Deep Undervolt/clock PC w/ a 4.7W CPU!
-
- Site Admin
- Posts: 12285
- Joined: Sun Aug 11, 2002 3:26 pm
- Location: Vancouver, BC, Canada
- Contact:
Deep Undervolt/clock PC w/ a 4.7W CPU!
SPCR reader Mark created a Deep Undervolt/clock PCand ended up w/ a 4.7W CPU! Its clock speed ramps up/down based on load.
Last edited by MikeC on Thu Jun 03, 2004 1:40 pm, edited 1 time in total.
Good article, shows what's possible with the right tools and motherboards. My Palomino has such a low leakage current that I'm able to run passive when it's ideling. The T-Bred and the Athlon 64 are 130nm and have more leakage current, so dynamic underclocking is the way to go.
Via platforms will certainly need less power since the PSU and the power regulations are more effective.
"Gentlemen, start your AutoFSB"
Via platforms will certainly need less power since the PSU and the power regulations are more effective.
"Gentlemen, start your AutoFSB"
Low energy and gaming
With my setup you can have low energy and gaming, except not at the same time. It will play Halo @1024x768 with everything at default values.
Mark
Mark
I don't know what mobo you need to use a centrino on the desktop, but for desktop, a good candidate may be the new 1.35v Mobile XP projected to be in stock any day now at Newegg. It listed voltage is 1.35v at 1667mhz. There's a good chance it will do 1.1 - 1.15v or so at 1ghz. By my calculations, an XP should produce 14.5 - 16 watts and still be quite fast at that speed.Dirge wrote:Can anyone tell me what processor would be most suited to extreme underclocking/undervolting with the best results? Would a mobile or centrino processor give even better results?
I have not got all the AutoFSB settings handy but the key one is the Min FSB at 60MHz. You get this by dragging the slider in 8rdavcore down. The button does not work below 100MHz. I would try this in stages until you know what your set up will cope with.
Given the design of the Zalman CNPS5100-AlCu http://www.zalman.co.kr/eng/product/vie ... 2&code=005
I prefer to keep a bit of air running through it. Particularly as the PSU is much louder than the CPU fan at 1800 RPM. If the ambient temperature is low enough I can run it fanless. As I said in the article if I was to build now I would proabaly fit a Nexus AXP-3200 if I was to move the PC around as much as I do now. If it was not moved I would probably fit something like a Swiftech MCX462-V.
Depending upon ambient temperatures and your CPU heatsink I think it could be possible to run Fanless at 5x Multiplier, 133MHz FSB and 1.1VCore. You would just need to watch your temperatures.
I hope other members of the SilentPCReview Community will tell us what they manage to accomplish with their CPUs. I supect that there would be little benefit from changing your XP1800, so I would wait and see what other people say. It might be that the options available with motherboards (and 8rdavcore) are more important.
Mark
Given the design of the Zalman CNPS5100-AlCu http://www.zalman.co.kr/eng/product/vie ... 2&code=005
I prefer to keep a bit of air running through it. Particularly as the PSU is much louder than the CPU fan at 1800 RPM. If the ambient temperature is low enough I can run it fanless. As I said in the article if I was to build now I would proabaly fit a Nexus AXP-3200 if I was to move the PC around as much as I do now. If it was not moved I would probably fit something like a Swiftech MCX462-V.
Depending upon ambient temperatures and your CPU heatsink I think it could be possible to run Fanless at 5x Multiplier, 133MHz FSB and 1.1VCore. You would just need to watch your temperatures.
I hope other members of the SilentPCReview Community will tell us what they manage to accomplish with their CPUs. I supect that there would be little benefit from changing your XP1800, so I would wait and see what other people say. It might be that the options available with motherboards (and 8rdavcore) are more important.
Mark
My XP-M 2500 will drop to 1.1v @ 60Mhz FSB.
Unfortunately my NF7 board doesn't seem to like dynamically changing the FSB. It will drop just fine, but occasionally BSOD's while attempting to raise the FSB back up. Seems to be a fault particular to my motherboard; it did the same thing with an XP2100 CPU, yet many other people report good success with the program on their NF7's.
If you've got a motherboard on the compatibility list, I highly recommend trying the program out. It's very slick, and gets better with each revision.
Unfortunately my NF7 board doesn't seem to like dynamically changing the FSB. It will drop just fine, but occasionally BSOD's while attempting to raise the FSB back up. Seems to be a fault particular to my motherboard; it did the same thing with an XP2100 CPU, yet many other people report good success with the program on their NF7's.
If you've got a motherboard on the compatibility list, I highly recommend trying the program out. It's very slick, and gets better with each revision.
I have an A7N8X Deluxe and when I choose the A7N8X deluxe setting I get a repeating error message in a foreign language. I didn't read it too closely, but on the surface it looked German, perhaps.
Anyway, I used Nvidia's system utility to change FSB via windows before. It would work for a bit, and then, 5 minutes later the system crashes. Haven't experienced anything like that with this program yet. It seems to be quite a useful tool. I wish there wasn't such a long delay in throttling the clock. Also, it doesn't go from low to high clock instantly (after delay). It gradually increases the FSB. Takes a few seconds to eventually reach max clock speed.
Unless the reason is because I haven't played with the settings enough, I'd not like to have to use a 300 mhz system under those conditions.
Poorman's Cool 'n Quiet? I like it very much! Thanks for the link to the program!
One question...on the default settings...the minimum delay is larger than the maximum delay. How does this make sense? Also, how does the program choose which delay or what delay to use?
Anyway, I used Nvidia's system utility to change FSB via windows before. It would work for a bit, and then, 5 minutes later the system crashes. Haven't experienced anything like that with this program yet. It seems to be quite a useful tool. I wish there wasn't such a long delay in throttling the clock. Also, it doesn't go from low to high clock instantly (after delay). It gradually increases the FSB. Takes a few seconds to eventually reach max clock speed.
Unless the reason is because I haven't played with the settings enough, I'd not like to have to use a 300 mhz system under those conditions.
Poorman's Cool 'n Quiet? I like it very much! Thanks for the link to the program!
One question...on the default settings...the minimum delay is larger than the maximum delay. How does this make sense? Also, how does the program choose which delay or what delay to use?
mpteach
When runnning Word (120,000 word thesis) and Win98 CD player there no significant loss in useability at 300MHz. I do not tend to use the AutoFSB to increase the clock. I tend to switch off AutoFSB and drag the FSB to 133MHz manually, when I have done this Neverwinter Nights plays fine.
crypt0r
I have not had any BSOD that I can remeber. When I set the min FSB below 60MHz the system would crash. It might have been the particular settings you used.
I agree the delay settings do seem strange and the documentation for 8rdavcore is not very clear, but the setup described has been running stably at 300MHz for Word and 665 for games for a month now.
When runnning Word (120,000 word thesis) and Win98 CD player there no significant loss in useability at 300MHz. I do not tend to use the AutoFSB to increase the clock. I tend to switch off AutoFSB and drag the FSB to 133MHz manually, when I have done this Neverwinter Nights plays fine.
crypt0r
I have not had any BSOD that I can remeber. When I set the min FSB below 60MHz the system would crash. It might have been the particular settings you used.
I agree the delay settings do seem strange and the documentation for 8rdavcore is not very clear, but the setup described has been running stably at 300MHz for Word and 665 for games for a month now.
-
- Posts: 1
- Joined: Fri Jun 11, 2004 3:30 pm
- Location: Sydney, NSW, Australia
- Contact:
G'day - Just came across your article linked off OCAU (http://www.overclockers.com.au)
I currently have a standard 2500+ (AQYHA 0409) @ 3200+ @ 1.39V for every day use. Full system details here...
Haven't underclocked as such but @ stock speed it is prime stable @ only 1.17V!
Might have a crack @ the discussed methods & see how low she can go.
Good article.
I currently have a standard 2500+ (AQYHA 0409) @ 3200+ @ 1.39V for every day use. Full system details here...
Haven't underclocked as such but @ stock speed it is prime stable @ only 1.17V!
Might have a crack @ the discussed methods & see how low she can go.
Good article.
Wait, what is your cpu speed at when your playing nwn? That game, on high settings hits me harder than ut2k4pod03 wrote:mpteach
When runnning Word (120,000 word thesis) and Win98 CD player there no significant loss in useability at 300MHz. I do not tend to use the AutoFSB to increase the clock. I tend to switch off AutoFSB and drag the FSB to 133MHz manually, when I have done this Neverwinter Nights plays fine.
-
- -- Vendor --
- Posts: 31
- Joined: Sun May 23, 2004 12:01 pm
- Location: oregon
- Contact:
fanless
depending on ambient temp and airflow, 10w is a potentially fanless option. i have a 1.35w 2200+ with abit AN7 running at that rating next to me as i typeDirge wrote:pod03 could you post your AutoFSB settings for us to take a look at.
I would ideally like to build a fanless computer. Do you need a fan to cool the processor at 4.7w?
Can you run an AthlonXP fanless when it hits 5X multiplyer, 133mhz fsb and 1.1V vcore? I think thats around 10w power dissipation
the an7 stock chipset heatsink needs to be replaced, but i just have an old junky heatsink with the fan unplugged.
might still be better off with a fanless via system - but this is an intriguing option! anyone out there running this completely fanless, and what's the max speed you can reliably run fanless?
jef
A little update. I had reason to overclock (Rally Trophy - best quality). This is what I managed easily.
Removing the resistive cable so the CPU cooler could run at full speed and setting everything else at default values I managed 2000MHz (15x133) with a CPU temp of 42oC at 22oC ambient. Obvoiusly at about 5000RPM the cooler is not quite but is software adjustable. I strongly suspect it would have gone to higher speeds with higher temperatures, but this was sufficient for my purposes.
So a range from 300MHz to 2000MHz with software control.
Removing the resistive cable so the CPU cooler could run at full speed and setting everything else at default values I managed 2000MHz (15x133) with a CPU temp of 42oC at 22oC ambient. Obvoiusly at about 5000RPM the cooler is not quite but is software adjustable. I strongly suspect it would have gone to higher speeds with higher temperatures, but this was sufficient for my purposes.
So a range from 300MHz to 2000MHz with software control.
And 5100AlCu is not very high performing... so anything above 2000Mhz could make climb temps and voltage so much more than mhz.
Anyway, nice achievement and aproximation to super-extreme undervolting and underclocking and at the same time changing it without restarting, and I'm glad my article has been useful for your purposes.
I guess the stepping of your CPU is not very new. It can be even a JIUCB as the first one I used for my article. I guess that thing because of your need of high (1.275) start up voltage. You NF7-S has (almost) the same power regulation as KX7-333R or KR7A I used in my article, or even my current KD7A, and with all of them I have been able to boot up from cold my 1.5V JIUHB CPU with 1.125Volts.
Today, although a lot of Socket A CPUs are multiplier locked, there are a lot of new steppings that are able to achieve even more impressive undervolts with stock voltage (I have done some new tests and seen some very interestings results in this forum), but we are in Athlon 64 and Cool'n'Quiet times, which makes this automatically, and in a better way
Anyway, nice achievement and aproximation to super-extreme undervolting and underclocking and at the same time changing it without restarting, and I'm glad my article has been useful for your purposes.
I guess the stepping of your CPU is not very new. It can be even a JIUCB as the first one I used for my article. I guess that thing because of your need of high (1.275) start up voltage. You NF7-S has (almost) the same power regulation as KX7-333R or KR7A I used in my article, or even my current KD7A, and with all of them I have been able to boot up from cold my 1.5V JIUHB CPU with 1.125Volts.
Today, although a lot of Socket A CPUs are multiplier locked, there are a lot of new steppings that are able to achieve even more impressive undervolts with stock voltage (I have done some new tests and seen some very interestings results in this forum), but we are in Athlon 64 and Cool'n'Quiet times, which makes this automatically, and in a better way
-
- Posts: 21
- Joined: Wed Aug 25, 2004 10:24 pm
Successful passive cooling!
Here's my stats:
Antec 3000BQE
AMD Athlon XP 2800+
Zalman 7000A-Cu (should have gotten the Al-Cu)
512 PC3200 DDR Ram
epox 8rda3i (nforce2 400 ultra)
radeon 9000 passively cooled
8rdavcore!!!! The best thing to ever happen
I have my system set to normal when I boot up w/ the fan running... but I like to leave my computer to download overnight.... fans can be annoying cuz my comp's in my bedroom. So.. I extrememly undervolt it and unplug all my fans.
Stock fan speeds:
PSU: Around 2,000... loud as hell (at least to me)
CPU: Lowest Fanmate 1 setting: approx 1300rpm....
Case: 120mm Antec runs at approx 1000rpm
Vcore: 1.65V
Vdimm: 2.76V
Vdd: 1.8V
FSB: 166MHz
Multiplier: 12.5
MHz: 2.087 GHz
CPU Temp: 50C
System: 40C
HDD: 40C
Undervolt info:
Vcore: 1.1V
Vdimm: 2.5V
Vdd:1.6V
FSB: 60
Multiplier: 12.5
MHz: 750
CPU Temp (Zalman 7000A fan off): 49C stable (as i'm using it now actively for about an hour)
System: 34C
HDD: 37C
PSU fan: 1000-1100 rpm due to less power draw!
Only fan running is the ANTEC SmartPower 350 fan @ almost silent level. Once I Panaflo L1A it out.. should be completely silent!!!
WOW!!!! Significant difference!! Only problem is that you have to make sure that you raise the Vcore to 1.65V before reinstating the original preset settings, or else my system will reboot. Takes about 5sec to change!!! Couldn't be any better!
Now I just need to play around with the Auto FSB settings in order to get the best performance vs load. I think changing it manually is better for me until I get a fan controller to be able to turn on my CPU fan from the outside... can't have my cpu heat up and die on me.... have my MB set to turn off @ 65C... so i'm not too scared... any replies or comments would be great! Thanks
Antec 3000BQE
AMD Athlon XP 2800+
Zalman 7000A-Cu (should have gotten the Al-Cu)
512 PC3200 DDR Ram
epox 8rda3i (nforce2 400 ultra)
radeon 9000 passively cooled
8rdavcore!!!! The best thing to ever happen
I have my system set to normal when I boot up w/ the fan running... but I like to leave my computer to download overnight.... fans can be annoying cuz my comp's in my bedroom. So.. I extrememly undervolt it and unplug all my fans.
Stock fan speeds:
PSU: Around 2,000... loud as hell (at least to me)
CPU: Lowest Fanmate 1 setting: approx 1300rpm....
Case: 120mm Antec runs at approx 1000rpm
Vcore: 1.65V
Vdimm: 2.76V
Vdd: 1.8V
FSB: 166MHz
Multiplier: 12.5
MHz: 2.087 GHz
CPU Temp: 50C
System: 40C
HDD: 40C
Undervolt info:
Vcore: 1.1V
Vdimm: 2.5V
Vdd:1.6V
FSB: 60
Multiplier: 12.5
MHz: 750
CPU Temp (Zalman 7000A fan off): 49C stable (as i'm using it now actively for about an hour)
System: 34C
HDD: 37C
PSU fan: 1000-1100 rpm due to less power draw!
Only fan running is the ANTEC SmartPower 350 fan @ almost silent level. Once I Panaflo L1A it out.. should be completely silent!!!
WOW!!!! Significant difference!! Only problem is that you have to make sure that you raise the Vcore to 1.65V before reinstating the original preset settings, or else my system will reboot. Takes about 5sec to change!!! Couldn't be any better!
Now I just need to play around with the Auto FSB settings in order to get the best performance vs load. I think changing it manually is better for me until I get a fan controller to be able to turn on my CPU fan from the outside... can't have my cpu heat up and die on me.... have my MB set to turn off @ 65C... so i'm not too scared... any replies or comments would be great! Thanks
brianLp518
Have you considered changing the multiplier for overnight downloading. It could reduce heat and noise considerably and should only take a reeboot.
deebass
Yes thanks for the article.
Yes any thing more that 2000MHz would be hard work.
From my investigations (search the forums for pod03) the A64 might adjust automatically better, but it will not go low as MHz so not as low energy. So socket A might be better for the purposes of adjustable from low energy/noise to high performance.
Have you considered changing the multiplier for overnight downloading. It could reduce heat and noise considerably and should only take a reeboot.
deebass
Yes thanks for the article.
Yes any thing more that 2000MHz would be hard work.
From my investigations (search the forums for pod03) the A64 might adjust automatically better, but it will not go low as MHz so not as low energy. So socket A might be better for the purposes of adjustable from low energy/noise to high performance.
-
- Posts: 21
- Joined: Wed Aug 25, 2004 10:24 pm
multiplier to 5x
I followed your advice pod3.. and put my multiplier down to 5x making my comp @ 300MHz. Dropped another 3-4 degrees C. Thanks!
Well... that's super-extreme.pod03 wrote: deebass
Yes thanks for the article.
Yes any thing more that 2000MHz would be hard work.
From my investigations (search the forums for pod03) the A64 might adjust automatically better, but it will not go low as MHz so not as low energy. So socket A might be better for the purposes of adjustable from low energy/noise to high performance.
So that's why I think Athlon 64 is excellent for normal users with its Cool'n'Quiet feature.
I can handle up to 15-20W CPU without a fan (and even up to 40W without hearing my Papst fan spinning below 1000rpm), so anything below that is secondary for me against other things I should worry in a computer. Those other things include power consumption of other components (and nforce2 is quite power hungry) and I/O and PCI issues of Nforce2 I'm aware and scared of. So I'll stay away of 8rdavcore with pleasure (because no VIA Socket A chipset will ever be compatible)
...but that's my personal point of view. That's why I consider your article a very nice achievement and I expect anxiously for your Athlon 64 revision of your article (where nForce3 250 is up to my expectations of what a chipset has to offer, and VIA's alternative finally has AGP/PCI lock, i. e., independent of bus speed)
Another update.
I finally got round to getting a Seasonic Super Tornado 350. I have removed the case fan and blocked the opening to ensure air is drawn past the HDD. Things are really quiet now. Typical temperatures (degrees C) with 20 degree ambient, the PSU fan at minimum and the CPU fan at about 1500RPM are CPU: 29, System: 27 and HDD: 35. Next upgrade sometime, perhaps a SI-97 which would run passively, I am sure.
Mark
I finally got round to getting a Seasonic Super Tornado 350. I have removed the case fan and blocked the opening to ensure air is drawn past the HDD. Things are really quiet now. Typical temperatures (degrees C) with 20 degree ambient, the PSU fan at minimum and the CPU fan at about 1500RPM are CPU: 29, System: 27 and HDD: 35. Next upgrade sometime, perhaps a SI-97 which would run passively, I am sure.
Mark
Another update,
I had a new multimeter which measures AC current for Crimbo. Here are the figures I got for the system excluding the monitor.
Default ~ 0.36A ~ 86W
667MHz (Normal Boot - GFX underclocked) ~ 0.25A ~ 60W
300MHz (Normal run using word - GFX underclocked) ~ 0.22A ~53w
300MHz (Screen shut down) ~ 0.19A ~ 47W
I also noticed that the PC draws 0.05A (~12W) when plugged in but switched off. I was not surprised by this as the CPU temp is already quite warm if the PC is switched on after having been plugged in for a while.
Mark
I had a new multimeter which measures AC current for Crimbo. Here are the figures I got for the system excluding the monitor.
Default ~ 0.36A ~ 86W
667MHz (Normal Boot - GFX underclocked) ~ 0.25A ~ 60W
300MHz (Normal run using word - GFX underclocked) ~ 0.22A ~53w
300MHz (Screen shut down) ~ 0.19A ~ 47W
I also noticed that the PC draws 0.05A (~12W) when plugged in but switched off. I was not surprised by this as the CPU temp is already quite warm if the PC is switched on after having been plugged in for a while.
Mark
Last edited by pod03 on Fri Jan 07, 2005 4:07 am, edited 1 time in total.
Thanks Mark.pod03 wrote:Another update,
I had a new multimeter which measures AC current for Crimbo. Here are the figure I got for the system excluding the monitor.
Default ~ 0.36A ~ 86W
667MHz (Normal Boot - GFX underclocked) ~ 0.25A ~ 60W
300MHz (Normal run using word - GFX underclocked) ~ 0.22A ~53w
300MHz (Screen shut down) ~ 0.19A ~ 47W
I also noticed that the PC draws 0.05A (~12W) when plugged in but switched off. I was not surprised by this as the CPU temp was already quite warm if the PC was switched on after having been pluggesd in for a while.
One thing to keep in mind is that the numbers you're reading will correspond to VAs, not Watts. Since you're using a Seasonic Tornado w/ APFC, these values will be virtually identical (VA * PF = W; ST has a PF of .97-.99), but you should keep this in mind for future AC measurements.
Another update
I have now installed a mobile XP2500, a Thermalright SI-97 with Artic Silver 5 and a Vantec Stealth SF9225L 92mm Fan (I know … but is was the best 92mm fan that they had got where the SI-97 came from). This CPU will also boot at 1.275V at default speed (1862MHz). It is worth mentioning that I cannot the get the multiplier to change on the fly in Windows 98; it should be possible in WinXP if I (up?)grade.
Here are some figures:
System running Word unless otherwise stated. Video card underclocked 150/150MHz and Video card fan ~ 2600RPM unless other wise stated.
~0.22A ~ 53W CPU@300 MHz (5x60) 1.1Vcore, CPU fan ~ 1000 RPM CPU 22ºC – System 22ºC
~0.21A ~ 51W CPU@300 MHz (5x60) 1.1Vcore, CPU & VGA fan Off CPU 29ºC – System 29ºC
~0.25A ~ 60W CPU@665MHz (5x133) 1.1Vcore, CPU fan ~ 1000 RPM
~0.25A ~ 60W CPU@665MHz (5x133) 1.1Vcore, CPU fan lowest setting CPU 29ºC – System 30ºC
~0.24A ~ 58W CPU@665MHz (5x133) 1.1Vcore, CPU & VGA fan off CPU 37ºC – System 33ºC
~0.25A ~ 60W CPU@665MHz (5x133) 1.1Vcore, CPU & VGA fan off – Running Game CPU 39ºC – System 34ºC
~0.23A ~ 55W CPU@840MHz (14x60) 1.1Vcore, CPU fan ~ 1000 RPM
Video card at default 250/200 MHz
~0.29A ~ 70W CPU@1862MHz (14x133) 1.275Vcore, CPU fan ~ 1000 RPM Video Fan 3400 RPM
~0.32A ~ 77W Default @1862MHz (14x133) 1.45Vcore, CPU fan ~ 1800 RPM Video Fan 5300 RPM
I suspect that the reduction in current compared to the previous setup at the same speeds is mainly due to the Stealth fan being much more efficient than the Zalman. According to the specs it is not as efficient as the Nexus, but pretty close, it is also quiet enough when undervolted below about 1000RPM. I might get a Nexus fan next time I am putting in an order to somewhere that sells them.
I have also tried some different video cards, here are some figure that I obtained at windows desktop 1024x768, 1862MHz CPU, 1.275Vcore, CPU fan ~ 1000 RPM:
Connect 3D Radeon 9000 with Asaka GPU fan at ~ 2600 RPM 0.28A ~ 67W
Nvidia MX440 0.27A ~ 65W
ASUS Radeon 9600XT/TVD 0.31-0.32A ~ 75W
If the MX440 uses ~8W (http://forums.silentpcreview.com/viewtopic.php?t=7753) then it appears that the 9000 uses ~10W and the ASUS 9600XT/TVD uses ~ 16W. Of the three the Radeon 9000 appears the best performance during games per Watt in 2D.
Here are some figures:
System running Word unless otherwise stated. Video card underclocked 150/150MHz and Video card fan ~ 2600RPM unless other wise stated.
~0.22A ~ 53W CPU@300 MHz (5x60) 1.1Vcore, CPU fan ~ 1000 RPM CPU 22ºC – System 22ºC
~0.21A ~ 51W CPU@300 MHz (5x60) 1.1Vcore, CPU & VGA fan Off CPU 29ºC – System 29ºC
~0.25A ~ 60W CPU@665MHz (5x133) 1.1Vcore, CPU fan ~ 1000 RPM
~0.25A ~ 60W CPU@665MHz (5x133) 1.1Vcore, CPU fan lowest setting CPU 29ºC – System 30ºC
~0.24A ~ 58W CPU@665MHz (5x133) 1.1Vcore, CPU & VGA fan off CPU 37ºC – System 33ºC
~0.25A ~ 60W CPU@665MHz (5x133) 1.1Vcore, CPU & VGA fan off – Running Game CPU 39ºC – System 34ºC
~0.23A ~ 55W CPU@840MHz (14x60) 1.1Vcore, CPU fan ~ 1000 RPM
Video card at default 250/200 MHz
~0.29A ~ 70W CPU@1862MHz (14x133) 1.275Vcore, CPU fan ~ 1000 RPM Video Fan 3400 RPM
~0.32A ~ 77W Default @1862MHz (14x133) 1.45Vcore, CPU fan ~ 1800 RPM Video Fan 5300 RPM
I suspect that the reduction in current compared to the previous setup at the same speeds is mainly due to the Stealth fan being much more efficient than the Zalman. According to the specs it is not as efficient as the Nexus, but pretty close, it is also quiet enough when undervolted below about 1000RPM. I might get a Nexus fan next time I am putting in an order to somewhere that sells them.
I have also tried some different video cards, here are some figure that I obtained at windows desktop 1024x768, 1862MHz CPU, 1.275Vcore, CPU fan ~ 1000 RPM:
Connect 3D Radeon 9000 with Asaka GPU fan at ~ 2600 RPM 0.28A ~ 67W
Nvidia MX440 0.27A ~ 65W
ASUS Radeon 9600XT/TVD 0.31-0.32A ~ 75W
If the MX440 uses ~8W (http://forums.silentpcreview.com/viewtopic.php?t=7753) then it appears that the 9000 uses ~10W and the ASUS 9600XT/TVD uses ~ 16W. Of the three the Radeon 9000 appears the best performance during games per Watt in 2D.