Performance boost with 5.50 drivers!!! (better AGP)

[mkaresh]

In an earlier thread Big Lar posted some benchmarks showing that the Max gave quite a boost to benchmark performance. Unfortunately, he didn't use the same drivers with both cards. This led me to suspect that the new 5.50 beta drivers in themselves boost benchmark performance. When another revision was released (5.50.010), I decided to give the new drivers a shot.

Well, here are the results:

667mhz (133fsb, 89agp, 3-2-2 RAM):

3dmark 2000:
5.41 3157 (Max at 157.5/210)
5.50 3365 (Max at 156.6/208.8)
5.50 3385 (Max at 159/212)
Increase: 6.6%, 7.2%

3dwinmark2000:
5.41 38.4 (Max at 157.5/210)
5.50 41.6 (Max at 159/212)
Increase: 8%

720mhz (144fsb, 96agp, 3-3-3 RAM)

3dmark 2000:
5.41 3243 (Max at 157.5/210)
5.50 3349 (Max at 156.6/208.8)
5.50 3373 (Max at 159/212)
Increase: 3.3%, 4%

3d winmark 2000:

5.41 38.7 (Max at 157.5/210)
5.50 41.9 (Max at 159/212)
Increase: 8%

Details on settings follow discussion of results.

So there is a definite increase, at least with these benchmarks. ANYONE ABLE TO TEST THE DIFFERENCES WITH ACTUAL GAMES?

The reason for the increase becomes clear in the CPU speed rating with 3dmark 2000. At 667, this increases from 201 to 246. At 720, it increases from 212 to 247.

It appears that 5.50 is making significantly better use of AGP, hence the much larger impact of my 3-3-3 (vs. 3-2-2) RAM latencies at 720mhz with the 5.50 drivers. Also, the high detail helicopter test scores go from about 13fbs to about 20fps. Suddenly, my system performs as well at 667mhz as at 720mhz.

Note 1: AGP2x and sideband enabled with both tests. Tests run at default screen settings of XGA 16 bit with 3dmark and XGA 32 bit with 3d winmark.

Note 2: Aperture at 64MB for all but 720/159/212, where it ws set to 128MB. Setting the aperture at 128 enables the 64MB textures test in 3dmark to run. This tends to reduce the overall score by about 7 points, a negligible amount.

Note 3: I had to test the Max at two different speeds with the 5.50 drivers because MGA Tweak suddenly ceased to hold my former settings. I'd set the PLL at 315 as before, but when I checked again after the tests it had always kicked back down to 313.2. I have no clue why this happened. So I tested at this PLL and again at 318. 318 is right at the limit of my Max. Above that I get artifacts in one of the 3dmark 2000 test sets (Adventure).

Note 4: Save the “you need better RAM” flames. My Max isn’t totally stable at 720mhz (96mhz AGP) anyway. About 25% of the time it will lock up when going through these benchmarks. It seems to depend on how cleanly Windows loaded. I generally run my system at 667mhz, where it is rock solid.


------------------
PIII 500e @667
Soyo SY6BA+IV
Iwill Slotket II
G400 32MB Max
Nokia 445Xi 21"
and POS 14"
Aureal Vortex Superquad
128MB PC133
IBM DPTA 20.5GB 7200rpm
Toshiba 8/40 DVD

[mkaresh]

Addendum:

It's possible that the new drivers make better use of AGP 2x and/or sidebanding. These are always enabled on my computer, so I have no easy way to test this. Anyone want to run tests with these features enabled and disabled, in both 5.41 and 5.50? Hopefully you can run some actual games, and not just synthetic benchmarks like I did.

Remember, it's for a worthy cause. (Though which one I don't know.)

[mkaresh]

Actually, I'm not sure about my previous theories. maybe someone else can tell from the detail scores what the new drivers are improving?

In 3dmark 2000, most of the test scores don't go up by more than 1-2%, which can be explained by a slightly faster G400 PLL. The big changes are (133fsb, old at 157.5/210 vs. new at 159/212):

CPU: 201 vs. 246
Heli low: 66.0 vs. 70.1
Heli med: 42.0 vs. 48.2
Heli high: 13.1 vs. 20.9 (up 60% !!!)
Env. bump map: 94.5 vs. 100.1

Again, all other scores, including texture rendering, do not change significantly. So I'm not sure about the AGP theory. So what has Matrox changed with the new drivers that would affect these scores and these scores only?

[MaXi]

How do you get your MAX to be stable at a 133MHz FSB? Mine keeps quiting 3DMark2000 and so on?

[mkaresh]

I don't do anything really. It's actually totally stable up to 138fsb, and fairly stable up to 144. I can do 2d up to 150.

Some people talk about boosting the 3.3v voltage a bit. I can't do this manually, but I notice in my BIOS that this motherboard (Soyo SY6BA+IV) supplies 3.52 automatically. So that might be why I'm doing as well as I am. Some people are even stable up to 150, though I think with some AGP features disabled.

[OblivioN]

Mine is quite stable at 100AGP speed with agp x 2 and sidebanding enabled,it is stable in qIII and 3dmark as well.

I had to increase the IO voltage to get it stable at 100agp though I went from 3.3 to 3.5

[Neil]

Does this driver run with DX6.1?

[MaXi]

Hmm, never thought of rasing that voltage, but I'll try that for sure! Hope it works. Right now it's stable at 128MHz. I also looked at those 3DMark scores and mkaresh you got about 3365 using 1024x768x16 I guess? Using Pentium III? I get the same, but I've got a PIII650E clocked to 832MHz, and shouldn't I get more?

[mkaresh]

I'm not sure you should. I thought I saw scores in the 4000 range, from Big Lar I think. But I also notice that at 720 most of my detail scores are virtually identical to those at 667. So I get the sense that with the newest drivers the card is the limiting factor at 667, or even lower. If this is the case (I need to run some more tests) then more CPU will not raise your score.

[JmsAndrsn]

My G400 MAX will run 3DMark 2000 at 145 FSB on a BF6 w/500E. The highest score I have got at 1024x768x32B is 3591 with G400 clocked to 166/220. At 140 FSB I am able to score 3535 with the same settings. I cant run Q3 demo at 145. Can run at 140 FSB with about 83 FPS at 1024x768x32Bit

[mkaresh]

Those are excellent marks, even for 16 bit, the default. Sure you're at 32 bit? Unless you set in manually in the program, it will be 16, no matter what you have in the display settings.

How much voltage is your Max getting? Are you using the 5.50 drivers? CPU speed in 3dmark?