Weird Vcore problem on cold XP-M
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
Weird Vcore problem on cold XP-M
I have the Athlon Mobile Barton XP2600+ (what a mouthful!) in a Abit NF7-S board (BTW, a great board!). The purpose of this setup is to run passively cooled. To acieve this end, I run the CPU at 200x5=1GHz @ 1.1V Vcore. This works fine and system is completely stable and reasonably cool (well, not too hot anyway).
Now the problem. If the CPU is cold, let's say, stayed off overnight, the system cannot be switched on. The board emits the "siren" sound and refuses to boot. I gather that the CPU is unstable and crashes the BIOS startup or something to that effect. Can someone confirm what this sound from Abit really means?
So, I clear the CMOS, start the CPU at 1.4 V Vcore, run 1/2 hour Prime95, reboot, set Vcore to 1.1V - voila! It all works again.
Any clue as to what this weirdness with cold and hot starts is about?
Now the problem. If the CPU is cold, let's say, stayed off overnight, the system cannot be switched on. The board emits the "siren" sound and refuses to boot. I gather that the CPU is unstable and crashes the BIOS startup or something to that effect. Can someone confirm what this sound from Abit really means?
So, I clear the CMOS, start the CPU at 1.4 V Vcore, run 1/2 hour Prime95, reboot, set Vcore to 1.1V - voila! It all works again.
Any clue as to what this weirdness with cold and hot starts is about?
Who would think that a CPU was so much like a car? This makes me wonder when they will introduce the injection CPUs
Right, the idea is to run it 24 hours a day anyway, so it is not a big problem. Thanks for confirming it.
The CPU will start reliably at 1.35 V but will fail at 1.3 V when cold. When hot, it has no problem starting at 1.1 V.
On the heat side, measured with the motherboard sensor:
1.35 V, 100x5=500MHz tops at 61 C
1.1 V, 200x5=1000MHz tops at 51 C
The case temp is about 40 C, using Zalman CNPS-6000 (?) without a fan. Tested with running Prime95 for a few hours.
Apparently, the voltage has a lot of influence on the wattage...
Right, the idea is to run it 24 hours a day anyway, so it is not a big problem. Thanks for confirming it.
The CPU will start reliably at 1.35 V but will fail at 1.3 V when cold. When hot, it has no problem starting at 1.1 V.
On the heat side, measured with the motherboard sensor:
1.35 V, 100x5=500MHz tops at 61 C
1.1 V, 200x5=1000MHz tops at 51 C
The case temp is about 40 C, using Zalman CNPS-6000 (?) without a fan. Tested with running Prime95 for a few hours.
Apparently, the voltage has a lot of influence on the wattage...
Well, that's the interesting part. The voltage seems to have much more influence on the power than I thought. That's why I am so keen on dropping the voltage as low as I can.AZBrandon wrote:Voltage does play a major role, but by my calculations, the power consumption should still be 33% higher at 1ghz/1.1v than 500mhz/1.35v.
Right now, running at 1GHz @ 1.3 V the CPU tops at around 70 C with Folding running. Today is a rather cool day so the ambient may raise another 10-15 degrees on a hot day. That would make the motherboard sensor 85 C and the internal CPU temp would be around 90-95 then. A bit too hot, would not you say?
Anyway, so raising the CPU Vcore from 1.1 V to 1.3 V at the same CPU speed gives a massive increase in temperature (because of fanless cooling). The temperatures jump:
CPU from 51 to 70 C
Case from 40 to 46 C
Have you tried using 8rdavcore. I had exactly the same experience http://www.silentpcreview.com/article164-page1.html but use that prog to get around the problem. I get it to boot at 1.275V, run 8rdavcore at start up and drag the Vcore straight down to 1,1V.
Mark
Mark
You said you were able to run it at 1.1v, 200*5 to top at 51C. Curious what if another say 20% or so drop in speed at the same voltagee would actually drop that much further.
About your linux delima have you tried freshmeat.net / sourceforge.net etc. There was a tool listed for just such a thing of dropping the voltage in linux via a driver.
http://freshmeat.net/projects/cpuspeedy/
About your linux delima have you tried freshmeat.net / sourceforge.net etc. There was a tool listed for just such a thing of dropping the voltage in linux via a driver.
http://freshmeat.net/projects/cpuspeedy/
It is really difficult to say. Dropping the frequency seems to help always. The CPU is very sensitive to the ambient temperature though now. At the moment, it runs at the same settings but at 57 C. I do not have the time to run an exhaustive set of tests at different frequencies and voltages now and the results from two different days do not seem to correlate at all. Maybe on Sunday I will try to run at various settings and plot a couple of graphs for the illustration of dependencies on the Vcore and frequency.Johnny Z wrote:You said you were able to run it at 1.1v, 200*5 to top at 51C. Curious what if another say 20% or so drop in speed at the same voltagee would actually drop that much further.
My problem is that I am a bit uncomfortable with the current setting of 1.1 V knowig that I would have to go and reset the BIOS should I need to switch off the machine. I figired out, actually, that there are 2 things it would not do below Vcore=1.3V: boot from cold and change the FSB. I have to set the Vcore to 1.4, reboot, change the FSB, reboot, set the Vcore back to 1.1V to make it work.
I also really like the current 200*5 setting for the frequency. The PC feels exceptionally fast for its "only" 1 GHz. When I drop the FSB to 100, the whole thing slows quite noticeably. I hope it works stable now in the hot days with these settings. When you change the multiplier, you change just the CPU but when you change the FSB, you change the speed of the whole thing 2 times. It feels marvellous. Man, I am dreaming. Never mind.
I am actually considering testing at the FSB of 133 and 166 to see if I get a much larger safety margin. The problem is that the board does not use the built-in CPU diode sensor but measures from under the CPU. It takes a very long time for the temps to stabilize. Ok, so maybe this Sunday...
Thanks for the tip. I tried it out but the Linux (2.6.5) does not seem to provide the information in those /proc/sys/cpu/0/* files, they are all 0. And the tool simply crashes on them. Besides, it seems to be able to control the speed of the CPU but not the Vcore. There is a couple more: cpudyn and cpufreqd, but they also seem to control only the frequency scaling. I think I may need to get down to writing this tool myself...Johnny Z wrote:About your linux delima have you tried freshmeat.net / sourceforge.net etc. There was a tool listed for just such a thing of dropping the voltage in linux via a driver.
http://freshmeat.net/projects/cpuspeedy/
I think I found a "sweet spot" of my CPU Apparently, both the frequency and the FSB stress the CPU (surprise, surprise ). Well, I was pushing hard to get it run at 200 FSB all the time and that was not a good idea. The result of this was:
- CPU heats considerably more
- the northbridge heats up noticeably
- CPU has troubles to start up at low voltage
After many cycles of experimenting, I found that the CPU will run rather cool at 166x6=1GHz and will start almost always at 1.2 V Vcore. This proves to be a very successful configuration for the moment. The CPU is now 48 C passively cooled and the case is 35 C while folding. Now I seem to have found a stable configuration and I am happy now
- CPU heats considerably more
- the northbridge heats up noticeably
- CPU has troubles to start up at low voltage
After many cycles of experimenting, I found that the CPU will run rather cool at 166x6=1GHz and will start almost always at 1.2 V Vcore. This proves to be a very successful configuration for the moment. The CPU is now 48 C passively cooled and the case is 35 C while folding. Now I seem to have found a stable configuration and I am happy now
Thanks for the testing was waiting to hear the results and they are interesting. I had wondered about the fsb being the voltage issue, guess you figured it out though.
What other fans did you happen to have in the case?
Yeah about your linux problem no clue though still. I'm still struggling with putting a linux from scratch system together my self.
What other fans did you happen to have in the case?
Yeah about your linux problem no clue though still. I'm still struggling with putting a linux from scratch system together my self.
Right, the system is as follows:
Antec 3700AMB, fan mounting brackets and guards removed
Zalman ZM-400B-APS
Papst 4412FL 12cm fan exhaust
Abit NF7-S, NB is cooled with Zalman ZM-NB47J
AMD Athlon XP-M 2600+
Zalman CNPS-6000Cu
Corsair PC3200 1GB
Geforce MX440
3COM 3c9xx (?)
HP CD-RW
Promise TX2 Ultra 133 IDE controller
CMD-649 ATA-100 IDE controller
Seagate Barracuda IV 60 GB
2x Seagate Barracuda V 80 GB
That's about it, I think...
The two biggest problems right now are:
1) The rear exhaust must be run at 12 V because otherwise the Zalman PSU speeds up. Papst is still quiet at 12 V but the next thing will be to try to duct the PSU and see how things will work out.
2) The hard drives. The airflow is very limited and they get rather hot. Now this is something that is not that easy to take care of...
(edit: forgot the cooler)
Antec 3700AMB, fan mounting brackets and guards removed
Zalman ZM-400B-APS
Papst 4412FL 12cm fan exhaust
Abit NF7-S, NB is cooled with Zalman ZM-NB47J
AMD Athlon XP-M 2600+
Zalman CNPS-6000Cu
Corsair PC3200 1GB
Geforce MX440
3COM 3c9xx (?)
HP CD-RW
Promise TX2 Ultra 133 IDE controller
CMD-649 ATA-100 IDE controller
Seagate Barracuda IV 60 GB
2x Seagate Barracuda V 80 GB
That's about it, I think...
The two biggest problems right now are:
1) The rear exhaust must be run at 12 V because otherwise the Zalman PSU speeds up. Papst is still quiet at 12 V but the next thing will be to try to duct the PSU and see how things will work out.
2) The hard drives. The airflow is very limited and they get rather hot. Now this is something that is not that easy to take care of...
(edit: forgot the cooler)