Announcement

**Joel** · 17 March 2001, 13:35

Found the formula that is used to calculate the 3DMark.

3DMark score = (Game1LowDetail + Game2LowDetail + Game3LowDetail) * 10 + (Game1HighDetail + Game2HighDetail + Game3HighDetail + Game4) * 20

For the rest of the article see www.hardwaremania.com/reviews_eng/3dmark2001/3dmark.shtml

Since the G400 can't do Game4 that is what hurts it's score compared to the GF3. Hopefully that will change soon,

Joel

**RAB** · 17 March 2001, 16:00

What makes more difference is texture compression. On my Radeon 32 DDR (K63+ 550, Win98SE, DX8) I get 1080 with texture compression turned on and 780 with texture compression turned off. Obviously, the lack of support for texture compression hurts scores as much as anything.

My G400 on my other machine (K6-3 450, Win98SE, DX8) gets 570, but it doesn't do texture compression. Nonetheless, it runs much like the Radeon with texture compression turned off (the Radeon doesn't run game 4 either so the scores are comparable).

Is there a way to enable texture compression on the G400 and get scores up? It looks like it adds about 30% to scores. I thought Matrox added some compression for textures in the 6.x drivers, but 3dMark2k1 won't allow it for G400s.

RAB

**Nuno** · 17 March 2001, 18:20

And now I got why Radeon are performing ahead of Ge-force´s: Game 1 and Game 2 HD are using triple texturing, and the 3 texel units on the Radeon can render it in one pass, while the Geforce only has 2 texel units for pipeline..

**superfly** · 17 March 2001, 20:08

I get 3050 3dmarks on the default bench using a dual p3 993+ Gf2 gts 64 meg.

What's curious though is that 3d mark recognizes both cpu's and actually uses them,but the way SMP is implemented isn't very well done since while it splits the workload between both cpu's,they never exceed 50% load each during either the benchmark or the demo run.

When i only use one cpu, it's load is always pegged at 100%,while the other isn't being used at all.

So in reality there's no performance advantage in running both since the score is about the same in either case.

[This message has been edited by superfly (edited 18 March 2001).]

**Admiral** · 18 March 2001, 13:49

1576

P3 650@900, (still) 128 ram, G400 vanilla @150/200 (1xAGP, 128 aperture), Win 98, DX 8.0a, PD 6.50

**alessandro** · 18 March 2001, 16:19

With a little more o/c of my Radeon 32 DDR 200/200,i'm now at 3388 3DMarks at 1024x768x32bit with 32bit Z-buffering.

Using 24bit Z-buffering like the GF2's, i get 3453 3DMarks .

Compare URL--- http://gamershq.madonion.com/compare2k1.shtml?441458

[This message has been edited by alessandro (edited 18 March 2001).]

**Indiana** · 18 March 2001, 17:03

Originally posted by Rags:
How about some Radeon Win2k results?

1689 here.

I expected as much.

Rags

Err, here I get:

- Duron800 only @893 atm, since those pencil-drawings to unlock the CPU have stopped working again...
- ABit KT7
- Radeon 64MB DDR OEM oc'ed to 200/200. No other specific tweaks, especially full 32Bit Z-Buffer
- Win2k SP1, Radeon 5.13.01.3114 Win2k driver

As you're also posting other problems with the Radeon in Win2k that I do not have, I think you should recheck your system config. ATIs Win2k drivers are far from being perfect but they're not as bad as you keep claiming, they're overall very stable here. Now, if ATI could finally resolve the lacking performance in specific areas, e.g. take a look at the high Polygon-count results. I guess here we'll have to wait for another trick like the BetaDX8-DLL that cured these low Radeon/Win2k scores in 3DMark2000

EDIT: Sorry for the late reply, but I didn't see your low score until now

[This message has been edited by Indiana (edited 19 March 2001).]

**cbman** · 18 March 2001, 21:01

I get 2516

Cele II 566@875
ASUS CUSL2 Bios 1006
256MB PC 133 at CAS 2
Creative Labs Ann. 2 GF2 GTS (210/360)
A whole buncha Extra Crap
and a nice trackball

I also think that because the Radeon does both EVBM and DOT3 BM that it also gets a boost in its scores... maybe this Benchmark is just for judging how many options is squeezed into a card... (I would post a pic.. but my webspace is gone..LOL)
so I will teadiously write it out by hand...

(Default Bench)

Results

3DMark Score 2561
Game 1 - LOW 53.1
Game 1 - HIGH 12.5
Game 2 - LOW 41.0
Game 2 - HIGH 15.6
Game 3 - LOW 56.7
Game 3 - HIGH 24.6
Game 4 - NOT SUPPORTED
Fill Rate (Single) 296.9
Fill Rate (Multi) 566.3
High Poly (1 Light) 16.3
High Poly (8 Light) 3.1
EVBM - NOT SUPPORTED
DOT3 BM - 47.3
Vertex - 20.4
Pixel - NOT SUPPORTED
Point - 8.2

And this Benchmark doesn't want to run on my classic Matrox Ultima (2MB)... (Go Figure)

------------------
Hang Low and Limber

**impact** · 19 March 2001, 03:39

Can't catch them!

**Indiana** · 19 March 2001, 11:41

Originally posted by cbman:
I also think that because the Radeon does both EVBM and DOT3 BM that it also gets a boost in its scores... maybe this Benchmark is just for judging how many options is squeezed into a card...

No, the EMBM and Dot3BM tests don't influence the final score, this depends only on the fps in the 4 game scenes with a bit stressing of the high detail scores. You can find the formula to calculate the score yourself somewhere in this thread (page2, I think).

Take a look at your high-detail fps and you'll know why your score is quite low.

Then take a look at the Fillrate-tests and you'll probably get the idea why the Radeon beats the GTS: 3DFake2001's engine apparently is one of the very first to actually use the Radeons triple texturing units, giving it a slight edge over the GTS with it's dual texturing.

If you want to see one of the Radeons drivers major weaknesses in Win2k, take a look at the high-polygon tests: the results are terribly low, lower in fact than when using software T&L. The same problem was apparent with the combination Radeon/Win2k/3DMark2000 but this could be resolved by using one DLL out of the Beta146 of DX8 instead of the release-version.

[This message has been edited by Indiana (edited 19 March 2001).]

**Rags** · 19 March 2001, 12:26

As you're also posting other problems with the Radeon in Win2k that I do not have, I think you should recheck your system config.

Uh, sorry. But ATI tech support has already reproduced my crashes and issues (every one of them), and they assure me they are looking into fixing them in the next release or the one after. So you can take that "system config" theory and push it to the side

ATIs Win2k drivers are far from being perfect

I have a better description than that

but they're not as bad as you keep claiming

Yes they are, in fact I am quite confident that if I used my Radeon more often, I would have no problems finding more problems.

they're overall very stable here.

Well, I am not there, so that means nothing to me.

Now, if ATI could finally resolve the lacking performance in specific areas, e.g. take a look at the high Polygon-count results.

The performance problem is only half of the equation.

Rags

**cbman** · 19 March 2001, 12:30

Yeah.. I read the post about how the score is created... but then why bother even including other tests in if they don't factor at all... makes me think this test may be flawed...

I do agree with you on the triple texturing though (Being a good thing if your card can render 3 pixels a pass as you get to see the full effect).. but then any other card owner can't get the actual fill rate of multitexturing on their cards... another thing to throw off the validity of the test I think...

Just MHO

Charles

------------------
Hang Low and Limber

**Joel** · 19 March 2001, 13:34

All of the different version of 3DMark has calculated it's 3DMark scores based on the fps achieved in the games. Final Reality was probably the last benchmarker that I have used that did take everything into consideration. But we also all know that 3DMark can be a very inaccurated benchmark. With a simple hack it can show scores to be alot more than they would be normally. Just check out this thread forums.murc.ws/ubb/Forum13/HTML/000015.html

Joel

[This message has been edited by Joel (edited 19 March 2001).]

**Indiana** · 19 March 2001, 13:35

cbman, I agree that all 3DMark benchmarks are seriously flawed, I've started a thread about this topic sometime ago in this forum.
What do you think of a benchmark that is "coincidentally" released just after a gfx card of a certain vendor is available and makes sure that it has one feature that is ONLY available on this specific card: the nature test, and this one in fact goes into the score giving the GeForce3 a much higher score even if it's not one frame faster than a GeForce2 in the other tests (BTW, they're doing this the second time now, think of the T&L support of 3DMark2000 being only available on the Geforce and boosting it's scores over the competition without any relation to real-world performance).
BTW, this thing with only the game-scenes fps going into the final score is there at least since 3DFake2000.

Rags,
seeing you posting a score of 16xx tells me that there IS something majorly wrong on your system (or do you have a P90?). I also can't reproduce your artifacts when using IE and I'm heavily using this with up to 20 windows open most of the time. I don't use any of the other programs you mentioned in the other thread so I can't double-check these, but here the only problem with ATIs Win2k drivers up to now has been the sometimes lacking performance.
Or maybe it's an Intel vs. AMD issue here?

**Joel** · 19 March 2001, 14:00

BTW, this thing with only the game-scenes fps going into the final score is there at least since 3DFake2000

Actually it's been doing it since 3DFake1999.

Joel

Announcement

3DMark2001 out TODAY!

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment

Comment