What was that about Cg *Not* favoring Nvidia Hardware?

[Doomtrooper]

That would be a hardware limitation IMO..otherwise it would have been fixed.

Nah. Vertical AA worked for months before a certain driver revision fixed it <g>. Ask around, most 8500 users know this.

I know I'm a 8500 user ..have been since the launch...and no it was never great with verticals, and when it was it was dog slow...since most games only allow 2x peformance anyways and I leave AF @ 16X..

Lets face it the 8500 offers superior IQ in FSAA due its SS and support for Alpha Textures, but its no FSAA card for speed and was one of my only dissappointements with this card...yet it wasn't until the Geforce 4 and 9700 did I consider FSAA a useable option in all games.

 

[KimB]

And Nvidia shouldn't have 'broken' AF + MT with their GF4s, making it (AF) basically unusable in a lot of newer games. Things happen, no product is perfect. But reaching for a user-coded mod of a game that's now 6.5 years old is a bit ridiculous, don't you think?

I never use lower than 8-degree anisotropic on my GeForce4 TI 4200. It's not broken in any way. It's just not as high-performing as it possibly could be.

I've been playing outdoors and haven't noticed any problems with models yet. Play NOLF1 on an Nvidia card lately? The menu has some nasty corruption in it. Try HoMM 4 when it first came out last spring? Crashed to the desktop with the then-latest WHQL Detonators. Ever notice the LOD changing in various Detonator releases? I sure have, and the increased texture aliasing (NWN really showed this). I could go right on nit-picking, but for the most part my GF3/GF4s worked with most games and gave good performance and pretty good IQ. The 9700 of course obliterates them, but that's only to be expected since it is a next-gen. product. It's not perfect, but when viewed fairly and in the large picture it's a damn fine piece of gaming hardware.

No, I haven't played NOLF or HOMM4. In fact, I don't really play that many games, which, to me, makes the problems even more unsettling. And yes, I did notice the texture aliasing in NWN with a couple of textures, but nowhere near the degree I noticed it on the Radeon 9700.

Anyway, yes, if it weren't for me needing to use Linux and nVidia not providing an AGPGART for their nForce chips in Linux, I'd probably leave the Radeon 9700 in my machine. The increased performance in UT2k3 is very noticeable, and the FSAA is fantastic. But that still wouldn't stop me from complaining about the Morrowind issues, the texture aliasing (most noticeable in NWN), and the bilinear/trilinear forcing with anisotropic (which almost makes NWN unplayable).

 

[Jamm0r]

That would be a hardware limitation IMO..otherwise it would have been fixed.

Nah. Vertical AA worked for months before a certain driver revision fixed it <g>. Ask around, most 8500 users know this.

I know I'm a 8500 user ..have been since the launch...and no it was never great with verticals, and when it was it was dog slow...since most games only allow 2x peformance anyways and I leave AF @ 16X..

Lets face it the 8500 offers superior IQ in FSAA due its SS and support for Alpha Textures, but its no FSAA card for speed and was one of my only dissappointements with this card...yet it wasn't until the Geforce 4 and 9700 did I consider FSAA a useable option in all games.

Heh. I ran every single game I had at 4x FSAA years ago on my V5. I know how you feel =) Those games were: European Air War, Falcon4, Grand Prix Legends, Red Baron 3D, Longbow 2, Rainbow 6 / Rogue Spear, System Shock 2, Deus Ex, Ultima 9, etc, etc, etc. When I switched to my 8500 128MB, 4x was out of the question, no matter how good it looked.

 

[John Reynolds]

I never use lower than 8-degree anisotropic on my GeForce4 TI 4200. It's not broken in any way. It's just not as high-performing as it possibly could be.

Well, its performance was certainly a step backwards from its earlier generations, so in that perspective I do consider it somewhat broken. And you use 8x AF and AA? In shooters and games like Morrowind? If so, you have a far better tolerance for poor frame rates than I do.

 

[KimB]

Well, its performance was certainly a step backwards from its earlier generations, so in that perspective I do consider it somewhat broken. And you use 8x AF and AA? In shooters and games like Morrowind? If so, you have a far better tolerance for poor frame rates than I do.

Yes, 8x AF and 2x AA. The only game that has real problems with this is UT2k3, so I run it at 800x600x32 instead of the usual 1024x768x32. Anyway, when playing UT2k3, the Radeon 9700 really shows its stuff, but I won't turn the settings down any further when I'm using the GF4. Can't stand to play anymore without FSAA/Anisotropic.

As a side note, I'm really disappointed with the Radeon 9700's performance in NWN. In particular, its performance with AA enabled is extremely low compared to the GeForce4, compared to how much better it is in other games. I just had to turn it down to 1024x768x32 w/ 2x AA and 16-degree anisotropic (I usually run NWN at the same res/FSAA and 8-degree aniso on the GF4). It does run faster than the GF4 at this setting, but I shouldn't have to turn it down that much. I have a sneeky suspicion that something that NWN uses removes the 9700's ability to compress the framebuffer (and, unfortunately, it looks like that might well be the stencil buffer, since Tenebrae has the same issues, which does not bode well for DOOM3 on the 9700).

 

[Joe DeFuria]

I have a sneeky suspicion that something that NWN uses removes the 9700's ability to compress the framebuffer (and, unfortunately, it looks like that might well be the stencil buffer, since Tenebrae has the same issues, which does not bode well for DOOM3 on the 9700).

Hmmm...let me have a go at this FUD thing...

Since NV30 also uses color compression, it most certainly will have the same horrific issues with NWN and Doom3. In fact, because NV30 enables color compression "full time", Doom3 performance on NV30 is going to be questionable at best, even without AA applied.

 

[Dave Baumann]

The issue is not with color compression, but rather the lack of it.

ATI's DX9 optimisation documentation states there is circumstances where enabling stencil operations will disable HyperZ on R300 (although I doubt that includes the full res early Z reject).

 

[KimB]

Hmmm...let me have a go at this FUD thing...

Since NV30 also uses color compression, it most certainly will have the same horrific issues with NWN and Doom3. In fact, because NV30 enables color compression "full time", Doom3 performance on NV30 is going to be questionable at best, even without AA applied.

Why? Why must color compression necessarily interfere with the stencil buffer? I see no fundamental reason why it must. Hopefully it doesn't.

Anyway, DOOM3 will naturally tend to be a bit more raw fillrate-limited than previous games, so the NV30 already has an advantage over the R300 here.

 

[KimB]

The issue is not with color compression, but rather the lack of it.

ATI's DX9 optimisation documentation states there is circumstances where enabling stencil operations will disable HyperZ on R300 (although I doubt that includes the full res early Z reject).

That's what I was suspecting. Well, if ATI cannot manage to fix this with a driver workaround, and nVidia's GeForce FX can run with full z-buffer and color buffer compression and other bandwidth-saving features enabled, then ATI is in quite a bit of trouble. We all know how much stock so many seem to place on id's engines.

And, depending on the direction that the gaming industry takes, it may bode poorly for the R300 and its derivatives longer down the line, too. Personally I'd like to see global shadowing in future games. If many games use the stencil buffer for that shadowing, ATI's in for trouble.

 

[Doomtrooper]

Anyway, DOOM3 will naturally tend to be a bit more raw fillrate-limited than previous games, so the NV30 already has an advantage over the R300 here.

Nv30 is not competing with a R300 ?? The R300 is almost 6 months old

 

[KimB]

Nv30 is not competing with a R300 ?? The R300 is almost 6 months old

ATI will have similar problems with the R350 if they can't get Hyper-Z to work properly with the stencil buffer.

 

[Doomtrooper]

Who said there is a problem ?? You are speculating again...and again...and again.

 

[Dave Baumann]

Actually, it doesn't appear to affect compression, just Heir-Z:

In addition, few other things interfere with hierarchical culling; these are – outputting depth values from pixel shaders and using stencil fail and stencil depth fail operations.

Early Z reject, AFAIK, is not part of the hierarchical culling either.

[Edit] Further it also states that the stencil buffer should be cleared at the same time as the depth (which makes sense I guess) for Fast Z clear to operate. Perhaps its on or both of these that NWN isn't doing.

As a side note, that document appears to be quite interesting. It states that lower level hierarchies are on chip which should make testing against this very fast.

 

[LeStoffer]

In accordance with what DaveBaumann is talking about, Huddy mentions this on directxdev: http://discuss.microsoft.com/SCRIPT...=directxdev&D=1&F=&S=&P=12046

>The same document states that stencil test harms quick z
>rejection. How much?

Lots! The fast Z rejects (or accepts) which result from the hierarchical Z
buffer can make things run about 4 times faster than you might naively
expect. That's quite a benefit - and you lose that all when you enable the
stencil test.

Thanks,


Richard "7 of 5" Huddy
European Developer Relations Manager, ATI

I understand that Doom III's engine first draw z-buffer in one pass, but xince the z buffer shouldn't change anymore, why would I need stencil test afterwards?

 

[SirPauly]

I don't really play that many games

I figured as much but it's nice to know for sure.

 

[KimB]

I understand that Doom III's engine first draw z-buffer in one pass, but xince the z buffer shouldn't change anymore, why would I need stencil test afterwards?

A stencil test must be done for each shadow hull that is rendered (to calculate what is in shadow and what is not), and the stencil buffer must be read pretty much all the time to see if a light should be rendered to a particular area or not.

 

[KimB]

[Edit] Further it also states that the stencil buffer should be cleared at the same time as the depth (which makes sense I guess) for Fast Z clear to operate. Perhaps its on or both of these that NWN isn't doing.

But if you have multiple light sources, you must clear the stencil buffer multiple times each frame. Otherwise you won't be able to tell which light is shining on which object.

 

[Dave Baumann]

But if you have multiple light sources, you must clear the stencil buffer multiple times each frame.

But you aren't clearing the Z-buffer at this point.

 

[Reverend]

First of all... what a long, drawn-out thread...

That's what I was suspecting. Well, if ATI cannot manage to fix this with a driver workaround, and nVidia's GeForce FX can run with full z-buffer and color buffer compression and other bandwidth-saving features enabled, then ATI is in quite a bit of trouble.

You're mentioning the ifs and the ifs... until we know (and if NVIDIA will let us know) the full details of the GeForceFX in this regard, you're speculating more than you really should be. Furthermore, tell us what the effects (and differences) are if the GeForceFX can do what you said and if the R300 can't do those. No, really!

You know nothing about the GeForceFX's performance in such situations and you know nothing about DOOM3's performance in either hardware's situations. Until you do (and can) while providing approximations with regards to performance and possible IQ issues, it is probably better to keep such basically-useless-rumor-inducing comments to yourself.

 

[KimB]

Yikes! After reading that paper, it really does look like there is a problem here:

In addition, a few other things interfere with hierarchical culling; these are - outputting depth values from pixel shaders and using stencil fail and stencil depth fail operations.

John Carmack's shadow volume algorithm depends on using stencil depth fail operations! This is also most likely the cause of the performance problem that I've been seeing in Tenebrae and NWN. Unless ATI has worked closely with JC to get a different, 9700-optimized algorithm implemented (if that is even possible), this does not bode well for the 9700 and DOOM3.

As a side note, the Radeon 9700 will still trounce the GeForce4 series in DOOM3, but it may have problems doing as well against the GeForce FX as it will do in more traditional rendering scenarios.