Quake3 HQ: Why isnt this used?

[Typedef Enum]

This is one of those deals that I've always personally wondered about...

We're all familiar with sites that still continue to use Quake3 timedemos, and used to seeing numbers in the 300+ FPS neighborhood...

However, there's one option which, if enabled, will most definitely bring those numbers significantly down, and would be in the area of HQ.

The option I'm referring to is the cg_shadows "2" variable.

Am I out in left field, or is enabling this variable considered wrong when dealing with HQ? I'm actually somebody that still continues to play Quake3 on a regular basis, and recently began playing with this option on, and basically had to drop the resolution back down to 12x10 due to the performance drop when 4xAA and 16xAF are both enabled...

Any thoughts?

 

[Nick[FM]]

It's such a long time since I played Q3A, and now I can't find my CD. Damn. Could you possibly posts shosts with/without the cg_shadows "2" variable. Interesting to see the difference.. Does it work with all cards?

 

[Typedef Enum]

Hmm...

I would, but I cannot recall the address of my webspace folder (just switched ISP's)...

When you enable the stencil buffer in Quake3 and set the shadows to "2", you go from that "blob" shadow to...well, until Doom III arrives, about as good a shadow as you're going to get.

In terms of framerate, it most definitely will drop you down pretty significantly. In fact, this has almost always been one of those options that went by the wayside going all the way back to its initial release simply because the framerate hit was so excessive.

 

[Entropy]

Two good reasons:

1. The setting wasn't used by players. It's obviously a major benefit for an application benchmark if it reflects application use. (As you noted, the setting still isn't used much, even though probably many players today have systems that can comfortably accomodate it.)

2. Once it had been around for a while, the wide use of Q3, at easily recognizeable settings, had generated a huge body of comparative data. This is wonderful for a benchmark, and given the inevitable problems with precise transferability of results from benchmark runs to actual gameplay use it is probably the most important property. Eventually we learned to what extent Q3 depended on CPU - FPU - Fillrate (single and multi) - T&L - Memory bandwidth - memory latency - AGP speed et cetera, et cetera

Compare, gentle reader, with your ability at this point in time, to interpret or predict for instance Splinter Cell benchmark results.

This ability to understand and interpret Q3 benchmark results have been aided by having a high degree of consistency in the settings used, and demos run.

(I'm afraid that Reverends sensible solution of having reviewers recording their own demos for benchmarking not only raises the spectre of non-repeatability by anyone else, but also comes at a significant cost in ability to interpret what the results actually mean on the part of the educated reader. OTOH, it would also systematically undermine the single figure of merit line of thinking, so it would have advantages as well, apart from making cheats more difficult.)

Entropy

 

[Hyp-X]

It does stencil / shadow volume shadows to dynamic objects, without correct lighting. (Just dimming the shadowed area)

IIRC it uses a fake directional "lightsource".

I haven't found it impressive at all.

 

[Nick[FM]]

Ok, I see. But isn't there some other console command too which improve the visual quality? IIRC there was some command that pumped up the polygons on screen? They were touting it when the first GeForce SDR & DDR was out.. I can't remember the command, but I recall that it made the scene look a load better. At least it looked better back then.

I wrote an article/review using these settings like years ago.. Heh, and now I can't even remember the commands.

 

[Ante P]

Another reason might be that lazy reviewers (like me hehe) use Q3 Bench or some other similar proggy where you don't have the option of editing the cfg yourself

 

[Nathan]

I've just been playing around with the shadow settings, and I was not terribly impressed. A bot ran over a bridge and I saw his shadow on the underside of the bridge, so its hardly Doom3 quality. Its quite ironic really, the shadow is technically more more accurate, but its not even close to being in the right place. :? I think I prefer wildly inaccurate shadows at a higher resolution.

 

[Typedef Enum]

Yeah...

What you want to change are:

- r_subdivisions
- r_lodCurveError
- r_lodBias

http://www.technologyvault.co.uk/ge...p;nr=113&catnr=3&prog=gef&lang=en

Given where we are today, doesn't it seem logical to enable such a setting today? I mean, the numbers that websites report are virtually pointless when it comes to Quake3. I mean, I can understand an argument as to whether or not Quake3 is even relevant in 2003...But if you're going to use it (I think it still has value), it seems to me that everything possible should be cranked up?

I can understand why this wasn't enabled back in the GF/GTS days...and probably 1 year thereafter...But at this point in time, it seems to me that the resulting framerate with this feature enabled, regardless as to whether or not it produces the "best" or "most acceptable" shadowing, might be more interesting to look at rather than 450 FPS vs. 425 FPS.

 

[Entropy]

I can understand why this wasn't enabled back in the GF/GTS days...and probably 1 year thereafter...But at this point in time, it seems to me that the resulting framerate with this feature enabled, regardless as to whether or not it produces the "best" or "most acceptable" shadowing, might be more interesting to look at rather than 450 FPS vs. 425 FPS.

I'm sorry, but I don't know how to put this very politely - it would seem to me that you are a bit confused (or at least unclear) as to why you want to run the benchmark in the first place.

Criticising the Q3 settings used in reviews because they produce high FPS values just doesn't make sense in and of itself.

Why would introducing a setting which hardly anyone uses produce benchmark results that are somehow more valid? Particularly since you loose the ability to compare with earlier results/systems? In what respect would they be "more interesting to look at."?

Entropy

PS. (And incidentally - all Q3 players I know personally see to that they get at least 125 fps absolutely solid - requiring 250+ FPS in timedemos. These framerates are what players always aimed for, now they want to have them at higher resolutions to aid long distance railing. Others go for 200 FPS absolutely solid - but this requires unusually competent monitors.)

 

[indio]

ahh Q3 . It is what in my opinion made Nvidia because it's GF2 was the first card that could render at a stable 125 fps and ATI couldn't till the 8500. The reason players want becuase it makes you jump farther and fast . The reason this is an advantage is pretty obvious. for further reference on how the framerate is integrated in the Q3 engine and affects jumping... http://ucguides.savagehelp.com/Quake3/FAQFPSJumps.html

 

[Typedef Enum]

Why would introducing a setting which hardly anyone uses produce benchmark results that are somehow more valid? Particularly since you loose the ability to compare with earlier results/systems? In what respect would they be "more interesting to look at."?

Well, if you're going to make that statement, then you might as well _never_ benchmark Quake3 using either FSAA or AF because hardcore Quake3 players wouldn't enable features that bring their performance down...

I know a couple of hardcore players that don't go beyond 640x480 for God's sake! And if that wasn't pathetic enough, they turn down all the significant IQ settings just to ensure that the performance never drops at all!

So, I don't know...To say that it shouldn't be done because the hardcore gamer wouldn't turn it on seems dubious to me.

Quite frankly, stencil buffer performance would be fairly interesting...if, for no other reason, because none of the manufacturers have had to worry about it too much because it's not used too often, if it all.

 

[Typedef Enum]

BTW, just to give you an idea on performance:

Card: 9800 Pro (430/370)
FSAA: Quality, 4x
AF: Quality, 16x

10x7: 140 FPS
12x10: 97 FPS
16x12: 60 FPS

 

[Nick[FM]]

BTW, just to give you an idea on performance:

Card: 9800 Pro (430/370)
FSAA: Quality, 4x
AF: Quality, 16x

10x7: 140 FPS
12x10: 97 FPS
16x12: 60 FPS

Try these settings:

/r_lodBias -2
/r_subdivisions 1
/r_lodCurveError 10000
/vid_restart

/r_stencilbits 8
/vid_restart

/cg_shadows 2
/vid_restart

What happens to the performance with those settings enabled? With FSAA 4X and AF 16x..

 

[Typedef Enum]

The numbers that I posted include all of those settings...

 

[Nick[FM]]

The numbers that I posted include all of those settings...

Oh, I see! Are there any other settings/commands that would tune up the level of detail or something? To put more workload of the GPU/VPU.

 

[Ante P]

http://www.nordichardware.se/UndaC/Q3Extra.png

Using these settings in my cfg:
r_clear "0"
r_drawSun "1"
r_ext_compressed_textures "0"
r_novis "1"
r_flares "1"
r_lodBias "-2"
r_subdivisions "1"
r_lodCurveError "10000"
r_stencilbits "8"
cg_shadows "2"

Using 8x FSAA and 8x Aniso on a 5800 Ultra
AA/AF basically had no impact on my scores using those cfg-settings.

 

[Nick[FM]]

Can you set a higher number for the r_lodCurveError "10000" ?? I mean, is 10000 the absolute max?

 

[Ante P]

Using the config posted above, demo "four":
1024x768: 162.7 fps
1024x768 with 4x FSAA and 8x Aniso: 99.5 fps

Using the standard high quality config:
1024x768: 310.5 fps
1024x768 with 4x FSAA and 8x Aniso: 211.5 fps


GeForce FX 5800 Ultra 44.10 drivers

This should be pretty GPU VRAM limited so could someone try it using the same settings on a ATi board, I don't feel like installing one just to try this one test.
For the record: Athlon XP 2600+, 768 MB PC2700 cas 2-5-2-2, nForce2

 

[Typedef Enum]

Likewise, on a 9800 Pro...

Using the config posted above, demo "four":
1024x768: 177 fps
1024x768 with 4x FSAA and 8x Aniso: 142 fps

Would you mind doing that test at 12x10 and 16x12? I'm not just saying this because I started this thread, but I think far more interesting things could come out of this sort of analysis than some ridiculously stupid "OK, here's Quake3...As you can see, the 5900 gets 500 FPS, compared to the 410 FPS of the 9800 Pro...Next, we have..."