Another Question ;-)

[CannyOne]

I want to overclock a G400-MAX. Should I perform the "memory optimizations" first? Or should I change dividers first?

In one of the other posts, DuRaNgO said, "you need to set the dividers, disable duty cycle, then o/c it." When he refers to "o/c" here is he talking about the memory optimizations?

Also I haven't applied any extra cooling solutions yet! So I don't think I want to try the settings listed in the other post. Does anyone have suggestions for some good starting settings, that are 'half way' safe?

[CannyOne]

Ok, "enhanced cooling" has been applied! And I think I understand what DuRaNgO was saying. But I'm still not sure if I should perform the "memory optimizations", before, or after, overclocking the core???

For memory settings I think I'll go with those recommended for the G400 in the PDF file. (Which are: 3,1,3,8,3,1,Fast,1,1,1,1.4,1.4) Except I'm not sure what to try for the "Memory Refresh Counter". No suggestions are given for this value, except "one step above or below" the standard value. I guess I'll just try one or the other. Hey, I guess this is all part of the adventure!

[anid]

Can't help ya with specific values, since I'm still stuck with a G200, but - you will get best results from the highest core/mem speed obtainable. After you see how high the card can go you can play with the mem ops and try and get that extra little bit out of it. As for the mem refresh, the lower the number, the faster the perfromance (I don't think it has a very significant effect, however). It is recommended to cahnge it because, in some situations, the clock settings can get reset to default - this provides a way of monitoring your OC and making sure that your card is, indeed, still OCed.

------------------
"I wrestled with reality for 27 years and I'm happy to say I finally won out over it."

[CannyOne]

Well, I "gave it a shot"!! I'm now running at 416.57 with dividers of 2.5,2,2.5. I optimized the memory as detailed above.

I then ran 3D Mark 99 MAX. My score, with a resolution of 1024x768x16-bit, went up from 6170 3D Marks to 6677. Fill rate went up from 257.7 Mtexels to 283.7 Mtexels. And the RasterMark increased from 2,503 to 2,743.

What should I try next? Just increase the "system pll clock" incrementally, or do I have to change the dividers first?

[dsp]

CannyOne,

You could change the system PLL if you want or change the dividers, if you have any level you want to go for, try them. I would suggest trying to increase the Memory speed a bit more but if you can, increase the core right along with it. Right now I'm running a G400 Non-MAX @ (see sig) so it would probably be better for you to increase the memory speed first.

dsp

------------------
AMD K6-2 350@400MHz
128MB Generic PC-133 RAM
13.6GB Maxtor 7200RPM Hard Drive (Yaaa!!)
Toshiba SD-M1212 6X DVD ROM
ACER 2X/2X/6X CD-RW
SB 128 PCI
Matrox G400 32MB Dualhead Vanilla @ 157/210/157
15" Daytek 1531D
PD 5.52
Win 98SE
10Base-T Ethernet card
DSL (Yahoo!!!!)

[Proton]

CannyOne, since you applied some special cooling to the G400 Max, I'd shoot for 450 with the dividers you're using. My Max without any additional cooling runs fine up to 435=174core/218memory clock. 450 is the maximum you can achieve with MGA Tweak AFAIK. I tried the suggested memory tweaks as well, but at that speed, I was pushing the memory already to its limits, which resulted in corrupted menus/mouse pointer in Q3A. In Q3A 1024*768 high quality settings except for trilinear my framerate went from 36,8 to 41,1. With your special cooling you should be able to reach 450 2,5/2/2,5(180Mhz core/225MHz memory). From my experience the memory is the limiting factor when overclocking the G400 Max. I tried 450 2,5/2,25/2,5(180Mhz core/200Mhz memory) to find out how high the core could go. At this setting I was able to run Q3A timedemo, but 3DMark2k crashed rather badly. There were no visual glitches at this setting, but the memory wouldn't go higher than 218MHz without some texture corruption. I think you attached some heatsinks to the memory, maybe it will do 225MHz.
Good luck and post some numbers.

Ole

------------------
My Sytem: P3 450@558, Asus P3B-F(Bios 1005), 128 MB PC-100(2,3,3,4, fast), G400 Max@174/218 (MGA-Tweak:2,5/2/2,5,435) PD 5.52wTGL1.30, 13GB Quantum Fireball CR, Pioneer 40*CD-Rom, SBLive!Value (Liveware 3.0)

[This message has been edited by Proton (edited 20 April 2000).]

[CannyOne]

Ok, guys! I'll disable the memory optimizations then bump the core some more. I was getting various artifacts in the rendering when I tried to go much higher than 418.5 with those "optimizations". It was rather disappointing. I guess I just didn't quite understand, that the memory thing was an auxiliary function. And yeah! I've been generating lots of "numbers" Maybe my OCR software can help shrink them into just one page!

Oh! and I think I have a "caffene hang-over". I should probably get some rest.

[CannyOne]

Well, at least I'm over the "Caffene Overdose"

I just replaced my power supply, a PowerMan FSP 300-60GT, with an Enermax EG-351P-VE. This may allow me to run stabily at higher settings because it is capable of sustaining higher "power" output along the 3.3v rail. However, these are the results at this point.

Highest "Stable" System PLL Clock - 437.40
Resulting in a:
Graphics Clock of 174.96
Memory Clock of 218.70
Warp Clock of 174.96
this is without SGRAM Memory Optimizations and No Miscellaneous Controls enabled.

Here are the results from testing with 3D Mark 99 Max.
- MGA Tweak version 1.20
- Matrox Driver Version 5.52
- under Windows 98 SE

Test 1 Test 6
Resolution 1024*768 1024*768
Color Depth 16-bit Color 16-bit Color
Frame Buffer Triple buffering Triple buffering
Refresh Rate VSync Off VSync Off
CPU Optimization AMD 3DNow! AMD 3DNow!

3DMark Result 6170.23 6720.51 3DMarks
Synthetic CPU 3D Speed 14850.26 14913.94 CPU 3Dmarks
Rasterizer Score 2504.82 2842.23 3DRasterMarks

Game 1 - Race 64.91 72.16 FPS
Game 2 - First Person 58.79 62.88 FPS

Fill Rate 256.83 293.88 MTexels/s
Fill Rate With Multi-Texturing 257.34 295.06 MTexels/s

2MB Texture Rendering Speed 278.56 306.66 FPS
4MB Texture Rendering Speed 227.32 248.67 FPS
8MB Texture Rendering Speed 155.12 168.31 FPS
16MB Texture Rendering Speed 117.37 126.33 FPS
32MB Texture Rendering Speed 56.86 59.13 FPS

Bump Mapping Emboss, 3-pass 79.90 87.69 FPS
Bump Mapping Emboss, 2-pass 106.21 116.96 FPS
Bump Mapping Emboss, 1-pass 198.71 218.91 FPS

PointSample Texture Filtering Speed 110.54 % 110.62 %
Bilinear Texture Filtering Speed 100.00 % 100.00 %
Trilinear Texture Filtering Speed 70.01 % 69.61 %
Anisotropic Texture Filtering Speed 0.00 % 0.00 %

6 Pixel/individual 2629.44 2632.01 KPolygons/s
6 Pixel/strips 3472.00 4017.53 KPolygons/s
25 Pixel/individual 1879.93 2012.12 KPolygons/s
25 Pixel/strips 2053.94 2246.24 KPolygons/s
50 Pixel/individual 1209.97 1306.97 KPolygons/s
50 Pixel/strips 1328.38 1454.41 KPolygons/s
250 Pixel/individual 501.72 549.37 KPolygons/s
250 Pixel/strips 526.27 576.84 KPolygons/s
1000 Pixel/individual 174.39 192.11 KPolygons/s
1000 Pixel/strips 186.33 205.39 KPolygons/s

That didn't work as planned! Basically, the first set of numbers is before the OverClock and the second set is after.



[This message has been edited by CannyOne (edited 22 April 2000).]