Welcome, Guest. Please Login 3dfx Archive
 
  HomeHelpSearchLogin  
 
Page Index Toggle Pages: 1
Send Topic Print
AAlchemy SX 8232SB Questions (Read 172 times)
sigtau
Junior Member
**
Offline


I'm Obsessed with 3Dfx!!!

Posts: 86
Houston, TX USA
Gender: male
AAlchemy SX 8232SB Questions
06.04.08 at 02:42:41
 
I'm the new owner of a Quantum3D AAlchemy SX PC-IG w/ a 8232SB card.  The unit looks and runs like it's brand spanking new.  It has dual P3 1 Ghz CPUs and 256 MB of RAM.  I've managed to get W2K updated and I loaded the final drivers for the 8232 that I got from Quantum3D. 

I loaded Quake 2 and Quake 3 and finally got them running after some tinkering.  I've done some benchmarking with these games and discovered something interesting.  The 8232 card runs much slower than I would have ever thought possible.  Benchmarking using Quake 3 "Four" demo @1024 x 768 yields only 34 fps.  This is slower than my Pure 3D II SLI setup on a P3 800 mhz test rig.  Dropping the resolution down to 800x600 yields about 56 fps.  Going down to 640x480 yields about 77 fps. 

It seems as though the fill rate of the card is limited to that of only one VSA-100 chip.  It doesn't matter which AA level is chosen, it's still the same performance.  I would have thought that this 8232 could at least match a V5 5500, which runs circles around it. 

So what am I missing?  Any ideas on how to boost the performance?  Is there some kind of registry hack or special combination of the control panel options that will increase the performance?

Thanks in Advance,
Ken
Back to top
« Last Edit: 06.04.08 at 04:48:43 by sigtau »  
 
IP Logged
 
RaverX
God Member
*****
Offline


Voodoo2 fan

Posts: 533
Oradea, Romania
Gender: male
Re: AAlchemy SX 8232SB Questions
Reply #1 - 06.04.08 at 05:12:08
 
I suspect your CPU limited, because Q3 uses only one CPU, you can force it by console command "r_smp 1", but there's a big chance that the system will freeze or crash after that when trying to run again Q3. But I think you can try it.

One more question, you sau you run 1024x768 ? 32 bit ? All maxed out (texture, geometry) ? Because if that's the case, don't expect too many fps in Q3, it's not a 3dfx friendly game, like UT.
Back to top
 
 
IP Logged
 
exxe
God Member
*****
Offline


^^

Posts: 1316
germany
Re: AAlchemy SX 8232SB Questions
Reply #2 - 06.04.08 at 09:50:49
 
Quote:
The big thing about AAlchemy is image quality- People see a board with 8 chips and assume it is going to be a speed monster. Performance is fair but where AAlchemy really shines is when Anti-Aliasing is enabled as it virtually has no impact on performance.


from http://www.thedodgegarage.com/3dfx/q3d_aalchemy.htm
Back to top
 

...
WWW 233274082  
IP Logged
 
sigtau
Junior Member
**
Offline


I'm Obsessed with 3Dfx!!!

Posts: 86
Houston, TX USA
Gender: male
Re: AAlchemy SX 8232SB Questions
Reply #3 - 06.04.08 at 19:00:03
 
I did more Quake 3 bench marking test with my V5 5500 AGP.  I didn't do a a super detailed thorough benchmarking test, just a quick look comparison.

I'm running the 5500 in a system that has a 2.8 ghz P4, Soyo P4S Dragon (SIS 645D chipset), 512 MB DDR SDRAM, and Windows Me.  I shouldn't have to worry about any CPU bottlenecks.  I also set it to use single chip rendering in 3Dfx tools.  I used medium settings for both geometry and textures.  Also did 32 bit rendering.  Here's what I found:

V5 5500 (single chip)
640x480 - 95 fps
800x600 - 63 fps
1024x768 - 39 fps
1280x1024 - 22 fps
(note: enabling "fasted performance" i.e. dual chips with no AA almost doubled these numbers on this test system)

Here are the numbers again for the 8232 w/dual P3 1 Ghz, Tyan S2567 mobo, and 256 mb PC-133 ECC SDRAM.  Enabling dual CPUs causes the system to hang, so no joy there.  I also ran it at 1280x1024 just to push it a little harder.

AA5 8232 (0 or 8X AA..makes no diff)
640x480 - 77 fps
800x600 - 56 fps
1024x768 - 34 fps
1280x1024 - 19 fps

My preliminary conclusion is the 8232 seems to perform much like a V4 4500 in terms of fill rate (This is a far cry from the fill rates quoted in the owner's manual).  However, I didn't actually dig out my V4 4500 to test that, but I'm sure the numbers will be similar to the non-SLI V5 numbers I quoted above. 

The small differences between the two could be because the single chip V5 is on the AGP bus and the 8232 is on PCI (which has been seen with  V4/V5 AGP vs PCI benchmarks).

I did my research beforehand, so I didn't expect twice the performance of a V5 6000 or anything like that.  There's very little info out on the internet concerning these Q3D systems.  I figured it would have at least been like a 5500 with no performance hit for AA.  The sad thing is my all of my 12 MB V2 SLI pairs will beat the 8232 by ~20% in raw performance running Quake 3.  This leads me to believe that some of Quantum3D's earlier products probably outperform the 8232 as well.  This is somewhat shocking considering the hefty price these units were fetching at the time (my card has a week 36, 2002 date on it). 

It still makes me wonder if there are any registry hacks that can modify the performance of this card.  It could also be related to the fact I had to cheat to get it to run openGL games in the first place.  I had to rename 3dfxOGL.dll to opengl32.dll and drop it in the Quake 3 folder.  I did try the wickedGL mini GL and it was slightly faster, but the image quality was very poor.  I also tried to rename the Q3D OpenGVS dll (as described in the owner's manual), but I didn't have much success there (Quake 3 crashed and Quake 2 sorta ran, but was very prone to crashing). 

Ken
Back to top
« Last Edit: 06.04.08 at 19:02:15 by sigtau »  
 
IP Logged
 
sigtau
Junior Member
**
Offline


I'm Obsessed with 3Dfx!!!

Posts: 86
Houston, TX USA
Gender: male
New Benchmarking Results w/ AAlchemy SX 8232SB
Reply #4 - 08.04.08 at 05:37:28
 
I must retract my previous statement about the 8232 card.  I did more tinkering and discovered a very interesting "feature" of the 8232 control panel.  All my previous tests were done with the AA setting set to 8x.  I assumed that when I hit the hot key to turn off AA, that the performance would change proportionally.  Since the performance didn't increase, I assumed the architecture was limited to produce quality, not quantity. 

Turns out I was WRONG!  If you set the AA mode to 0X or 4X in the driver control panel, performance will increase...significantly.  I reran the same benchmarks as I posted in my previous post above.  Since I'm only working with a 1 ghz P3, I skipped the lower resolutions.  I ran my V5 on my 2.8 ghz P4 rig in dual chip mode for a comparison.

V5 5500 (Dual chip mode, Quake3 "Four", 32bit, medium geometric detail)
1024x768 (0X AA) - 53 fps
1024x768 (4X AA) - 34 fps
1280x1024 (0X AA) - 31 fps
1280x1024 (4X AA) - 18 fps

Aalchemy 8232SB (Quake3 "Four", 32bit, medium geometric detail)
1024x768 (0X AA) - 71 fps
1024x768 (4X AA) - 63 fps
1280x1024 (0X AA) - 69 fps
1280x1024 (4X AA) - 42 fps

This is a significant increase in fill rate from my previous tests.  I experimented some with switching to 16 bit rendering and the best I get at 1024x768 was 71 fps (for both 0x and 4x AA).  1280x1024 gave me 69 fps for AA=0x, 65 fps for AA=4x, and 36 fps for AA=8x (just for fun).

Very Cool.  Most of these numbers are better than the V5 6000 benchmarks Gary has on his webpage.  This seems to disprove the myth the Quantum3D Aalchemy cards are all about quality and not much about speed.  It just took some tinkering to discover the tricks to balancing speed with image quality.  And clearly, I'm flirting with a CPU bottleneck that may be muddying up my results in the 60 - 70 fps range. 

I would also like to try the 8232 on a faster mobo/CPU combo.  Who knows how much these numbers will rise with a faster CPU to feed the transformation and lighting data to it.  I just need to do more research on compatible motherboards before I put this gem at risk.  I know my Soyo P4S Dragon is a safe board to run a V5 6000 on, so I would guess the same applies to the 8232SB.  I just need confirmation from someone who has first hand knowledge of these cards.

Ken   
Back to top
« Last Edit: 08.04.08 at 15:11:05 by sigtau »  
 
IP Logged
 
Page Index Toggle Pages: 1
Send Topic Print