I'd be interested to see what the 3Dmark score looks like using the default settings. Running it with everything maxed out doesn't really give me a good frame of reference.

As for overclocking, yep, you're thinking right: leave the adapter card set to 100, and just switch the motherboard jumpers to 112. While on the subject, I forgot to mention before, but be aware that the 112FSB setting will give you a 37.5mhz PCI speed. Usually that's not a problem, but you can occasionally run across a PCI device that doesn't like it.

Old Thrashbarg wrote:

I'd be interested to see what the 3Dmark score looks like using the default settings. Running it with everything maxed out doesn't really give me a good frame of reference.

I should have mentioned it on another thread, but Thrashy's right. Cranking up the AA/AF and even resolution is testing the card more the rest of the platform in a lot of cases. This can lead to scores on different systems being much closer than they really should be, and should be only be taken for reference when comparing one graphics card to another.

If your trying to compare one system to another, best to leave the filtering off and stick to 1024x768 or lower.

OK, I have run some more benchies. I'm still using the Radeon 9800 Pro 128mb card, with the ATI control panel software settings all set to their default initial installation settings. The only change I have made is to set v-sync = "off". Also, at this moment the mobo is not OC'd. Memory used is 1 stick of pc-133 SDRAM. Finally, I'm using 2 PSUs simply because I've got 2 of them lying around and I may as well use them both just to help out with the Radeon card.

Pcpbench mode 100 (640x480, lfb) = 138.3 (All these pcpbench tests run inside windows 98 )
Pcpbench mode 103 (800x600, lfb) = 94.5
Pcpbench mode 105 (1024x768, lfb) = 67.4

3DMark 2001 s.e. (all settings on default, which means 1024x768, no AA, etc) = 8415
3DMark 2001 f.e. (all settings on default, which means 1024x768, no AA, etc) = 8338
3DMark 2000 (all settings on default - I simply clicked on the "default benchmark" button) = 8121
3DMark 99 Max (Freeware version, not Pro version. All settings on default - just 800x600, etc) = 8936, 18874

I rerun the 1st test again, just to see if the overall system performance has degraded after running these 4 tests above.
3DMark 2001 s.e. (all settings on default, which means 1024x768, no AA, etc) = 8433. (That's OK, no problems.)

This is a bit odd. I'm sure these scores (especially test number 4, the "8936" 99 Max score) beats my XP 2000+ rated VIA mobo system. Perhaps I need to mess about with changing its VIA 4-in-1 driver? I've read somewhere that if you change this driver for an older version, it can improve system performance.

I will now attempt to overclock the mobo. I will try 112 fsb. Incidentally, I found this article on the net about OC'ing a Powerleap based system. Duh Voodooman explains that when the mobo's fsb is increased, he has to increase the voltage settings on the Powerleap adapter.

http://duhvoodooman.com/powrleap/OC/CeleronOC1.htm

Edit: Also, after I overclock the 440BX based P2B mobo, I will set up the same components (Radeon graphics, PC-133 RAM) inside a VIA KT133A based system, and for its CPU, I will use something which has approximately 1400 mhz real clock speed. A bit like a "head to head" test, I suppose. I've got a t-bred 1700+ rated CPU I could use. That has a real clock speed of 1467mhz. I think that's close enough.

Before OC'ing the P2B mobo, I increased the Powerleap's voltage jumper by one "notch", from its default setting of 1.550-1.575v to 1.600-1.625v. I then set the P2B's mobo fsb jumper setting from its 100.3 fsb setting to 112 fsb (37.33 PCI). I switch on power, and the BIOS POST screen tells me that the CPU is running at 1567 mhz. Previously, it was 1402 mhz. Windows 98 boots normally. Before rerunning all of the tests listed in the post message above, I run Sandra 2002 Pro -

CPU = 4317, 2107. That's excellent.
Multimedia = 8331, 10107

3DMark 2001 s.e. (all settings on default, which means 1024x768, no AA, etc) = Immediately fails. Reboot time.

I think I will increase the Powerleap adapter's voltage jumper by another one "notch", from 1.600-1.625v to 1.650-1.675v.

Edit: I just noticed on the Asus P2B revision 1.12 mobo, there are 2 jumpers:

VIO
VCORE

They both have 3 pins. They are both in the 2-3 pin position. I wonder if I need to mess about with them, now that I have increased the mobo's FSB setting from its default value of 100, to 112?

Things seem to be going a *bit* better, now that the PL adapter's voltage has been increased to 1.650-1.675v

Sandra 2002 Pro runs OK. She tells me the CPU and multimedia scores OK. But...

3DMark 2001 s.e. (all settings on default, which means 1024x768, no AA, etc) = Fails on the nature test. Screen freezes. Reboot time.

On this webpage, at the bottom where it says "Overclocking, anyone?" -

http://duhvoodooman.com/powrleap/PLspeed2.htm

This guy is OC'ing his rev 2.0 1.4GHz PowerLeap. He was successful at 112 FSB. He only had to increase the PL's voltage by "one notch", to 1.600-1.625. He also successfully OC'd the mobo's FSB to 117 FSB. To do this, he increased the PL's voltage by another "one notch", to 1.650-1.675v. (That's what my current setting is now.) So, he managed 117 FSB @ 1.650-1.675, and I could not manage 112 FSB at the same PL core voltage. That's a pity. Maybe the Radeon 9800 card is not happy with the mobo's FSB increase?

Anyway, I decided not to increase the PL adapter's voltage setting again. I thought that increasing it by another notch, for a fairly modest 112 fsb overclock seemed a bit excessive. So, I left the PL adapter's voltage on "2 notches higher than default" = 1.650-1.675v, and instead I lowered the mobo's FSB setting from 112 to 105. Unfortunately, there is no option where I can adjust the FSB value in 1 mhz increments. The closest lower value to 112 on my mobo is 105.

Now, the BIOS POST screen tells me that the CPU is running at 1469 mhz. Unfortunately, no great OC there. In fact, it hardly seems worth it. Sandra 2002 Pro reports:

CPU = 4039, 1972
Multimedia = 7794, 9456

3DMark 2001 s.e. (all settings on default, which means 1024x768, no AA, etc) = At last, it does not fail. I get 8804. (It was 8415)

Edit: This is interesting. On that webpage link I gave earlier above ("Overclocking, anyone?"), he says right at the bottom of that page -

"It seems that the Voodoo5 5500 video card in the PC just wouldn't get along with 3DMark2001 SE, and the test kept crashing back to the desktop."

But, he has been using his V5 OK for all other tests, including 3DMark2001 SE. Strange. In other words, the V5 wouldn't work properly with the FSB OC'd, but it would work OK with the FSB not OC'd. That's the problem I am having, with the FSB OC'd + Radeon 9800. However, my system does not jump back to the desktop, it usually locks the machine and I have to reboot. But I guess it equates to the same thing: instability.

Edit 2: I could abandon the AGP Radeon card, and use a PCI card? Also, I just spotted 2 more jumpers on the mobo:

PCIRATIO
AGPFS

The 3 pin PCIRATIO jumper has been completely "soldered out", meaning that you cannot put jumpers on it.
The 3 pin AGPFS jumper is set to pins 1-2 = "on". The settings for AGPFS are -

1-2 AGPCLK = CPUCLK * 2/3 (Currently Enabled = "on")
2-3 AGPCLK = CPUCLK

I wanted to increase the mobo's FSB back up to 112 again, from it's current OC'd setting of 105. But I needed to make a graphics card change, because the Radeon 9800 did not work correctly at that 112 fsb setting. When I say "did not work correctly", I mean it failed on one of the "demanding" 3DMark 2001 s.e. tests. So, I replaced it with a "crappy" AGP voodoo 3 2000 model. I reran 3DMark2001 s.e. (again using default settings). This time it did not fail, but unfortunately a lot of the demanding tests were skipped, because the "crappy" V3 did not have the facility to perform some of these tests. The score I get is 1989, which is rather poor.

It is probably a better idea to run an earlier 3DMark benchmarking program. I will try '99 -

3DMark 99 Max (Freeware version, not Pro version. All settings on default - just 800x600, etc) = 6813, 20929

(For this '99 test, I ran a utility called V-Ctrl, in order to switch off the V3 2000's v-sync setting. When I ran it, V-Ctrl told me that the core clock was 166. Unfortunately, I don't know whether it has always been 166, or whether V-Ctrl has decided to overclock it.)

Before I remove the V3 2000 card, I run Pcpbench. Note: these Pcpbench tests were done before running V-Ctrl.

Pcpbench mode 100 (640x480, lfb) = 157.1 (All these pcpbench tests run inside windows 98 )
Pcpbench mode 103 (800x600, lfb) = 105.0
Pcpbench mode 105 (1024x768, lfb) = 75.5

These Pcpbench scores are good.

Now, I could either do -

1) Leave the FSB at 112, and remove the V3, and replace it with something more powerful (but not a Radeon 9800). Or
2) Leave the V3 in the mobo, and attempt to increase the FSB to either 115 or 120 or 124. (The next FSB OC'ing option on this mobo is 133)

If I decide to replace the V3, I could try a GF 4 Ti4200. Would that be a good choice?

I decided to leave the V3 in the mobo, and increase the FSB from 112 to 120. I left the PL adapter's voltage on "2 notches higher than default" = 1.650-1.675v. Now, the BIOS POST screen said that the CPU was running at 1679 mhz.

Sandra CPU = 4622, 2256
Sandra Multimedia = 8917, 10819

Pcpbench modes 100, 103, 105 tests = 162.5, 110.0, 79.4

So far, so good! But then this happened:

3DMark 99 Max (Freeware version, not Pro version. All settings on default - just 800x600, etc) = immediately failed.

I will reduce the FSB from 120 to 115, and rerun all tests...

Edit: I set the mobo's FSB to 115 (from 120). The BIOS POST screen tells me the CPU is now running at 1609 mhz

3DMark 99 Max (Freeware version, not Pro version. All settings on default - just 800x600, etc) = 6816, 20208

Sandra 2002 Pro CPU score = 4427, 2161
Sandra 2002 Pro Multimedia score = 8542, 10364

Pcpbench modes 100, 103, 105 = 152.4, 105.2, 76.3

3DMark 2000 (default settings) = 3229

3DMark 99 Max (retested, just to check system stability) = 6796, 20458

3DMark 2001 f.e. (default settings specially "downgraded" 16 bit version) = 1975

115 FSB appears to be stable. I will leave it at 115. I will now remove the V3 card, and replace it with a GF 4 Ti4200 card, and hope it works OK with the mobo's OC'd FSB.

Edit 2: I decided not to try the GF 4 Ti4200. Instead, I put the Radeon 9800 back in to the mobo. I left the FSB at 115. Unfortunately, the Radeon went mad at the desktop, so I had to decrease the FSB from 115 to 112 and try again.

3DMark 99 Max (default settings) = 10090, 20885

Pcpbench modes 100, 103, 105 = 157.0, 105.1, 75.6

3DMark 2000 (default settings) = Failed. It froze on one of the "adventure tests".

Now, I will definitely try the Ti4200. I will try it at 112 fsb to begin with. If that works, I'll increase it to 115...

More testing -

Mobo FSB: 112
AGP graphics: nVidia Ti4200 64mb (45.something driver)
Rivatuner setting: V-sync = "off"

3DMark '99 (default) = 10185, 20856
3dmark 2000 (default) = 9078
3dmark 2001 fe (default) = 8158
3dmark 2001 se (default) = 8229

Pcpbench modes 100, 103, 105 = 152.7, 102.8, 73.7

All seems stable. Next test to do - increase FSB from 112 to 115, and then redo all tests above...

The Tualatin chips did not respond very well to voltage increases. Usually they could be increased to their limit (around 1.6-1.7ghz) without any increase in voltage, and upping the voltage would often hinder the overclock. So it might not have been the fault of the video card when the test failed.

Edit: Ah, wait... y'know what though, the 9800 might've been one of the ones that was really picky about AGP speed. I know there were some cards that often didn't like anything above 70mhz AGP speed.

I couldn't be bothered with 115 fsb. I tried 120 fsb instead -

3DMark '99 (default) = 10263, 21040
3dmark 2000 (default) = Failed. It froze about 2/3rds the way through.

I then reduced the fsb from 120 to 115, and tried again. Also, I just read Old Thrashbarg's post above, suggesting that I reduce the PL's core voltage. I've just done that. I have reduced it by 1 "notch". It is now set to 1.6v, which is only 1 "notch" above its default setting. I reckon that should be OK.

3DMark '99 (default) = 9772, 20115 (That's strange. These 115 FSB scores are less than the 112 FSB scores.)
3dmark 2000 (default) = 8499 (Again, that's strange. This 115 FSB score is less than the 112 FSB score.)

I rerun the '99 test again = 9992, 20657. That's a bit better, but really these scores should be greater than the 112 FSB scores.

I decide to increase the PL's core voltage back up 1 "notch" again, and rerun these tests again -

'99 test = 9753, 20444

Well that didn't work - ie, the voltage increase made no difference at all. I will now reduce the FSB from 115 to 112, and see if that produces faster results! (Also, I have reduced the PL's core voltage back down 1 notch again.)

'99 test = 10321, 20880 (That's better!)

So there you have it: 112 FSB is faster than 115 FSB. That is, if you are using an nVidia Ti4200 AGP graphics card.

Edit: I have now set the PL's core voltage back to its default value. So, no increase in volts for it. I have also increased the FSB from 112 to 115. I run the '99 test again, and get 9895, 20366. Again, this is strange. (I increase the FSB from 112 to 115, and the '99 scores decrease.)

Tualatins overclock better on 815 mobos. At least that's what I've found with experience with my Abit BF6 + Slot T vs. ASUS TUSL2-C. I have a Celeron 1200 and a PIII-S 1400. Neither really respond to added voltage on the BX board, but on the 815 the PIII-S goes up to 1650 MHz with some extra volts.

440BX has subtle flakyness issues that show up as you go above 100MHz. The thing isn't designed for it. And I think that the Slotkets aren't good enough quality. It's all supposition but I think there are likely signal quality issues going on.