PDA

View Full Version : what benchmarks do you care about?


3Dfx_Sage
12-23-2003, 12:13 AM
I'm doing an article on a pro workstation board vs their gaming equivalents and was wondering what benchmarks you all care about the most. There are a few obvious choices, of course, but what about the less common ones? (feel free to vote for the obvious ones as well, that way I can see how much importance people place on them relative to others)

dvornik
12-23-2003, 06:55 AM
Max:

The benchmarks on the second CD of max itself. Both viewport performance and rendering. Description and some results can be found here:
http://www.discreet.com/support/max/videocards/ (or just look around max support page)
SPECapc for max (viewport and some rendering):
http://www.specbench.org/gpc/apc.static/max42info.html
Greg's benchmarks on 3dluvr (rendering)
http://www.3dluvr.com/content/maxbench.php

[edit] When testig Max make sure to include information on the viewport drivers. Custom drivers (Maxtreme for Nvidia, Maximum for ATI) are generally the fastest but quite often don't really work. There's no Maxtreme for max 6 for instance. Ideally a review would include custom, OGL and Direct3D results on the same page.

Maya:

SPECapc for Maya (mostly viewport but also i/o and is very processor dependant):
http://www.specbench.org/gpc/apc.static/maya5info.html
Highend3d rendering:
http://highend3d.com/tests/maya/testcenter/
Maya Testcenter rendering and viewport (results database no longer maintained unfortunately):
http://www.maya-testcenter.de/

I don't care much about non-application-specific benchmarks, but Specviewperf is a good "professional" benchmark that definitely needs to be included in any workstation review.

Would be great if you post a link to your article when it's up.

Novakog
12-23-2003, 08:49 AM
3dfx nailed my Maya ones, I was gonna say those exact 3.

imashination
12-23-2003, 09:21 AM
www.cinebench.com

and

http://www.imashination.com/bench.html
The last column shows comparable gfx card speeds.

dvornik
12-23-2003, 10:52 AM
I've never paid attention to cinebench but actually it looks very interesting. It would be nice to be able to sort the results by column to come to some sort of conclusions. imashination, could you post them in excel format or something over there?

One thing that stands out is that in Cinema 4D "pro" cards have no speed advantage whatsoever compared to gaming cards. I believe the situation is the same in Maya but can't really prove it.

imashination
12-23-2003, 03:59 PM
Originally posted by dvornik
I've never paid attention to cinebench but actually it looks very interesting. It would be nice to be able to sort the results by column to come to some sort of conclusions. imashination, could you post them in excel format or something over there?

One thing that stands out is that in Cinema 4D "pro" cards have no speed advantage whatsoever compared to gaming cards. I believe the situation is the same in Maya but can't really prove it.

Unless you're using CAD or badly programmed software (no names mentioned ;-)) The gaming cards will serve you just as well as their expensive brothers.

I'll put up an excel file, or get a database web thingy as soon as I can, you're not the first to ask. For now they are simply sorted by rendering power.

Emmortal1
12-23-2003, 05:24 PM
"Unless you're using CAD or badly programmed software (no names mentioned ;-)) The gaming cards will serve you just as well as their expensive brothers."

Sorry but that's a blatently false statement. Stick a Geforce FX 5900 in a system and try tumbling around the viewport with a model of 500,000+ faces with verts showing and tell me you will get the same performance as a Quadro FX 3000 does. You won't in Max and you certainly won't in Maya.

Emmortal

imashination
12-23-2003, 06:45 PM
With equivelent (sp?) gpus, yes, mostly. ie Radeon 9800Pro and FireGL X2. GF2 and Quadro 2 etc.

There will be differences due to the high end cards typically having better timed memory, but these are marginal differences.

Regarding the Quadro 3000, I don't know enough about this one to comment.

Oh, and max is hardly an ideal display speed benchmark, it struggles with 500,000 polys regardless of the gfx card used.

dvornik
12-23-2003, 07:01 PM
I think you can't make general statements like that without benchmarks to support them - that's what they are for.

As far as max goes Maxtreme will not work with Geforce FX cards and therefore they would be MUCH slower than Quadros. In max 6 they would perform about the same (slow that is) cause there's no Maxtreme for it.

Other than that I think there's a lot of convincing evidence to support imashination's statement. Benefits of pro cards in DCC software are not in any way obvious.

GregHess
12-24-2003, 01:27 PM
And just for comparison sakes...

Here's a nice review of a whole slew of cards with various bits and pieces of these benchmarks.

http://www.3dchips.net/content/review.php?id=63&page=1

imashination
12-24-2003, 02:33 PM
Originally posted by dvornik
I think you can't make general statements like that without benchmarks to support them - that's what they are for.

As far as max goes Maxtreme will not work with Geforce FX cards and therefore they would be MUCH slower than Quadros. In max 6 they would perform about the same (slow that is) cause there's no Maxtreme for it.

Other than that I think there's a lot of convincing evidence to support imashination's statement. Benefits of pro cards in DCC software are not in any way obvious.

I have benchmarks to back them up ;-) www.imashination.com/bench.html

dvornik
12-24-2003, 02:39 PM
Yep. That link Greg posted shows similar (but not so drastic) results in Maya. My Quadro card was visibly slow on that cinebench T&L.

Ice Czar
12-24-2003, 02:41 PM
http://www.iozone.org/
http://www.iozone.org/docs/IOzone_msword_98.pdf
http://www.iozone.org/src/current/Windows2000.xls (excel)
filesystem benchmark

3Dfx_Sage
12-24-2003, 04:41 PM
Originally posted by imashination
I have benchmarks to back them up ;-) www.imashination.com/bench.html that's only cinebench though. there is a huge difference between gaming an pro cards (at least from nVidia) when using OpenGL. Why? Becuse nVidia's Quadro OpenGL drivers are a hell of a lot better than their GeForce ones. I boosted the speed of my GeForce4 in GAMES (that used OpenGL) by SoftQuadro'ing it... and I got upwards of a 75 percent boost in some!!!!

dvornik
12-24-2003, 04:59 PM
That's why I prefer application-specific benchmarks. I wouldn't jump to conclusions as to which drivers are "better". Normally gaming cards outperform Quadros in gaming benchmarks (3dmark is an obvious place to look). Quite often people who use Softquadro complain that it hardly made any difference at all in their application.

imashination
12-24-2003, 06:37 PM
Originally posted by 3Dfx_Sage
that's only cinebench though. there is a huge difference between gaming an pro cards (at least from nVidia) when using OpenGL. Why? Becuse nVidia's Quadro OpenGL drivers are a hell of a lot better than their GeForce ones. I boosted the speed of my GeForce4 in GAMES (that used OpenGL) by SoftQuadro'ing it... and I got upwards of a 75 percent boost in some!!!!

Im going to call your bluff. Softquadro has never in any test every performed, improved a game frame rate. In fact quite the opposite, they often have a very negative effect on the speeds.

Give me a card model, game and driver number which improve performance over standard drivers.

3Dfx_Sage
12-24-2003, 07:27 PM
card: GeForce4 Ti 4200 64MB overclocked to 305/620, driver version I don't remember- long time ago. game: UT2003 beta (using OpenGL beta renderpath, was faster than OpenGL and Direct3D in GF mode) game: Doom3 (Alpha), benchmark: codecreatures.

imashination
12-24-2003, 08:15 PM
http://www.hothardware.com/hh_files/S&V/qfx3k(4).shtml

I win.

dvornik
12-24-2003, 08:31 PM
I'm always amazed by the fact that Quadros outperform Geforces in synthetic benchmarks by such a large margin while being almost the same in applications benchmarks (if you disregard Maxtreme). I wish you could model in Specviewperf.

3Dfx_Sage
12-24-2003, 11:33 PM
Originally posted by imashination
http://www.hothardware.com/hh_files/S&V/qfx3k(4).shtml

I win. ???? you win? you win what? and what is that link supposed to prove?

dvornik
12-24-2003, 11:42 PM
Come on , there's no point to get competitive over this. The link shows GFFX 5900U outperforming Quadro FX 3000 in a Quake 3 OGL test while underperforming in Direct3D in Unreal Tournament.

CgFX
12-25-2003, 03:41 AM
Originally posted by 3Dfx_Sage
there is a huge difference between gaming an pro cards (at least from nVidia) when using OpenGL. Why? Becuse nVidia's Quadro OpenGL drivers are a hell of a lot better than their GeForce ones.

NVIDIA only has one driver. There are parts of that driver that benefit from on-chip Quadro goodness and there are parts of that driver that are only for Quadro but it is still just one source tree and one driver that gets built.

That being said, I would be interested in what differences you saw in OpenGL games since the on-chip and driver features and optimizations aren't really targeted at areas that would benefit a game. Additionally, most games are fill-limited so your bottleneck is typically the memory bus clock within the same architecture family and not GeForce vs. Quadro.

CgFX
12-25-2003, 03:50 AM
Originally posted by 3Dfx_Sage
???? you win? you win what? and what is that link supposed to prove?
Actually, his link unintentionally proves my point that the difference in OpenGL games between Quadro FX and GeForce FX will be dominated by clockrate differences and not by any Quadro vs. GeForce differences. In the Quadro4 and GeForce4 era that you discuss the clocks were the same or closer so Quadro vs. GeForce differences might have more of a role. Again, I doubt what you claim and certainly not the that level because much of what Quadro is about doesn't really benefit a game.

BTW, Light-06, DX-08, and DRV-09 are pointless even for sythetic benchmarks and need to be shot. They are so old and irelevant that they do more harm then good. 3dsmax-02 is also a problem because 3dsmax's performance is so CPU-bound without a vendor specific driver plug-in (e.g. maxtreme).

dvornik
12-25-2003, 04:00 AM
So... Which part of Specviewperf IS relevant for DCC in your opinion? Are proe and ugs relevant to us? (I agree about the max part and don't know anything about the rest).

The first paragraph went over my head.

CgFX
12-25-2003, 05:08 AM
Originally posted by dvornik
Are proe and ugs relevant to us? (I agree about the max part and don't know anything about the rest).


I do think a large CAD benchmark like UGS is relevant to DCC when using large models and large scene blocking and layout.

There is also a new SPECapc benchmark for Maya, although that is currently focused on fixed function graphics features and doesn't differentiate between cards that are doing more work (128 bit FP vs. 32 bit Integer). There needs to be an apc benchmark in the future that factors in the programmable shaders aspect and the higher precision of the workstation cards that have come out over the past year.

3Dfx_Sage
12-25-2003, 05:36 PM
Cg-FX, I think that, for once, you are right. On newer nVidia drivers there is little to no difference in games. However, those numbers were back in the det 2x.xx days. I havn't really ran any benchmarks recently, but as soon as I get ahold of a decent system I am going to.

Might end up having to run the benchies on an Athlon64 3200+ rather than a dual Xeon like I was hoping :(

CgFX
12-26-2003, 01:35 AM
Originally posted by 3Dfx_Sage
Cg-FX, I think that, for once, you are right.
I am always right. Sometimes I just come up short in my effort to get you to understand that.

Happy Holidays! :wavey:

3Dfx_Sage
12-26-2003, 03:22 AM
<the ancient Sage begins preparations for a spell that will summon up an army of engineers, programmers, and hard-core enthusiasts to stand by her side...>

Now why didn't I think of doing that before? :surprised

edit: spelling :p

GregHess
12-26-2003, 02:01 PM
CgFx isn't always right....

What do you mean "isn't always?"

Try to imagine all life as you know it stopping instantaneously and every molecule in your body exploding at the speed of light.

Ya so sometimes, just sometimes :).

Happy Holidays.

:beer:

CGTalk Moderation
01-15-2006, 01:00 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.