StarDestroyer.Net BBS

Get good parts.

Flash with the newest BIOS for your board. In my case (P5B deluxe) it was BIOS 0614.

Set the memory divider to the lowest available (1:1 aka DDR2-533, in the specific case of my board. You probably can't get a lower divider on any other board though). Then relax the timings a bit, to what's specced for DDR2-800 operation (I mentioned good parts, didn't I?)

Set the CPU voltage(VCore) to the max speced voltage (1.35 V). (This will be lowered back down once you figure out what's actually stable)

Boot into windows. Get Clockgen, Orthos 2004, and the speedfan beta.

Start up Orthos, and get it running a blend stress test. Start up clockgen. See if it recognizes the PLL on your board. If it doesn't, check the website to see if it has one listed. If that doesn't work, you get to reboot alot. Use speedfan to monitor the CPU temps. If it goes too high (too high being a variable amount depending on your cooling, ability to handle noise, and thoughts on how hot stuff should be (in my case, around 60 C)), you'll want to either get better cooling, or stop increasing the overclock.

About every ten to fifteen minutes, increase the FSB speed 10 MHz. If Orthos goes from green to red, the OC isn't stable, and you want to dial it back some. If the computer locks up, remember what the FSB speed was, and mark that as your high point. Dial back 10 MHZ and increase by a smaller amount, next time.

Continue this until you've found the max FSB.

Let Orthos run for 12 hours or so. If it fails at any point, you need to dial it back a bit, and re run Orthos. Continue dialing back until it stops failing. At this point, you have the highest stable overclock.


Now it's time to start reducing the VCore. This involves a bit of rebooting. Nothing for it, but at least it doesn't take as many as for the FSB. The point of this is to reduce heat and extend CPU life.

Decrease the VCore by a couple steps (4 or so). Boot into Windows. Continue until Windows fails to boot.

Increase the VCore a step. Boot into windows. If fail, increase again. Repeat until windows actually boots.

Run Orthos, again with the blend test for 12 hours. If it fails, increase Vcore and repeat.


This is the point where you activate windows, BTW.


Next, it's time to start messing around with the RAM timings. For this, you'll need a bootable copy of Memtest86+. Again, this will involve alot of rebooting. This is also a step I have not yet undertaken, so, no further comment will be made. It's a relatively low impact on performance anyway.

And people wonder why I don't OC.

Beowulf wrote:Get good parts.

Flash with the newest BIOS for your board. In my case (P5B deluxe) it was BIOS 0614.

Set the memory divider to the lowest available (1:1 aka DDR2-533, in the specific case of my board. You probably can't get a lower divider on any other board though). Then relax the timings a bit, to what's specced for DDR2-800 operation (I mentioned good parts, didn't I?)

Set the CPU voltage(VCore) to the max speced voltage (1.35 V). (This will be lowered back down once you figure out what's actually stable)

Boot into windows. Get Clockgen, Orthos 2004, and the speedfan beta.

Start up Orthos, and get it running a blend stress test. Start up clockgen. See if it recognizes the PLL on your board. If it doesn't, check the website to see if it has one listed. If that doesn't work, you get to reboot alot. Use speedfan to monitor the CPU temps. If it goes too high (too high being a variable amount depending on your cooling, ability to handle noise, and thoughts on how hot stuff should be (in my case, around 60 C)), you'll want to either get better cooling, or stop increasing the overclock.

About every ten to fifteen minutes, increase the FSB speed 10 MHz. If Orthos goes from green to red, the OC isn't stable, and you want to dial it back some. If the computer locks up, remember what the FSB speed was, and mark that as your high point. Dial back 10 MHZ and increase by a smaller amount, next time.

Continue this until you've found the max FSB.

Let Orthos run for 12 hours or so. If it fails at any point, you need to dial it back a bit, and re run Orthos. Continue dialing back until it stops failing. At this point, you have the highest stable overclock.


Now it's time to start reducing the VCore. This involves a bit of rebooting. Nothing for it, but at least it doesn't take as many as for the FSB. The point of this is to reduce heat and extend CPU life.

Decrease the VCore by a couple steps (4 or so). Boot into Windows. Continue until Windows fails to boot.

Increase the VCore a step. Boot into windows. If fail, increase again. Repeat until windows actually boots.

Run Orthos, again with the blend test for 12 hours. If it fails, increase Vcore and repeat.


This is the point where you activate windows, BTW.


Next, it's time to start messing around with the RAM timings. For this, you'll need a bootable copy of Memtest86+. Again, this will involve alot of rebooting. This is also a step I have not yet undertaken, so, no further comment will be made. It's a relatively low impact on performance anyway.

New BIOS out for P5B deluxe, supports crossfire now (but poorly compared to the 975x boards). Mine is OCed to 375MHz FSB, but it's probably capable of 400 (not playing w/ any 3D rendering yet, not really worth the bother of spending time to OC more- I'll bother when I start either doing that again (unlikely, no time), or upgrade my graphics card and throw in a more powerful CPU fan).

It seemed more stable actually when I cranked voltage down a bit (passed 15 hours of dual prime on torture on Linux which should be equivalent to Orthos blend which failed after 9 hours, while at 1.38V), though that's probably just luck- right now it's at 1.35V in the BIOS- interesting, according to you I might be able to lower it- I thought you had to increase it above 1.35 due to the motherboard actually sending a lower voltage than set...

On P5B, remember to lock the PCI Express and PCI busses to 100 and 33.

EDIT: Eh, lowered voltage and changed to 400FSB anyway, I'll do 10 minutes of orthos on Windows and play some TA Spring, then switch to Linux and dual prime it (I seem to have this peculiar problem where folding or prime will bog down Firefox but nothing else, even though they are at priority 19- maybe related to the fact that my firefox has the flash 9 beta or something).

Arthur_Tuxedo wrote:And people wonder why I don't OC.

Too many reboots? I think it only took me like 8. And most of those were in the voltage dropping part. It was a bit of a hassle, but not much, and I've got a 25% faster processor out of the deal, for effectively free. Such performance, if I'd bought a processor that was binned at that speed, would have cost 3x more.

Pu-239 wrote:New BIOS out for P5B deluxe, supports crossfire now (but poorly compared to the 975x boards). Mine is OCed to 375MHz FSB, but it's probably capable of 400 (not playing w/ any 3D rendering yet, not really worth the bother of spending time to OC more- I'll bother when I start either doing that again (unlikely, no time), or upgrade my graphics card and throw in a more powerful CPU fan).

It seemed more stable actually when I cranked voltage down a bit (passed 15 hours of dual prime on torture on Linux which should be equivalent to Orthos blend which failed after 9 hours, while at 1.38V), though that's probably just luck- right now it's at 1.35V in the BIOS- interesting, according to you I might be able to lower it- I thought you had to increase it above 1.35 due to the motherboard actually sending a lower voltage than set...

On P5B, remember to lock the PCI Express and PCI busses to 100 and 33.

EDIT: Eh, lowered voltage and changed to 400FSB anyway, I'll do 10 minutes of orthos on Windows and play some TA Spring, then switch to Linux and dual prime it (I seem to have this peculiar problem where folding or prime will bog down Firefox but nothing else, even though they are at priority 19- maybe related to the fact that my firefox has the flash 9 beta or something).

I haven't played with the new BIOS yet. It's stable right now. Maybe if I got a good reason to do so. Crossfire isn't. Yeah, the MB sends a lower voltage than is set. I've got mine set at 1.3V, but it's getting 1.26 +/- .004V. You'll note that I didn't actually raise any voltages out of spec for mine, which is why my OC is a good 35MHz FSB lower than yours was. The PCIe and PCI busses were already locked to 100 and 33. They have a seperate PLL loop to clock them, so they don't change when you change the FSB.

Beowulf wrote:

Arthur_Tuxedo wrote:And people wonder why I don't OC.

Too many reboots? I think it only took me like 8. And most of those were in the voltage dropping part. It was a bit of a hassle, but not much, and I've got a 25% faster processor out of the deal, for effectively free. Such performance, if I'd bought a processor that was binned at that speed, would have cost 3x more.

This is why I don't OC:

1. The time spent setting up, rebooting, and not being able to use your computer while it is stress testing.

2. The constant uncertainty. My computer just crashed. Was it a normal event or was it because of my OC'ed processor? How can I be sure?

3. Small gains in actual performance. Unless you do a lot of CPU-intensive stuff, there's no point, since most CPU OC's give precisely 0 extra gaming performance. GPU overclocks might show some extra FPS, but unless they allow you to put the settings noticeably higher, you probably won't notice.

4. It's not as cheap as people think. A perfectly good motherboard costs roughly $80. A perfectly good mobo that OC's well is more like $150. 2 GB of perfectly good RAM is $215, good OC RAM is $300. If you use aftermarket cooling, it gets even worse. A system that OC's well typically costs ~$200 more than one that doesn't. $200 is the difference between a 7900 GS and X1950 XTX. An otherwise equivalent system with an X1950 XTX at stock clocks will slaughter an OC'd one with a 7900 GS.

5. There's always a risk factor in OCing. You might go to all that trouble and expense only to find that your stuff didn't OC worth a shit. And it's always possible to fry, break or damage something, even if you're careful. Or it might work beautifully and rock stable for two years and then die just as the new gee-whiz components that you wanted to wait for are about to come out.

This isn't meant to be a diatribe against OCer's, as I can see why some choose to do it, but OCer's often act like anyone who doesn't OC is mentally deficient, and there are very good reasons not to OC.

Wot? No billion-degree fireball with an exajoule (that's ~240 megatons!) of pure unrestrained energy?!

Beowulf, you didn't OC it right.

Well, orthos seems stable for at least 3 hours on 1.32V, clocked up to 400FSB- maybe higher voltages aren't necessarily always better- we'll see

And Arthur, you do have a point- I skimped and went w/ a Radeon 9200 . Of course, I am waiting for DX10 cards, and the extra 50.2% CPU (assuming this OC at 3.2 works) is worth it for compiling/rendering (don't do either of those much now though, due to lack of time to fiddle w/ art, and compiles for my own small projects aren't that long anyway (I refuse to run Gentoo)).

And of course, bragging rights .

Improvements in 711 BIOS:

P5B Deluxe Release BIOS version 0711
1. Fix DRAM frequency incorrectly display problem under auto, standard and AI N.O.S mode
2. Add check ATI CrossFire VGA card routine
3. Add ASUS Cross Graphics Impeller (ASUS C.G.I) feature
4. Reduce system full reset of times
5. Patch some VGA card can''t found when NB Vcore value has been set too high
6. Fix DDR frequency 889 unstable by SPD
7. Enable support for NVIDIA 7950GX2 cards under recovery mode
8. Fix CPU frequency not correct when use ES CPU and lock free fuction enable
9. Fix memory display message not correct under multi-language
10. Add new CPU Ucode to support Kentsfield B3 stepping CPU
11. Fix chassis fan 2 speed not correct when fan speed too low
12. Fix ATI CrossFire can''t work when using with 800FSB Prescott CPU
13. Enhance Chassis Q-Fan sensitivity
14. Improve memory compatibility

Hrm, 8 might be related to why /proc/cpufreq is incorrect when speedstep is enabled. Too lazy though (no flash drive on hand) .

Beowulf, what was your final OC?

My E6600, P5W DH Deluxe and Corsair XMS DDR-800 hit 3.2 GHz CPU, 890 MHz RAM @ 4-4-4-12. The CPU was stable at 3.4 GHz, but it put out a bit much heat (it needed a good amount of vCore at that point; 3.2 is stable with just a slight increase).

Arrow wrote:Beowulf, what was your final OC?

My E6600, P5W DH Deluxe and Corsair XMS DDR-800 hit 3.2 GHz CPU, 890 MHz RAM @ 4-4-4-12. The CPU was stable at 3.4 GHz, but it put out a bit much heat (it needed a good amount of vCore at that point; 3.2 is stable with just a slight increase).

My final OC was 340x9MHz. All my voltages are at stock or lower (memory is 2.0V, because that's the spec on the RAM I got).

That's a very good OC for stock voltages.

Well, mine looks stable running dual mprime (I keep referring to this as prime95 for some odd reason, that's only the Windows version) for 7 hours at 400x8 @ 1.32v on torture mode - it did die before on Orthos at 375x8 @1.38 after 9 hours though, but it's run 15 hours of dual mprime at 375x8 @ 1.35 - I did change the RAM settings from 1.95 to auto though, so that's probably it.

At 400x8 @ 1.32v, temps seem to be 56C w/ TAT, and 53C on Orthos (unfortunately temperature sensors don't work on Linux). Improved a bit after I slid the HSF on the CPU around to squish the heatsink compound around rather than just squashing it and clipping it on.

EDIT: Well, it failed after 14 now... probably my northbridge and/or RAM, since it died under large FFTs.

EDIT: Changed FSB to 402, since according to http://www.thetechrepository.com/showth ... =32#post32 , the NB is most unstable between 360 and 400. Upped RAM to the maximum specced voltage of 2.1. Might need to increase voltage on the NB, but for now I'll leave it at auto and see if the FSB increase did anything.

EDIT: It's the CPU now that the FSB is bumped to 401, had to bump vcore to 1.35

Pu-239 wrote:At 400x8 @ 1.32v, temps seem to be 56C w/ TAT, and 53C on Orthos (unfortunately temperature sensors don't work on Linux). Improved a bit after I slid the HSF on the CPU around to squish the heatsink compound around rather than just squashing it and clipping it on.

Good readings; it's approximately exactly the same as my old Celeron (Northwood-class) and its malfunctioning heatsink underway at full power.

GG Plutonium.

Odd, orthos fails almost instantly when the voltage on the memory exceeds 1.9V w/ no other settings changed on the RAM, yet it is rated for 2.1V - this seems completely independent of CPU speed, which seems pretty stable just sitting on 1.32 (though w/ RAM >1.9V and crashing orthos, increasing voltage on the CPU consistently gives me 5 more minutes). Perhaps I need to update to the latest BIOS.

In any case, the system is stable after 10 hours of prime and fails after 15, so I'll leave it at 401x8@1.32vcore, 1.9vdimm @ 4-4-4-15 802MHz and test.

EDIT: Seems stable at the above settings after 11 hours on Orthos (I'll keep running til 24 if possible) - any explanation on why the RAM works better at lower voltages?

EDIT: Now almost 15 ... looks like I have the fastest proc here after Arrow (who has more cache), muahahah

... well, really only equivalent to Beowulf's thanks to his extra cache, and I'm lacking a decent GPU at the moment.

EDIT: 17 hours.

Well, it's been stable for 18 hours at 401FSBx8 (3.2GHz), 1.32vcore, 1.9vdimm @ 4-4-4-15 802MHz . Probably could go higher, but I don't really feel like bothering w/ many more reboots and tests.