How to reach your max stable fsb quickly, without hosing your OS.
 

I've received some positive feedback after posting this on XS. I thought maybe one of our local members might benefit from this info too:

It seems to me like many of you people still use Prime95 / 3Dmark or any other windows based app to test stability while searching for the highest stable fsb. I'm not saying this is wrong, but there is a far faster way to find your highest stable overclock, without the chance of hosing your OS.

Step 1:
download memtest86 floppy or CD ISO image, and write the floppy/cd. Make sure "boot from cd/floppy" has a higher priority in your bios than your HD.

Step 2:
Overclock. :) Increase FSB by a few hz, change memtimings, whatever floats your boat.

Step 3:
Loading memtest86 only takes a few seconds. When it's loaded, select to loop test 5 (press "c","2","5","5","0","esc" i think). If it passes test 5 two or three times without generating errors, you've got a very, very good chance your system is stable. One loop should be a minute, depending on your ram size/speed. It only checks for memory errors, but experience tells if your cpu/chipset can't cope the high fsb, it will generate errors too.

Step 4:
As long as no errors where dedected, reboot and go back to step 2.

Step 5:
If you think you've hit the ceiling, boot from OS and start stressing with prime95 / 3D apps to ensure stability.


I've been using this method for years. It may not be the best method if our aiming for the highest 3Dmark/overclock (surely, your system won't crash instantly when memtest generates only a handfull errors, those can be corrected), but it's an excellent method to quickly find your optimal 24/7 overclock. If you got multiple motherboards to test on a limited time, this is by far the best method in my humble opinion. :)

I'm sure many of you guys already know this, but I haven't seen it posted anywhere. I've stumbled upon it by accident and found out test 5 was the hardest to pass, so I just want to share my knowledge.

1) A single memtest error can generate a LOT of windows stability problems. However test 5 isn't a too reliable test (as stated on the homepage) If you want to take some time, run by the entire test. Test 4 and 5 are most likely to crap out. I think you can identify the exact problem depending on the test it fails (be it chipset-speed or memory-timings) But I'd have to do more testing.

2) memtest, as the name implies, only tests memory; The CPU and other chipset subsystems aren't stressed at all. It can get you on the way to identify memory & basic chipset problems.
I had a SD module running 150 MHz stable in memtest, but windows load hangs within 5 seconds (on an old P3), it's definitly not the holy grail.

This is a way to quickly find your optimal overclock. It doesn't really matter what fails. Either it's stable or not. Normally you shouldn't overclock more than 1 thing at a time, so you should know what fails.

Experience tells me that if the CPU/chipset can't cope with the overclock, memory errors will be created. Which is only normal in my humble opinion.

I've tested this method with over a dozen motherboards since the KT266A. It never failed on me... but of course no method is flawless.


You don't have to do it this way... but someone who doesn't know any other decent method should try it. :)

I think it is a very good way, and better then booting XP and realising you clocked to high and b0rked the install :-/

bootable CD + 30min of testing gives you a raw estimate of the overclocking potential of PC system X

With fixed AGP-PCI speeds it's fairly accurate chipset measurement. Without the AGP and PCI stresses might cause problems. Even with fixed AGP-PCI there might be internal chipset timings that will cause problems.

Judging max CPU speeds is simply impossible with this tool. It might test some on-die cache but that's it.

Not that it's a bad thing, I frequently use memtest to determmine the max speeds of my system memory. But you shouldn't go expecting that the entire systems has to run perfectly just because memtest works (let alone only on test 5)

topic title "without hosing your OS."

is it the best way :)

"HOW-TO : reach your max stable fsb quickly, without hosing your OS."

I did not mention CPU overclock. ;)

It is also easy if you are playing with your memtimings, you can easily see which will have the best performance.
Memtest86 really rocks imo.

RichBa5tard, I've got some questions:
don't take this wrong but in my humble opinion, you're being a bit contradictive here.

Anyway, since fsb affects more then just the memory, wouldn't it make sence that if you relax your timings, and drop your cpu/dram ratio, that you'll actually be testing for the max stable fsb that your proc can handle, going by your experiences, or is this untrue, because the prog only tests the ram.

btw its "c","2","5","5","ENTER" , "0","TEST STARTS" "esc"
the esc = exit memtest


in my case , i never got errors ,
i got 1) everything is fine or 2) system freezes.


my stick 512 Mb twinmos 'mtec' 60B really sucks , only 190 mhz @ relax timings , no improvement 2.8V <-> 2.9V