you need to worry about your rep though

I'm still waiting in line to get the little bugger... :?