It's time to overclock! What do I do?

Tim

Well, what I mean is what do I do without blowing the computer up. I've played around with overclocking a little bit, but always got nervous about it saying it had to change the BIOS. When I was done I'd put it back to stock settings.

Here's what I have to work with:

1. Abit NF-7 V.2 MB. A regular NF-7, no extra letters after the 7.
2. AMD Athlon XP 2500+ Barton core unlocked week 31 processor.
3. 2 X 256 MB sticks of Samsung PC3200 memory.
4. Generic 400 watt power supply.

Right now, everything is stock / default. I checked the BIOS and got these numbers:

2500+ 166 / 11.0 multiplier
AGP 66
Default power
CPU core 1.65 V
DDR SDRAM 2.6 V
Chipset 1.6 V
AGP 1.5 V
CPU FSB / DRAM Ratio - auto (whatever that is)
CPU interface - enabled
Memory settings - optimal 2-3-3-7

Stock speed is 1830 Mhz. I'd like to move it to 2.00, then if all is well, 2.10 or 2.20 or 2.25 Ghz.

Right now, with 100% cpu load on the folding program, the temperatures are CASE - 26.0 C, CPU - 44.0 C.

Ideas?

lemonlime

I would give it a whirl at 11x200 for 2.2Ghz (3200+ speeds).. Most Barton processors will do this without too much effort. You shouldn't need to up the voltage. If you run into any stability problems, you can kick it up a bit higher. Your temps seem quite reasonable, so I dont think heat will be a problem..

I don't have much experience with the NF7, but It should be able to handle that clock..

I'm not sure what your stock memory timings are at 200MHz FSB, but you may want to loosen the CAS to 2.5 before cranking the FSB up to 200MHz.. Once you are there, give it a try at 2 again..

You'll get aquainted with the CLR-RTC jumper real quick.. I think I can short those pins blindfolded now

Let us know how that goes

Tim

What is the CLR-RTC jumper and what does it do?

primesuspect

Tim, keep an eye on that power supply. A "generic" 400W may be a very big limiting factor in how high you can go...

lemonlime

Tim wrote:

What is the CLR-RTC jumper and what does it do?

CLR-RTC is also sometimes referred to as the 'clear CMOS' jumper. When you place the jumper in a specific position, it will restore the BIOS defaults. If you attempt to overclock too high, the system will sometimes fail to post (You will not even reach the BIOS). The only way to get back into the bios to correct the settings is to restore the BIOS defaults, through the use of the clear RTC jumper. Take a look at your motherboard's owners manual, you'll find more specific information there.

edcentric

IF you over-do it you may need to clear (reset) the bios in order to get the machine to boot.
1.Keep notes, all setting that you change from default need to written down.
2.Choose benchmarks to run. If you play one game a lot then use a clip from it for testing. A lot of us use F@H in order to test both stability and speed.

You will find that higher fsb/memory speeds will be well worth reductions in memory timing and maybe even not quite maxing out the cpu. Memory settings CAS of 2.5 or 3 are not the end of the world. One of my boxes runs 3-4-4-10 at 220 just fine.

madmat

On your board there is an alternative to clearing the CMOS if you fubar the OC, simply shut the power off (the fans will be runnining etc.) and restart it while holding down the Insert key on your keyboard.
Abit and several other mobo makers have started implementing that with their boards to avoid having to crack the case.
My IS7 worked with that, my SN41-G2 worked with that as well and I've read somewhere that the NF7 and NF7-S both work with it as well.
I don't guarnatee it'll work but it's worth trying before diving into your case to clear the CMOS.

GrayFox

madmat wrote:

On your board there is an alternative to clearing the CMOS if you fubar the OC, simply shut the power off (the fans will be runnining etc.) and restart it while holding down the Insert key on your keyboard.
Abit and several other mobo makers have started implementing that with their boards to avoid having to crack the case.
My IS7 worked with that, my SN41-G2 worked with that as well and I've read somewhere that the NF7 and NF7-S both work with it as well.
I don't guarnatee it'll work but it's worth trying before diving into your case to clear the CMOS.

Yah on my asus p4p800 all i have to do is turn the psu switch off then back on the bios resets to stock settings if it detects the system failed post last time

Tim

So far no one has told me what I need to know, which is something like "set the memory settings to XXX, and put the multiplier on XX and the bus speed on XX, etc. Then adjust the CPU voltage to XX if it is unstable.

I tried setting it to the 3200+ speeds of 200 X 11. Didn't want to work. After the POST beep, I'd get a screen saying a windows system file was missing or corrupted. Is that considered unstable?

I hit the reset button on the case, then hit delete to go back into the BIOS, and changed it to 166 X 12 from 166 X 11. Started up just fine. That gets me 2004 Mhz and the temperatures haven't changed much. Maybe a degree. I'm at 27 C case and 44 C cpu right now at 100% folding load.

How far can I raise the multiplier without changing / affecting anything else? If I jacked it to 13 or 13.5 or 14, what would be the first part to have problems?

Can I try raising the FSB to unusual numbers like 170, 175, 186, 194, etc? I mean numbers not divisible by 33, 66, or 100.

Doesn't 1 Ghz = 1024 Mhz? Or do we just always go by 1000 Mhz = 1 Ghz?

madmat

Leave the memory speed as set by SPD, lower the multiplier to 10 and raise the FSB to 200, set the vCore to user adjustable, set the vCore to about 1.7v (if memory serves me) and reboot.
You should be able to lock the AGP/PCI bus at 66/33 regardless of the fsb.

Tim

Ah, now that's the kind of information I need. Thanks. Anyone else got any ideas for different speed levels?

madmat

First make certain it's going to be stable at 2.0ghz and if it is move the mult. up to 10.5 or 11...if it becomes unstable add a bit more vCore which usually helps but keep an eye on your temps.
I wouldn't attempt raising the fsb too far past 200 with the ram you have seeings as it's more than likely not aimed at the enthusiast market but more likely mainstream judging by the numbers in the cas list you gave.
I'm pretty sure with the ram running at "by SPD" your cas numbers will end up being cas 2.5-3-3-7 or so but with the fsb increase your memory bandwidth will still be better than it is so that's not really an issue.
I've heard that good memory running at cas 2 levels on the NF7 series of boards will give better performance at 2-2-2-11 but to run straight 2's you'd need better ram such as Corsairs ultra low latency series of PC3200 which, if you're thinking of an upgrade, might be a good future choice. The 1 gig kit would give you a healthy boost in newer games like HL2 and Doom3.
You never mentioned what HSF you're running on there either, if it's the stocker I'd suggest getting a better model but since I'm an Intel kinda guy ATM I couldn't tell you what's the best choice to run, you might consider shooting Geeky1 a pm and asking him.
I also noticed Prime advising you to replace the generic PSU, that's deffinitely a good idea as clean stable power is a must-have for obtaining solid OC's and also the added stress of running your rig OC'ed might very well cause your "400w" PSU to bite the big one and sometimes that kills other parts of your rig as well, not a pretty situation.

Tim

I just put on an all-copper heat sink for the CPU. It dropped temperatures considerably over the stock aluminum one I had before. That's why I'm finally trying to raise the speed a little.

The current 166 X 12 is at stock voltages.

Yes, I'm going to replace the power supply soon, just need to find some extra cash.

Is changing to 200 X 10 less stressful on certain parts, or does it just plain run better somehow? I notice you said the voltage had to be increased to get the same overall speed level.

madmat

I said to raise the voltage because it's usually needed to promote stability, if it's stable at 12x166 it should be stable at the same speed with a higher bus speed. (10x199 or 10x200)
Going with a higher bus speed is better than a multiplier OC because it gives you more memory bandwidth.
Besides, the highest your 2500+ will allow you to go via strictly multiplier based oc's is 2075mhz because a 2500+ will only do 12.5 multiplier on the top end without work.

Tim

I tried resetting it to 10 X 200, but that didn't work. After saving the settings and getting the post beep, I'd get a black screen and message saying that the System32\Drivers\Ntfs.sys file was missing or corrupted.

Raising the VCore to 1.675, 1.7, and 1.725 didn't change that.

So I reset the voltages to stock / default, and changed it to 166 X 12.5.

That worked. I'm at 2088 Mhz now. Temps went up slightly to 28 C / 45.5 C.

I wonder why it won't work at 10 X 200? What's so special about those settings? Could it be that the memory isn't very good? Maybe I should try 180 or 190 FSB with a 10 multiplier and work up on that?

primesuspect

Probably your memory.

lemonlime

I agree.. It appears that your memory is holding you back. You did mention that you were running it at CAS2 latencies, which could be limiting you from taking the memory beyond ~166MHz. I would try to set the timings to something like 3-4-4-11, and start raising the FSB a few MHz at a time, until you find the ceiling. In order to do this, set the multiplier fairly low, (9 or 9.5). With any luck and loose timings, you may be able to get a bit higher than 200MHz. Once you appear to have found the highest FSB frequency, and have determined that it is running stable, start kicking up the multiplier at 0.5 intervals to find the CPU's maximum clock. Those temps you mentioned are still fairly good, so I would not be concerned about heat. Once you get your FSB and multiplier to the sweet spot, I would then try to tighten up the ram timings a little, and see if you can run them at a cas 2.5. I'm sure there is some more life in that chip give it a go

Also, once you get into windows, you'll have to verify the system's stability. Take a look for a program called 'Prime95' if you have not already done so. If you can run a Prime95 torture test for 24+ hours without any failures, then you should be good to go.

Tim

I just downloaded version 1.25 of CPU-Z, which seems to be an upgraded version of CPU ID that I've had for over a year now.

In the SPD section, when I click to see what each memory stick says, it tells me that the speeds for 166 Mhz are 2.5-3-3-7, and for 200 Mhz it is 3-3-3-8.

My memory sticks are in slots 2 and 3 on my NF7.

Perhaps I should reset my memory to 3-3-3-8? Or go with 3-4-4-11 if that's a more conservative starting point?

If the memory will handle it, what are the benefits of moving from 3-4-4-11 to 3-3-3-8? A slightly faster speed? Would it be anything significant?

I'll start with 3-4-4-11 and go from there.

Tim

I set the multiplier / FSB to 9 X 200. By setting the FSB to 200, the BIOS changes the memory timings from 2-3-3-7 to 2-3-3-8.

I set the memory to 3-4-4-11. Computer started up and worked fine.

I set the memory to 3-3-3-8 as specified by CPU-Z. The computer started and worked fine.

I tried it at 2-3-3-8. Error screen.

I tried 2.5-3-3-8. That worked, but to keep things a little conservative I then reset it to 3-3-3-8 and changed the multiplier / FSB from 9 X 200 to 10 X 200.

All is working good so far.

Tim

I've noticed that the computer isn't as fast with the 200 Mhz FSB.

When I had it running at 2088 mhz @ 12.5 X 166, it would fold a Tinker frame in 5 minutes 4 seconds.

With it running at 2104 Mhz at 10.5 X 200, it takes 5 minutes 33 seconds. That's with 3-3-3-8 and 2.5-3-3-8 memory timings.

Why is this? It was all tested on the same Tinker core, so it should be right.

lemonlime

I dont think the increased FSB will make much of a difference in Tinker folding. Anyone feel free to correct me if I'm wrong, but the overall clock speed should play a larger factor.

You may want to look at running the Sandra memory benchmark and the Sandra CPU arithmetic/multimedia benchmark as an overclocking reference point. I think folding timing may be somewhat inconsistant depending on WUs.