Q3Test2 Benchmarking!! Post your scores!

[CoolBreeze]

I got a G400 reg 32MB single head...and i get about 25 fps on demo 1 and 30 fps on demo 2. This is on a C366 not overclocked and 128mb ram. At 1024x768x32...with all eye candy turned up and high quality sound...only 3 things i change are:

1. High quality sky to low
2. Disable V sync
3. Type cg_simpleItems into the console for
speed improvement. It only changes the
items icons...weapons, health, armor,
etc.

I know i dont have the fastest PC in the world but i find these rates to be satisfactory, because the image quality more than makes up for it. I also believe that there will be major driver improvments and overclocking utilities in the very near future. And another thing that i find very remarkable is that even with these framerates i dont get anykind of slowdowns in a game of like 20+ ppl in heavy combat. Oh well...my $.02

CoolBreeze

[acobra]

I'm not an expert, so I'm not sure about this, but wouldn't Hi-Res tests also be affected by the quality of the drivers?

(i.e. better implentation of certain hardware features for less of a hit?)

[voyager]

I did some q3test 1.06 benchmarks at 1024x768x16 with 16 bit textures. My system is a C366 overclocked to 578Mhz, 256MB oF CAS-2 RAM, Win98, and G400 32MB DH o.c to 142.5/190Nhz. I am using PD 5.13.

All game options were turned on except sync every frame. In graphic options I used bilinear texturing with geometric detail set to high and texture detail set to position 3 of 4 (counting from left to right).

In Q3testdemo1, I get 43.7 fps. In q3testdemo2 I got 50.9fps.

The same as above with trilinear textures yielded 35.6/22.8 fps.

Restoring biliinear textures and putting texture detail at the rightmost position the results were 42.1/43.4fps.

Turning all the game options to NO, yielded 57.6/55.0 fps.

The above with geometic detail set to low and texture detail set to position 2/4 yielded 58.1/56.1 fps.

These are 'slightly' better results than what other people with slower CPUs are posting. :-)

If there is a 15 to 20 percent improvement in the drivers I think 50-60fps would be tolerable at 1024x768.

I haven't had a chance to do other resolutions.

Greg

[This message has been edited by voyager (edited 07-21-99).]

[Stil]

acobra: you are correct in that high-res tests will also be _slightly_ affected by the quality of the driver, but to a much lesser degree than low resolution tests.

When benchmarking a card's quality there are basically two factors that should be of importance to gamers. One is the card's fill-rate, and the other is the driver's triangle rate. Triangle rate is a measure of how fast the driver can get triangles from the game to the card, and fill-rate is how fast the card can fill triangles once they get there.

Benchmarks can be tuned to test one factor or the other, or a combination of the two factors. For example: A benchmarking app that just sents millions of tiny (10 pixel or so) triangles to the card is primarily excersising the driver's triangle througput. The card's fill-rate is not really tested in this case, because the triangles are so small.

Conversely, a benchmarking app that sends large textured triangles to the card is testing the card's fill-rate, triangle rate is virtually insignificant when the card has to draw, say, 500k pixel triangles.

The point:

On today's hardware (G400, TNT2) low-resolution Quake tests (1024x768 and under) tend to flex the driver's triangle rate, because their drawing engines are so fast. The fill rate is also partially a factor, but not as much. On the other hand, Quake tests at higher resolutions tend to flex the card's fill rate, because the cards take much longer to draw each triangle. The triangle rate is partially a factor, but again... not as much.

The conclusion:

A benchmark that tested purely fill-rate would put the G400 ahead of say the TNT2, because the G400's raw hardware is very fast. On the other hand, a benchmark that tested purely OpenGL triangle throughput will show the TNT2 beating the G400 soundly, because the G400's OpenGL drivers need work.

Using Quake as a benchmark gives a good "overall" performance reading that encompasses the hardware _and_ the drivers, but puts the focus on neither of them.

[Jake]

Hey Stil!

That's one solid post, dude!! Gotta give you a big slap one the back for that one.

Man, did somebody say this thread was ignorant and did not belong among the 'true' hardware issues? Way to go!

We need more people like you posting here. Holy cow, I think I actually learned something about 3D performance! Unreal!

Regards,

Jake

[Wraithy]

Hello to all. I'm new here, 'been reading this forum some 3 weeks now. I heard about G400 and got interested. I have used old Millenium as second graphic adapter and the dualhead in G400 would be marvelous.

I was too impressed about the Stil's mail. I definitely learned something. Is it really true that G400 has so much better HW than TNT?

I have that TNT so I wont bother sending my framerates, those have been seen many times now. I still read with interest, what G400 gets. BTW, MURC has put new framerates in News-section, with 5.2 drivers. Those are little better, eh? How long before users get those?

Matrox has always had best image quality. My friend bought g200 some time ago (7-8 months) and got very annoyed because of the og-drivers. He said he won't buy another Matrox product anymore. I hope (as have again many others said before me) that Matrox gets _good_ og's ASAP.

Well, this is my first posting to (almost) any forum, so if there are mistakes, ignore them

Wraith

------------------
AbitBH6; c300a/oc450; 320MBcas2pc100; 2940scsi; 2*6,4+4,3GBide; 4,5Gb7200scsi; idecd+scsicd; TerratecEWS64s(34Mbmem); DMonsterMX300; MMillII(pci); Asus3400TNT(agp); 19"Compaq; 17"Nokia; 3Com10/100NIC; Zip100par; HP820Cxi; HP4100scan; Videocam. Going to get scsi dvd and cdrw.
And G400.

[Buuri]

Stil,

After reading this
"A benchmark that tested purely fill-rate would put the G400 ahead of say the TNT2, because the G400's raw hardware is very fast. On the other hand, a benchmark that tested purely OpenGL triangle throughput will show the TNT2 beating the G400 soundly, because the G400's OpenGL drivers need work"
I couldn't help wondering that how do you know that slow triangle throughtput is because of the bad drivers?

Why can't it be of the hardware? Sounds like speculation to me.

B

[This message has been edited by Buuri (edited 07-22-99).]

[acobra]

Stil,

Actually, I misread your post... I thought it meant More graphics options on (HI-RES) vs. More or All options off (LO-RES). In that case would what I said be true??

I'm not trying to discount your very descriptive and informative post, BTW. I definetly learned something. I'm just correcting myself!!

[Stil]

Burri: Given the G200 and G400's Direct3D performance at low-resolution-high-triangle tests, we know the card is _capable_ of high triangle rates, so one can only assume that the low-resolution problems in OpenGL are driver-related.

acobra: as far as "graphics options" go, I assume you are referring to the different tweaks that Quake3 lets you adjust. It really depends on the option. An option that instructs Quake3 to send more triangles (i.e. higher quality geometry) will probably slow the G200 down slightly.