V5 faster than GF2-U ? new drivers w/surface removal reported

[cobosdan]

Whaddaya know !


http://www.voodooextreme.com

Excerpt:

V5 Hidden Surface Removal 4:33 PM (your time) - Jasen "Torrid" Torres - FAQ(s)/Tips/Guide(s)/Software Tweaks - (56)
In the latest set of V5 drivers, 3dfx slipped in minimal beta support for HSR (Hidden Surface Removal). Word on the street is that it improves Quake 3 Arena benchmarks to the level of a GeForce 2 Ultra and above. Here's a little juice on how to try it, but be careful with the registry stuff:

1. First make sure you have the latest 3dfx beta drivers. Get them here.
2. Get the little registry hack here(the guy who made the registry hack) or even here (from the folks at Rivastation) You can input the values yourself using "regedit" from Start/run, or just double click the file to install it. Make sure the display adapter number is the same as the V5 in your system. Look at the string HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\Class\Display] in regedit, and make sure the registry hack file has the same number- for most it will be [HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\Class\Display\0000]
3. The setting "value" determines the strength of the effect, 0 being disabled and 4 being strongest. This can also be achieved through the 3dfx drivers under advanced openGL settings, and then hidden surface removal.

If it doesn't appear to be doing anything in Quake 3, you can make this little batch file to start Q3A. The value for FX_GL_HSR= can be 0, 1, 2, 3, or 4. The higher the setting, the greater the HSR effect, and the more graphical errors you are prone to get. It's a trial and error to see what works best for you.
@ECHO OFF
PROMPT $P$G
SET FX_GL_HSR=4
QUAKE3.EXE
I'm going to crunch out some benchmarks here, so expect me to update this with some numbers, and maybe a screenshot or two if it looks as bad as I think it might.

End excerpt


Doesn't faze me at all even if proven true cause I don't care for bots' games like Q3, but then again, Elite force, Alice, Fakk, hmmmm ...

Time will tell

[amos]

Nice try but Vodoo 5 will not bench mark out as fast!

[pidge]

Does it only improve OpenGL performance or only Quake III Arena.

[Biohazard]

give up on that quest. It's like trying to get a camero to out run a porche. While the camero IS fast, it's just not as fast.

[Reg Kmet]

hehehe voodoo 5 would need magical drivers to do that!

[cobosdan]

Guys, take it easy, I'm just an expectator like you, no opinions, waiting for confirmation or whatnot, maybe hoping, but who wouldn't?

Just remember than T-buffer is what permits 3DFX's kind of FSAA.

Excerpt

"AirGibson 11/30/00 09:02 Hope you all are ready for a long read. Here it comes!

I spent several hours yesterday playing with HSR options, and quite frankly, it is the biggest performance increase I've ever seen or heard of. First, let's see the bad news:

- At your highest resolutions such as 1600 x 1200, the higher HSR settings cause graphical problems that I can't get rid of. In some cases (I.E. 1600 x 1200 with HSR 4) the problems are so bad it is unplayable. Sure, your FPS go through the roof and then some, but that obviously doesn't mean squat if you can't play the game.

- Direct 3D games get nothing from HSR.

- These beta drivers are "nVidia-esque" in the sense that sometimes they're difficult to set up for people who aren't familiar with a bit of technical jargon. Either a registry hack or a hand-made batch file is needed to work with HSR.

Keep in mind these *are BETA* drivers, so certainly some of these problems are to be expected.

Damn...with all this bad stuff, the results better be pretty friggin’ good, right? On to the benchmarks.

My system:
P3 700 (O/Ced to 815)
128 MB PC 100 SDRAM
ASUS P3V4X mobo (VIA 133a Chipset, latest BIOS and 4 in 1 drivers)
Voodoo 5500 (not O/Ced)

Just a quick note...when you see the markings (1) and (2) in my results, this is what I mean:

1 - Graphical glitches were very apparent at first, but disappeared when I enabled V-Sync and set 32 bit Depth Precision to "fast".

2 - Graphical glitches were very apparent at first, but disappeared when I enabled V-Sync, set 32 bit Depth Precision to "fast", and hard-capped the frame rate at about 15% below the fps average I was getting from the demo.

Here's what I got from Quake 3 Arena using timedemo 1, demo 001 across several different resolutions, FSAA settings, and HSR settings.

HQ - 800 x 600 – 2x FSAA – No HSR – 59.2 – Ran Perfect
HQ - 800 x 600 – 2x FSAA – HSR 2 – 80.3 – Ran Perfect
HQ - 800 x 600 – 2x FSAA – HSR 3 – 87.1 – Ran Perfect

An extra 28 FPS here at no cost to any image quality at all.

HQ - 800 x 600 – 4x FSAA – No HSR – 27.8 – Somewhat Playable (poor FPS)
HQ – 800 x 600 – 4x FSAA – HSR 1 – 36.2 – Very Playable (so-so FPS)
HQ - 800 x 600 – 4x FSAA – HSR 2 – 52.9 – Ran Perfect(1)
HQ - 800 x 600 - 4x FSAA - HSR 3 - 71.8 - Ran Perfect(1)
HQ - 800 x 600 - 4x FSAA - HSR 4 - 78.2 - Unplayable (glitchy)

Averaging 71.8 fps with absolutely ZERO graphical errors in 4x FSAA mode here.

High Quality - 1024 x 768 - No FSAA - No HSR – 68.9 – Ran Perfect
High Quality - 1024 x 768 - No FSAA - HSR 1 - 75.5 – Ran Perfect
High Quality - 1024 x 768 - No FSAA - HSR 2 - 71.3 – Ran Perfect
High Quality - 1024 x 768 - No FSAA - HSR 3 - 67.7 – Ran Perfect
High Quality - 1024 x 768 - No FSAA - HSR 4 - 67.9 – Ran Perfect (1)

My system ain’t the best in the world, but it’s pretty good. With PC 100 memory and a VIA chipset, I’m CPU limited in this case, hence, HSR doesn’t amount to a hill of beans for me except for HSR 1 giving me 7 extra FPS. This isn’t true with other folks who aren’t CPU bound at this level.

HQ - 1024 x 768 - 2x FSAA - No HSR – 36.7 – Ran Perfect
HQ - 1024 x 768 - 2x FSAA - HSR 1 - 45.6 – Ran Perfect (1)
HQ - 1024 x 768 - 2x FSAA - HSR 2 - 63.2 - Ran Perfect (1)
HQ - 1024 x 768 - 2x FSAA - HSR 3 - 80.6 - Ran Perfect (1)
HQ - 1024 x 768 - 2x FSAA - HSR 4 - 86.9 - unplayable (glitchy)

Check that out, guys...1024 x 768, 2x FSAA, High Quality running at 80 FPS average. That is *unreal*.

HQ - 1024 x 768 - 4x FSAA - No HSR – 15.2 - unplayable(low fps)
HQ - 1024 x 768 - 4x FSAA - HSR 1 - 20.7 - unplayable (low fps)
HQ - 1024 x 768 - 4x FSAA - HSR 2 - 32.9 - unplayable (glitchy)
HQ - 1024 x 768 - 4x FSAA - HSR 3 - 48.3 - Ran Perfect (2)
HQ - 1024 x 768 - 4x FSAA - HSR 4 - 61.6 - Ran Perfect (2)

MUHAHAHA...4x FSAA at HQ 1024 x 768 with frame rate hard capped at 50 fps. Can the GF2 ULTRA even do this with a comparable FSAA setting?

HQ – 1280 x 1024 – No FSAA - No HSR – 42.4 – Ran Perfect
HQ – 1280 x 1024 – No FSAA – HSR 3 – 80.4 – Ran Perfect

Yes, that's right, HSR 3 doubled my frames per second and I did not see a glitch in the 5 minutes I played at that resolution. No need to tweak anything.

HQ – 1280 x 1024 – 2x FSAA - No HSR – 20.5 – unplayable
HQ – 1280 x 1024 – 2x FSAA – HSR 3 – 80.4 – unplayable

I didn't have time to test the other 1280 x 1024 settings, but I will soon enough. It appears that FSAA at this resolution is tough to deliver with HSR, but, like I said, I haven't tested all the combinations yet.

So, overall, here’s what I gained with my system:
- Went from 59 to 87 fps in 800 x 600 / 2x FSAA
- Went from 28 to 71.8 fps in 800 x 600 / 4x FSAA
- Minor gains in 1024 x 768 with no FSAA
- Went from 36 to 80.1 fps in 1024 x 768 / 2x FSAA
- Went from 15.2 fps *average* in 1024 x 768 / 4x FSAA to a solid 50 fps.
- Went from 42.4 to 80.4 fps in 1280 x 1024

I've read several articles saying "HSR is too glitchy, it's unusable, it's worthless, etc..." and I can say beyond a shadow of a doubt that they're full of it. All I can do is SWEAR that the above is accurate based on my system. On all of the ones that I marked "Ran Perfect" I mean that I saw absolutely zero graphical anomolies while running the demo and then playing the game for several minutes.

Here's a quick comparison that should shake the foundations of the hard-core gaming world:

Quake 3, High Quality, 1280 x 1024, No FSAA

Voodoo 5500 on my P3 815: 80.4
Leadtek GF2 GTS O/Ced on similar system: 46
ELSA Gladiac GF2 ULTRA O/Ced on 1.1 GHz Athlon: 89.5

I think we'll agree when we say that Quake 3 is the "crown jewel" for GF2 benchmarking. It's one of the few games currently supporting T&L and it is Open GL, the realm where GF2 has pretty much ruled. I'm shocked!


Now, here is what I'm REALLY interested in: FSAA.

Running HQ, 1024 x 768

Voodoo 5500 on my system, 4x FSAA: 61.6 avg, capped at 50
GF2 GTS on similar system, 2x FSAA: 16.7 avg

Notice that I'm running *4x FSAA on a Voodoo 5500* while the GF2 is only running their 2x. Not to mention that their 2x FSAA is debatably inferior.

In one word, WOW. This is as good as getting a brand new video card! nVidia will certainly try to respond, but I wonder how successful they will be. Their FSAA response proved to be a bit inferior to 3DFX's FSAA. I wonder if HSR will tell the same story should nVidia develop their own version."

End of excerpt

I'll keep waiting, cause I like the "see to believe" school ...

See ya

[Coco]

This seem too good to be true... but if it that HSR driver only improve Quake3, it doesn't make much sense... It's like 3dfx rewriting the code to optimize for Q3, or even using a "Cheat" method to get higher Benchmark score.

I am not buying it until I see a general improvement in MOST games...

[Hipnotic_Tranz]

Oh my god! That's amazing. I'd love to see it work on other games. man, I just can't see what 3dfx did to make such a performance gain. I mean, that's incredible. Though, I'm already getting an ATI AIW 32meg DDR, so unless it proves to not run on my Abit KA7-100, I'm stickin' with 'er!

[legion88]

As I mentioned in this thread ( http://www.pcabusers.net/forums/showthr ... 8#post4458 ), 3DGPU's forum has a snapshot of what the V5 has to offer ( http://forums.nv10.net/ubb/Forum1/HTML/002873.html ).

There is an example of the image glitches often mentioned. The author of the screenshot notes that he had to take several shots in a row first before he was able to get the one with the glitch. Apparently, these glitches happen in "flashes". That is, the screen looks good for about a few second, then boom, the screen screws up and lasts about 1/2 a second or two (long enough for you to notice and take a screenshot), then returns to normal. It sounds like the card won't render the frames properly ever so often.

There is also another screenshot where there is no image glitch. It has a frame rate counter enabled. It shows 222 FPS. That looks great. But it was also noted that the GF2 Ultra can achieve that performance at 1024x768x32 HQ. Thus, it appears that the V5 with just that one screenshot equaled and maybe surpassed the Ultra's performance. It could be, though, that for that part of the map, the CPU is the bottleneck and not the video cards.

If someone can tell me the name of the map that screenshot was made, I can post the frame rate I get on my 366 million pixels per second Radeon.

[cobosdan]

Just saw a post by him in http://www.3dfxgamers.com. in a thread giving steps to getting perfect graphics with HSR enabled in Q3.

Good to see the ole' people waggin their tail still !

"Post a reply
wupass 12/1/00 11:40 In the V5 advanced setting for opengl set the v-sync to software controlled. When you enter Q3 you can turn v-sync on or off in the game options. It worked fine for me with v-sync enabled from within Q3 but not if I set v-sync on in the V5 advanced properties.

[DirkBelig]

[yawns]

GIVE IT UP!!! 3DEFECTS IS NEVER GONNA CATCH UP!!!!!

[goes back to sleep]

[Hipnotic_Tranz]

Never say Never You can't argue with that kind of performance jump. If this was a GeForce and you got a hack to where it performed like a GF2 Ultra, you all would be shittin' your pants right now and laughing, "haha, I didn't even have to pay the "ultra" price!"

Just cause the 3dfx lable doesn't mean it's false, not legit, or not good. Look at the frame jump there. That is definitly a noticable jump. 3dfx has not gone under nor do I think they will for a while. They are in "third place" in my mind. (behind ATI and nVidia) but that's not bad, to me. I'm willing to give 3dfx a shot if they pump out another graphics card that can get some better fps in the games I play.

If this fix happens to be only Q3; then that's great for the q3 players that play with a V5. If it works in all games, even better! Me on the other hand will be recieving my ATI AIW 32meg Radeon shortly and am not going to switch for a while unless the V5 proves to be as good as a GeForce2 Ultra in all games and outwights all the features the All-In-Wonder has. Just my humble opinion.

[Biohazard]

you can get the performance of the Geforce 2 ULTRA if you overclock the right GTS. Not real difficult.

I hate to sound like I'm flaming the V5, cuz I'm not. But it's just not worth farting with it anymore. A radeon or GTS can out perform it without any crazy tweaks to it. I'm sorry, but "glitchy" and "unplayable" aren't on most people's minds when they sit down at thier computer, they want speed and reliability NOW (I happen to be one of them ). I think more people will care about those things. Not to mention less and less software is actually supporting glide. I mean who needs it with the other cards doing so well in openGL and D3D?

[Bellz]

If you read Firing Squad's article about it <a href="http://firingsquad.gamers.com/hardware/ ... efault.asp
">HERE</a>, you see that they came to some VERY NEGATIVE conclusions when enabling this feature. Nice try Cobosdan, but NO CIGAR. 3dfx isn't going to be impressing anyone anytime soon, unless Rampage can pull them back from the grave.
Snippet from FS:
"The huge HSR gains are very deceptive. The higher the resolution, and the more aggressive the tiling, you start encountering serious in-game glitches, mostly flickering. You're able to see objects behind walls or below floors as those areas flicker. We tried to take screenshots, but both the in-game screenshot function and Hypersnap failed to work. We had to resort to the digital camera.

Frankly, the flickering makes the game unplayable, and we don't trust the framerate counter very much. We noticed that the framerate spikes up whenever the screen flickers. You can reduce the flickering by lowering the maxfps, but you'll still be playing with the occasional flicker if you set the maxfps too high.

It is nice to see that the 1.04.01 drivers do provide a small performance increase over the 1.03s at normal, non-HSR settings.

Yes, the benchmark numbers are impressive, but they are misleading. The HSR glitches make the game very hard to play, and they may be the cause of the inflated scores. We couldn't accurately judge the framerate at the more aggressive settings because the flickering was too great. One way to reduce the flickering is to cap the framerate, but you have to set the cap fairly low.

Even with these less than impressive results, it would be stupid to label 3dfx's HSR as a failure at this point in time. It's an undocumented, unfinished feature. We just have to step back and accept this beta HSR feature as an indication of things to come. It'll be interesting to see what kind of playable framerates HSR will yield once 3dfx officially releases the feature."

[Edited by Bellz on 12-02-2000 at 08:16 PM]

[legion88]

Over at 3DGPU and 3dfxgamers forums, there is 'Sharkfood' trying his best to advertise this brand spanking new HSR drivers. Amusing what some people would do.

In this thread, cobosdan hasn't said anything yet that I can see. But that is because, he is quoting other people and letting them do the talking. Sharkfood with no balls.