Does downclocking *always* work ?

jones_r · Post by **jones_r** » Thu Aug 09, 2007 4:52 pm

My 7VM400M-RZ Gigabyte motherboard (for Athlon XP CPUs), has a jumper that switches between 133mhz bus, and 100mhz bus.

Now, assuming I buy a CPU which suppose to run at 266 (133x2), and I set the motherboard jumper to 100mhz, will I get 25% downclocking for sure? (which means 25% drop in TDP).

aaa · Post by **aaa** » Thu Aug 09, 2007 5:14 pm

Yes.

Undervolting helps alot more of course. Pinmod if the bios doesn't have an option.

jones_r · Post by **jones_r** » Thu Aug 09, 2007 5:28 pm

Pinmod if the bios doesn't have an option.

What is the "Pinmod" ?

My current 7VM400M-RZ motherboard, paired with Athlon XP 2400 (Thoroughbred), and no PCI cards or VGA card, takes 85W at idle (when bus is jumpered to 100mhz). Is there anything I can do to this system to make it idle at ~30-40watts ?

I thought about buying a Duron 600mhz, which us 27W TDP, but I'm not sure if this is enough to cut down more than 45W in total power consumption. Could it be that my motherboard simply uses extremely large amounts of watts ?

tehfire · Post by **tehfire** » Thu Aug 09, 2007 7:12 pm

A pinmod is shorting certain pins on the CPU in order to change things like multipliers, voltages, etc. You have to find a certain site with your exact processor to do one. I've never done it, but it's relatively common.

85W is pretty good. What is the rest of your setup? Can't really comment on the system until I know more about it.

jaganath · Post by **jaganath** » Fri Aug 10, 2007 12:45 am

will I get 25% downclocking for sure? (which means 25% drop in TDP).

25% drop in frequency will not necessarily drop power consumption by 25%, because leakage power is not affected by freq. changes. undervolting is most powerful way to reduce consumption.

jones_r · Post by **jones_r** » Fri Aug 10, 2007 3:55 am

85W is pretty good. What is the rest of your setup? Can't really comment on the system until I know more about it.

85W @ idle is good ?, my Pentium 3 system consumes 38W @ idle, which is much more to my liking.

My entire AMD system consists of:

1. 7VM400M-RZ motherboard
http://www.gigabyte.com.tw/Products/Mot ... uctID=1764

2. Athlon 2400 XP (Thoroughbred, AXDA2400DKV3C), with TDP of 68W, but I use the bus at 100mhz instead of 133mhz, so the TDP is lower (I'm getting 86W idle, instead of 100W idle, which could indeed say 25% reduction in CPU consumption, right ?).

3. One 40GB 7200RPM desktop hard drive.

4. Two 256mb memory modules.

5. Cheap 500W chinese PSU.

That's it.

I don't understand why with the same 40GB hard drive and chinese PSU, I'm getting only 39W @ idle with the Pentium 3 system, a 47W decrease !!!. What's taking so much power on the AMD system ?, even if I completely remove the CPU power consumption from the AMD system, I think that the system will still draw something like 65W (I'm assuming that at 86W idle, the CPU is still using something like 20W), and I'm not sure why so much...

jojo4u · Post by **jojo4u** » Fri Aug 10, 2007 4:17 am

Try http://www.stargaz0r.nm.ru/

jones_r · Post by **jones_r** » Fri Aug 10, 2007 4:23 am

What is that link ?

Ragarding the Pinmod,
My AXDA2400DKV3C CPU runs natively at 2000mhz (133x15), and needs 1.65V. Now, assuming I use Pinmod to get it to 1000mhz (133x7.5), what is the new voltage I should use, can anyone give me a clue ?

jones_r · Post by **jones_r** » Fri Aug 10, 2007 4:43 am

Try http://www.stargaz0r.nm.ru/

Holly cow, I downloaded this program and enabled bus disconnect, and idle power consumption dropped from 86W to 60W !. What is the meaning of this, that my system without CPU power consumption at all, draws 60W ???

Edit:
Ok, I think I was correct. When using Orthos, the power consumption went up from 60W, to 113W, which is 63W, very close to the published TDP of my CPU (which is 68W). This means: 1. this program shuts off the CPU entirely when it is not needed (I think), 2. My system draws ~60W, without factoring in CPU at all (this means that if I under voltage my CPU, it will have no further effect on idle power consumption), and my question is why do my system draws so much power without CPU ???? and can I make it draw less ?

jones_r · Post by **jones_r** » Fri Aug 10, 2007 8:43 am

YEAH !!!
I successfully undervolted my non-undervolting motherboard !!!

I decided that the pinmod is not going to give me enough downvoltage (maximum undervolting with the pinmod is 1.475v), so I decided to mess with the L11 bridge.

Using this excellent website:
http://fab51.com/cpu/barton/athlon-e23.html#L11 I was able to see that the configuration of my CPU's L11 bridge is "C:CCC" (meaning closed, open, closed, closed, closed), which translates to 1.65v.

My strategy was to do the most minimal change to the CPU. The most simple change seemed to be => "::CCC", which only required to open the first bridge (since the second bridge was already open). This new state suppose to give me 1.25v. Now, since I've already underclocked the CPU from 2000mhz to 1500mhz (using the simple 133mhz to 100mhz mobo jumper), I guessed that 1.25v for 1500mhz, should be ok.

Anyway, on the first try to open the first bridge of the L11 bridges, I used a knife, and applied wrong pressure, so the knife slid and created a big scar on the entire CPU. I thought game-over, but suprisingly the CPU still worked (but the first gate was still closed).

On the second try, I used a sharp needle instead of a knife, and pinched the first bridge with increasing force, until the needle sank a bit in the CPU. Afterwards I installed the CPU again, and HURRAY!, idle system power consumption dropped from 86W, to 62W !!!. Ah, the joy...

I then checked with Orthos, and max power consumption dropped from 114W to just 72W, WOW!!!.

And now the big surprise: I was sure that using S2kCtl won't make any change, since I already got 62W idle, and S2kCtl brought me to 60W idle before I undervolted, and I was under the impression that S2kCtl completely took the CPU out of the equation at idle. Boy was I wrong !, once loading S2kCtl, the system power consumption went to 49W !!!.

Amazing, my 1.5Ghz Athlon XP T-Bred system, is now 49W @ idle, and 72W @ max load, compared to default voltage state of 86W @ idle and 114W @ max load.

Now, I calculated that with the picoPSU and laptop drive, I should get to ~30W @idle and 49W @ max load. I couldn't ask for more. We're talking Pentium III power levels here, and my AMD system is light years faster than my Pentium III system.

jojo4u · Post by **jojo4u** » Fri Aug 10, 2007 8:47 am

Did you play around with the dividers in s2kctl?

Btw, s2kctl basically reduces you frequency by dividing it with the divider. Leakage current is only reduced by lowering vcore. So both method coexist quite nicely. The big saving is quite a surprise, but a Barton seems to have much leakage.

jones_r · Post by **jones_r** » Fri Aug 10, 2007 8:59 am

Did you play around with the dividers in s2kctl?

Nope, I just use defaults and did nothing else since it worked fine. Can I gain anything else by playing with the dividers ?

Btw, s2kctl basically reduces you frequency by dividing it with the divider. Leakage current is only reduced by lowering vcore. So both method coexist quite nicely. The big saving is quite a surprise, but a Barton seems to have much leakage.

But I have T-Bred, not Barton (it's not the same core, right ?). Anyway I guess you're right, and my T-Bred is horribly inefficient at 1.65V.

Vicotnik · Post by **Vicotnik** » Fri Aug 10, 2007 9:05 am

Nice work, jones_r!

What you could try next is to lower the multiplier so you won't have to lower the FSB.

jones_r · Post by **jones_r** » Fri Aug 10, 2007 9:09 am

What you could try next is to lower the multiplier so you won't have to lower the FSB.

I know.

Actually, 1.5Ghz is way to fast for this computer which suppose to act as file server. I can easily live with 800mhz or so, and this will allow me to further downvoltage.

But the thing is, I know I had luck today, and like in a Casino, I think a good gambler should know when to quit.

Juventas · Post by **Juventas** » Fri Aug 10, 2007 5:56 pm

I have a mobile barton and my biggest problem was getting a low enough vcore. Asus is all big on higher vcores for overclocking, so I had about 20 higher vcore selections available, but only 2 that were lower than default. It still runs, but it's actually higher than the cpu's spec, which undoubtedly makes it run hotter.

If you're underclocking I'd assume you'd have a similar problem. Namely any frequency that is lower than that line of processor normally starts at. Just my two cents.

Aris · Post by **Aris** » Fri Aug 10, 2007 8:39 pm

Dropping the frequency doesnt drop the power consumption. The only reason you underclock is because when you undervolt your new lower voltage isnt enough to support your current clock speed. So you underclock to improve stability at your lower voltage setting.

Voltage is ultimately what determines power consumption and subsequently heat dissipation.

jones_r · Post by **jones_r** » Sat Aug 11, 2007 4:39 am

Dropping the frequency doesnt drop the power consumption

But this is not true, when I drop the bus from 133mhz to 100mhz, the power consumption of the system drops from 103W to 86W.

Something else: Is dropping the CPU frequency from 2000mhz to 1500mhz, justifies a vcore reduction from 1.65v to 1.25v ?, do people usually reduce the CPU frequency more in order to cut 0.4v from the vcore ?, also, what is the best way to check for system stability ?

aaa · Post by **aaa** » Sat Aug 11, 2007 5:00 am

The less speed you use, the less vcore you can get away with.

prime95 torture test to check.

jones_r · Post by **jones_r** » Sat Aug 11, 2007 5:28 am

The less speed you use, the less vcore you can get away with.

Ok, but do you think that a drop from 2000mhz to 1500mhz, generally justifies a vcore drop from 1.65v to 1.25v ?, or is this vcore drop too much.

I think my question is in place, because if it was the other way around, meaning if I wanted to overclock a stock 1500mhz, 1.25v CPU, to 2000mhz, I would *never* would have increased voltage to 1.65v, that's insane.

prime95 torture test to check.

I used Orthos for like 15 minutes and the computer was stable. Is Prime95 better than Orthos in checking stability ?. I'm a bit worried because I get a lot of "Windows has encountered a problem and needs to close" alerts (like when browsing the internet, but not just). I'll try to reinstall the OS and see if it continues.

Edit:
I've read the pinmod guide again, and saw that in order to change the multiplier on my T-Bred CPU, I don't have to do any dirty work (i.e connect/disconnect bridges), rather just connect some CPU pins with wire (my T-Bred was manufactured before the 40th week of 2003, so multiplier shouldn't be locked, or that's what I understand). I will do that in order to cut CPU frequency to something like 800mhz, and see if the "Windows has encountered a problem and needs to close" alerts go away. It's weird though that Orthos reports everything fine, yet I get these problem alerts... maybe they're not related to the undervolting...

jones_r · Post by **jones_r** » Sat Aug 11, 2007 9:47 am

I succeeded to pinmod my T-Bred to 7X multiplier (from 15X), so now I'm at 933mhz (using 133mhz bus).

Orthos now runs better than before (with 1.25v at 1500mhz), before was stable too, but now each iteration of the test takes 1-2 seconds, and not 1-2 minutes (I didn't even notice it to be a problem before).

Still, downclocking to 933mhz (and I even tried 700mhz, after setting the bus to 100mhz), I was still not able to eradicate the "Windows has encountered a problem and needs to close" alrets. Moreover, I can't install a fresh OS, since during the process I get a blue screen of death.

So, to conclude, even though I was succesful in doing the downvoltage and reducing the multiplier, I was not able to make the system stable, which is a shame. I'll probably buy a new CPU, with lower TDP.

aaa · Post by **aaa** » Sat Aug 11, 2007 10:56 am

The difference between 'working but unstable' and 'not booting at all' is very small. So the fact you were able to boot up at 1500/1.25v means at least you are very close to the ideal voltage, and lowering to 933 should have eliminated the instablility. Problem is, instability sometimes corrupts Windows permanently, making the errors stay even after you get rid of the original problem.

jones_r · Post by **jones_r** » Sat Aug 11, 2007 11:14 am

The difference between 'working but unstable' and 'not booting at all' is very small. So the fact you were able to boot up at 1500/1.25v means at least you are very close to the ideal voltage, and lowering to 933 should have eliminated the instablility. Problem is, instability sometimes corrupts Windows permanently, making the errors stay even after you get rid of the original problem.

What you say makes sense, but at the current state, I can't install a fresh copy of the OS, since I'm getting an error in the beginning of the installation.

Do you think I should try lowering the CPU frequency even lower ?, say to 500mhz ?

aaa · Post by **aaa** » Sat Aug 11, 2007 11:52 am

If it still has problems then I'm pretty sure it's something other than the voltage/speed.

jones_r · Post by **jones_r** » Sat Aug 11, 2007 12:04 pm

Apart from this, the only thing I can think of is that scratch I did on the CPU.

FlorisNielssen · Post by **FlorisNielssen** » Mon Aug 13, 2007 10:54 pm

Maybe you could try to run Prime95 -> Torture Test.
If you get any errors from Prime95 (especially when you get them within a few minutes), it means the CPU has made a wrong calculation, and that means your voltage is too low for your frequency.
I also recommend you to run it for a couple of hours, so you know for sure.

But lowering the speed to 933MHz should solve te problem if your voltage was too low. And if it doesn't (which seems to be) I think the CPU is making wrong calculations anyhow (you should try Prime95 to determine that, I guess) and that indeed can be due to the scratch.

pliz22 · Post by **pliz22** » Tue Aug 14, 2007 5:35 am

My own experience has taught me that if you get BSODs when installing/ reinstalling then it is almost always memory problems. I mean your RAM is set too fast in the BIOS. Make the timings slower and giving it more voltage will probably help.

jones_r · Post by **jones_r** » Tue Aug 14, 2007 7:52 am

My own experience has taught me that if you get BSODs when installing/ reinstalling then it is almost always memory problems. I mean your RAM is set too fast in the BIOS. Make the timings slower and giving it more voltage will probably help.

1. I did not change anything regarding the memory, all I did was play with the CPU.

2. My motherboard doesn't even have an option to play with memory settings.

vortex222 · Post by **vortex222** » Tue Aug 14, 2007 10:36 am

1.25 volts for that CPU should be right in the ballpark. Some people even get them as low as 1.0 volt at ~1ghz.

I second the fault with ram thing. Try removing 1 stick of ram and see if that helps. Also if the motherboard allows for Latency settings to be changed, try setting the ram to Cas 3, (higher meens slower)

pliz22 · Post by **pliz22** » Wed Aug 15, 2007 5:27 am

[quote="jones_r"]
1. I did not change anything regarding the memory, all I did was play with the CPU.

2. My motherboard doesn't even have an option to play with memory settings.[/quote]

Well, my experience definitely isn't the be all end all anyway. Possibly clearing CMOS, and then putting your new voltages back could help.

So how is the heat anyway? At such low voltages can it run fanless?

jones_r · Post by **jones_r** » Wed Aug 15, 2007 5:49 am

Well, my experience definitely isn't the be all end all anyway. Possibly clearing CMOS, and then putting your new voltages back could help.

Getting the voltages back to normal means to connect the cut bridge, not something easy, and in any case, I have no wish to use the mobo/cpu at the original settings.

So how is the heat anyway? At such low voltages can it run fanless?

Yes, I think it can definitely run fanless (it's practically cold to the touch at idle), assuming you have a decent heatsink, which is hard to find for Socket A.

Does downclocking always work ?

Does downclocking always work ?