Vanilla Scalability

[paulcs]

Anand reviewed that weird blue Gigabyte GA-MG400 last week. It was the first major review site I've seen with benchmarks of the Vanilla 32 Mb board and a fast processor. Anand posted some remarkable numbers: over 74 FPS in 800x600/16-bit, using a PIII 733.

I upgraded one of my systems over the weekend and had a golden opportunity to test the G400 Vanilla I bought last July at multiple CPU speeds. I tried quite a few settings in Quake3, but I could not duplicate Anand's numbers. He was using the 5.50 Beta drivers, while I stuck with 5.41 and TurboGL, so that could explain part of it. (It's not clear if Anand used TurboGL or not. He recommended downloading them, but did not mention them in his specs.) My numbers often look like the marks Anand was getting with the Matrox 16 Mb Vanilla and the Gigabyte board.

It may be a settings issue. Maybe it's the drivers. I don't know. This board has always been a bit fussier than my Max, and that's why it's in a secondary system.

In any case, I used a PII 400 and a PIII 600 Coppermine at multiple front side bus speeds. Unfortunately, the PII wouldn't remain stable at 500 MHz. CPU speeds of 400, 468, 600, 672, 702, and 744 MHz were used. I'll list my specs after the benchmarks.

When benchmarking in Quake3, I generally start at the default "Normal" settings, increase the resolution to 800x600 or 1024x768, and manipulate the color depth and texture quality.

Sound was set to low and enabled. Vsync was disabled.

Just for fun, I also tested with Marks on Walls and Dynamic Lights disabled, as they can slow things down, and I think it's a good idea to disable them when playing online. Here are the numbers, sorted to demonstrate scalability:

800x600 16-bit
P2 400 - 43.2
P2 468 - 49.8
P3 600 - 57.4
P3 672 - 60.7
P3 702 - 61.7
P3 744 - 62.2

800x600 32-bit
P2 400 - 41.7
P2 468 - 46.2
P3 600 - 49.2
P3 672 - 49.7
P3 702 - 50.0
P3 744 - 50.1

1024x768 16-bit
P2 400 - 38.7
P2 468 - 40.3
P3 600 - 40.3
P3 672 - 40.6
P3 702 - 40.6
P3 744 - 40.7

1024x768 32-bit
P2 400 - 31.3
P2 468 - 31.4
P3 600 - 31.3
P3 672 - 31.5
P3 702 - 31.5
P3 744 - 31.6

800x600 16-bit, Marks on Walls/Dynamic Lights Disabled
P2 400 - 51.4
P2 468 - 59.1
P3 600 - 66.7
P3 672 - 68.8
P3 702 - 69.4
P3 744 - 69.7

800x600 32-bit, Marks on Walls/Dynamic Lights Disabled
P2 400 - 49.7
P2 468 - 54.3
P3 600 - 55.9
P3 672 - 56.5
P3 702 - 56.5
P3 744 - 56.6

1024x768 16-bit, Marks on Walls/Dynamic Lights Disabled
P2 400 - 44.4
P2 468 - 45.1
P3 600 - 45.0
P3 672 - 45.3
P3 702 - 45.3
P3 744 - 45.4

1024x768 32-bit, Marks on Walls/Dynamic Lights Disabled
P2 400 - 35.5
P2 468 - 35.6
P3 600 - 35.5
P3 672 - 35.7
P3 702 - 35.7
P3 744 - 35.7

Well, I hit a wall and Anand didn't. Not knowing the exact settings he used in Quake3, I'm not prepared to speculate as to why.

System specs:

G400 Vanilla @ Spec
PII 400 MHz (4 x 100 and 117)
PIII 600 MHz Coppermine (6 x 100, 112, 117, and 124)
AOpen AX6BC Pro II Millenium
256 MB Siemens CAS RAM
Turtle Beach Montego Xtreme (Vortex 1)

Overclocking stuff:

600E/256/1.65 V S1
90030007 - 0444 MALAY
SL3NA

I used Intel's CPU fan and a Fortron Source 300 Watt power. In addition to the power supply fan, I used three case fans.

Paul
paulcs@flashcom.net

[This message has been edited by paulcs (edited 07 February 2000).]

[Rags]

Hi Paul,

Is it possible that maybe you had dual head enabled when you booted your system before running the benchmarks. I realize you can uncheck extend desktop to second monitor, but unless you reboot, half your memory is allocated for the second display.....just a thought.

Rags

[paulcs]

Thanks Rags.

I've never enabled the dual head feature, although I checked it after you posted. (Poltergeists. They manipulate my power settings when I'm at work.) It's still disabled.

I have a Max on a 600 MHz machine I never overclock, and the Vanilla almost catches up with it at 700 MHz (800x600, 16-bit/16-bit). Then my Vanilla hits the wall.

The 1024x768 numbers, almost identical from 400 to 744 MHz, may indicate a malfunction of some sort. Again, this board was never the same after I overclocked it in July or August.

It should be interesting to see what the mysterious G450 will do in this system.

Paul
paulcs@flashcom.net

[CrazyBee]

Is it possible to re-flash the BIOS in a G400? Maybe that will "restore" it to how it was before..

AGP settings?

hmmm..

CB

[paulcs]

Hi CrazyBee. I got the board fairly early, and the sudden slowdown was the topic of a fair bit of conversation. Someone at Matrox was gracious enough to to send me a BIOS update, and then I flashed it again when the BIOS update on Matrox's site was released.

I forced AGP 2x at the request of someone with the same motherboard who was having problems. It's been running stable with this setting and an AGP aperture of 128. I might fiddle with this later.

To tell you the truth, I'm not sure anything is wrong. I have only Anand's benchmarks as a point of reference. There are people on the forums with Vanillas on machines running in excess of 700 MHz, but they all seem to have successfully overclocked their boards to Max-like speeds.

Paul
paulcs@flashcom.net

[Maggi]

Don't worry, Paul ...

It's PD5.50 being faster than PD5.41

But I wonder how Anand made PD5.50 working, because afaik do the Gigabyte G400 series require special drivers and are now at rev. 5.25 !!!

Maybe that just got luckky to get it working with PD5.50 ... don't know, but at least you - Paul - should try the 5.50s and report again your scores ...

Cheers,
Maggi

[paulcs]

Thanks Maggi.

At first, Anand had the same problem Sharky's reviewer did. It's been a few days since I read the piece, but I believe during his first attempt at installing the 5.50 drivers, he selected an option specific to the Gigabyte board. This resulted in all sorts of problems. He reinstalled and selected the Matrox G400 option, and these drivers worked for him.

I considered trying the 5.50 betas yesterday. However, I wasn't entirely sure the system would run stable with a 124 MHz FSB, and I didn't want to confuse potential problems associated with overclocking with problems related to the beta drivers. At that point, I needed to minimized the variables as much as possible.

I will try the betas soon and report back.

Paul
paulcs@flashcom.net

[paulcs]

Anand devoted a page to drivers stuff. Here's what he had to say:

"The GA-MG400 is a supported device in Matrox’s current driver release (v5.50.005), however if you select the GA-MG400 device you won’t be able to successfully boot into Windows.

"Matrox has identified the problem and say that it is most likely related to the fact that the MG400 does not feature the same G400 BIOS that is present on the other G400 cards. While they are working on a solution for the problem, there is an easier way to use the Matrox drivers with the Gigabyte card.

"Instead of allowing Windows to search for the best driver for your card, manually select the Matrox G400 AGP driver from the list of all drivers contained within the G400 driver set. A quick reboot later and you’re back in business, running with Matrox’s driver on your Gigabyte card. You will have to download the TurboGL driver separately and use that alongside the drivers to get the maximum performance out of your setup."

See that reference to TurboGL? That's what confused me about whether or not he was using TurboGL or the new ICD.

Paul
paulcs@flashcom.net

[paulcs]

I got home from work, uninstalled PD 5.41 and TurboGL, and installed the 5.50 beta drivers. I got a couple of hard lockups while running the Quake3 timedemo, so I forced AGP 1x (remember, I'm running on a 124 MHz FSB), and I was on my way.

800x600 16-bit/16-bit: 62.8
1024x768 16-bit/16-bit: 40.8

A modest improvement at 800x600, but a nine FPS increase at 1024x768. I then disabled the framerate hindering Marks on Walls and Dynamic Lights options.

800x600 16-bit/16-bit, Marks on Walls/Dynamic Lights Disabled: 70.2
1024x768 16-bit/16-bit, Marks on Walls/Dynamic Lights Disabled: 45.6

My pokey little Vanilla getting over 70 frames per second in Quake3. Who would have thunk it? Just for fun, I turned everything off in the Game Options menu, although sound was still enabled.

800x600 16-bit/16-bit, Game Options Disabled: 77.6
1024x768 16-bit/16-bit, Game Options Disabled: 50.5

Well, that's a little faster than Anand's scores. Again, I don't know what settings he used.

I know for a fact that the Vortex 1-based soundcard was slowing things down by a couple of FPS on my primary system. (I ran some benchmarks before and after upgrading to an Aureal Vortex 2 board.) Just as an aside, when I disabled sound, my scores actually went down.

G400 32 MB Vanilla Retail
PIII Coppermine 744 MHz (6 x 124)
AOpen AX6BC Pro II Millenium
256 MB Siemens CAS RAM
Turtle Beach Montego Xtreme (Vortex 1)

Paul
paulcs@flashcom.net