Quad Q8400S undervolt 0.925V 24W one fan need to be removed

[electrodacus]

I have a Q8400S is an engineering sample from eBay.
I will like to go lower than 2Ghz but it will not boot for lower than 1.8Ghz.

The CPU at 2GHz (6x333) and 0.925V is using 4W idle and 24W at 100% load.


Some measurement power consumption for complete system.
.
Old 275W SFF ATX [Watt] see http://electrodacus.com/photo/ATX/PS275W.jpg

*BIOS 42 W
**Idle 39.6 W
***100% all 4 cores 54 W

New DD-24AX V220-2 + 180W 24V AC adapter [Watt] see http://www.electrodacus.com/photo/ATX/24vatx-1.jpg

*BIOS 34.2 W
**Idle 31.2 W
***100% all 4 cores 46.8 W

you can see that the power supply makes a big difference.

Test computer components: see http://www.electrodacus.com/photo/ATX/power130W.jpg

MB Gigabyte GA-EG31M-S2
CPU Q8400S (65W TDP) underclocked @ 2GHz and undervolt to 0.925 V
PCI-E DVI adapter Silicon Image
RAM 4GB (2x2GB OCZ2G8004GK)
SSD 32GB Samsung MCBQE32G5MPP-0VA (max 0.42W)
80mm 7V Fan
Multi card reader

All power measured at wall AC 120V the monitor is not included see
http://electrodacus.com/photo/ATX/800-idlePS275W.jpg

* Power consumption when in BIOS
** Idle power in Ubuntu 9.04 (Linux)
*** 100% on all 4 cores using Blender 3D to render a scene using 8 threads.


My question is why is not possible to go lower with underclocking and if you know a better motherboard for power consumption this one uses G31 and is not possible to use passive cooling even if I added an additional heatsink about 120g additional to the original 49g aluminum heatsink. see the link (photo) from test system.
I still use one fan is the only moving part in this PC and I hope to eliminate can you give me any advice on this I relay hate dust and noise. If I do not use the fan then the temperature on northbridge will be at about 75C
Picture from the top

[smilingcrow]

I have no idea how much Blender 3D stresses a CPU and running using 8 threads might actually lower performance. I’d be curious to know how much the system consumes when running Orthos as there’s more data available to make a comparison.

So it boots at a FSB lower than 333 but not below 300! You might just be unlucky with your combination as I’ve under-clocked the FSB on recent Intel chips in the past but never tried with a quad.

[electrodacus]

I have no idea how much Blender 3D stresses a CPU and running using 8 threads might actually lower performance. I’d be curious to know how much the system consumes when running Orthos as there’s more data available to make a comparison.

So it boots at a FSB lower than 333 but not below 300! You might just be unlucky with your combination as I’ve under-clocked the FSB on recent Intel chips in the past but never tried with a quad.

I'm not a big fan of windows but I did some tests with prime 95 under win XP and also under windows 7 64bit version the highest power consumption for cpu is under windows 7 64bit is 2.03A on the 12V line (more precise 12.21V) and 1.94A on windows XP 32bit it may be the tolerance but I think the 64bit version of prime will stress the CPU harder.
The Blender 3D is the real life application that stress the CPU close to prime 95 and is 1.8A so the total system difference is max 3W additional if i use prime 95.
Hope this answered your question.

My system is very basic main loads are:
Q8400S CPU max 24W (underclocked to 2GHz and undervolt to 0.925V)
G31 with integrated video max 15.5W min 7.4W probably close to min in my tests since Blender 3D is not using the video card for render.
ICH7 max 3.3W
4GB RAM I have no idea not more than 2 or 3W I guess.
32GB Samsung SSD max 0.42W
And an undervolted fan max 1W.

some tests done with the same system but with the old power supply and a normal 3.5" Segate HDD using prime 95

[alecmg]

pretty interesting chart

Shows two things that I suspected before
1. New intel cpus are quite efficient in idle and you can't do much to tweak power consumption significantly lower at idle
2. At reasonable voltages frequency ceiling is good without horrible tradeoff in power. Between your Low and Medium settings you gain 50% more CPU performance(maybe more depending on memory freq) while sacrificing 50% more (total) power consumption at load and only 3 W at idle.

In my case, I run highest possible frequency at default voltage and activated CnQ.

[electrodacus]

pretty interesting chart

Shows two things that I suspected before
1. New intel cpus are quite efficient in idle and you can't do much to tweak power consumption significantly lower at idle
2. At reasonable voltages frequency ceiling is good without horrible tradeoff in power. Between your Low and Medium settings you gain 50% more CPU performance(maybe more depending on memory freq) while sacrificing 50% more (total) power consumption at load and only 3 W at idle.

In my case, I run highest possible frequency at default voltage and activated CnQ.

2. If you are looking at system power you are correct but I want a silent PC and the CPU is more efficient at 2GHz than at 3GHZ if you look at CPU power Is hard to cool passively even 24W.
I need the best performance per Watt but the worse part is North bridge. If I can find a motherboard that take less power will be great maybe on AMD platform but I guess they do not have a CPU as efficient /W as this one this is even more efficient then mobile Atom

[smilingcrow]

I need the best performance per Watt but the worse part is North bridge. If I can find a motherboard that take less power will be great maybe on AMD platform but I guess they do not have a CPU as efficient /W as this one this is even more efficient then mobile Atom

What actually happens if you run completely fanless! Does this system crash? Which temperature gets too high, the NB or CPU? What temps do you have for the NB and CPU at idle in fanless mode? What orientation do you have the case in? It looks upside down in the photo but maybe you have it laying flat!

The 3 series chipsets were never very efficient and the 4 series are a noticeable improvement so you might want to try a G41/43/45 board.

[electrodacus]

I need the best performance per Watt but the worse part is North bridge. If I can find a motherboard that take less power will be great maybe on AMD platform but I guess they do not have a CPU as efficient /W as this one this is even more efficient then mobile Atom

What actually happens if you run completely fanless! Does this system crash? Which temperature gets too high, the NB or CPU? What temps do you have for the NB and CPU at idle in fanless mode? What orientation do you have the case in? It looks upside down in the photo but maybe you have it laying flat!

The 3 series chipsets were never very efficient and the 4 series are a noticeable improvement so you might want to try a G41/43/45 board.

The case is small from factor and is laying flat.
If I run completely fanless the CPU si OK if run at 2GHz 0.925V or less ruining prime 95 I get max 66 Celsius a bit more than I prefer but is fine I use the AXP-140 heatsink as you can see in the photo.
The North bridge is getting to hot I don't know the exact temp but is much more than 60 C since I can not keep my hand on the heatsink but it never crashed.
the ICH7 south bridge is also hot.
G3x is hot but from spec G4x is a higher TDP than G3x can somoane confirm that G4x will be able to run passivity (no fan on CPU or case).

[smilingcrow]

If I run completely fanless the CPU si OK if run at 2GHz 0.925V or less ruining prime 95 I get max 66 Celsius a bit more than I prefer but is fine I use the AXP-140 heatsink as you can see in the photo.

I usually prefer to use RealTemp or CoreTemp to check how far the CPU is from Tj Max as that’s a more reliable method of monitoring CPU temperature.
If it really is 66C that means it should be 34 or 39C below Tj Max.

The North bridge is getting to hot I don't know the exact temp but is much more than 60 C since I can not keep my hand on the heatsink but it never crashed. the ICH7 south bridge is also hot.

My North Bridge maxes at 73C according to Speedfan which is fairly high but many chips are safe to 100C although ideally I’d like to keep everything below 70C.

G3x is hot but from spec G4x is a higher TDP than G3x can somoane confirm that G4x will be able to run passivity (no fan on CPU or case).

The 4 series is more powerful so when the IGP is fully loaded it will likely hit a higher consumption; you aren’t loading the IGP though are you?
From my experience and looking at the reviews the 4 series certainly seem to consume less power. Cheap G4x series boards may well have puny heatsinks so in a passive setting you may well need to upgrade the chipset heatsinks or at least replace the thermal compound.

[electrodacus]

I used RealTemp and Speedfan to measure temperature and is no more than 66C Tj for CPU and I think was no indication of temp for north bridge.
Usually I do not load IGP but I want to be safe even if the IGP will be used and I do not need a powerful IGP the x3100 that I have now is more than enough.
It will be great if I can find a board with a mobile north bridge.

Today I did some more power consumption tests maybe you are interested .
There are 3 major changes from the last setup.
1 power supply is now a 130W passive DD-24AX V220-2 + 180W brick
2 Old 3.5" 250Gb Segate HDD was replaced by a 2.5" 32GB Samsung SSD
3 I upgraded the RAM from 2x1GB 667 to 2x2GB at 800

Now I can go down to 1.2Ghz with CPU frequency and I guess the RAM made the difference.

[smilingcrow]

I used RealTemp and Speedfan to measure temperature and is no more than 66C Tj for CPU.

Do you mean 66C below Tj Max or an absolute temp of 66C?

I think was no indication of temp for north bridge.

Does Speedfan show any other temperatures as well as Core 1/Core1? They are usually labelled Temp 1, Temp 2 etc.
If so monitor them as you load the CPU and if one of them rises significantly it may well be the NB. The way to confirm is to add extra cooling to the NB and see if lowers the temp you are investigating.

What type of task are you doing with Blender when you measure power consumption?

[electrodacus]

Do you mean 66C below Tj Max or an absolute temp of 66C?

Is 66C absolute temperature.

Does Speedfan show any other temperatures as well as Core 1/Core1? They are usually labelled Temp 1, Temp 2 etc.
If so monitor them as you load the CPU and if one of them rises significantly it may well be the NB. The way to confirm is to add extra cooling to the NB and see if lowers the temp you are investigating.

Yes I see more temperatures but are not all labeled I see all 4 cores temperature usually the Core1 is at the highest temperature about 66C the other cores are at around 60 to 63C absolute temp.
There is no temperature indication for north bridge because all other temperatures are under 50C even if I do not use any fan and the northbrige is for sure more than 60C because the heatsink is to hot to touch. When I talk about speedfan or temperature measurements all this are done in windows using prime95
See this screen shot is in Windows 7 64bit fanless http://electrodacus.com/photo/axp140/axp140load.jpg
You can see that speedfan is at least 10C lower than real and in linux I use lm-sensors for temperature and is similar with realtemp about 10C more than speedfan.
My room is always hot in summer 26 to 30 Celsius
see also the idle temp http://electrodacus.com/photo/axp140/axp140idle.jpg

What type of task are you doing with Blender when you measure power consumption?

I render a 3d scene it will produce the same load on CPU as Cinebench on windows using all cores it will produce similar temperature and power consumption as prime 95. Prime 95 will produce more heat but not more than 2 - 3C and CPU power consumption will be max 10% more comparing with Blender or Chinebench

[electrodacus]

I did today a test to see the power consumption on all lines is not as bad as I expected I think last time a did this test I got bigger number on 5V because I had a HDD not an SSD.

[lobuni]

Wow the efficiency looks surprisingly high. Even too high.
Any idea how efficient are these things (AC-DC adapter and DC-DC psu) separately?

Specs for the DD-24AX V220-1 state 88%, couldn't find the ones for DD-24AX V220-2.

Did you also measure the AC voltage?
At these current levels the accuracy of this measuring device (UT203?) is playing a big role.

[khaakon]

I am very excited to read about your finds, awesome science and very well written and documented. 'No moving parts' - way to go, future here we come. I wonder about just one small (?) thing that is unclear to me, have you attached an additional heatsink to the existing NB heatsink ? If, so - why didn't you rather replace the original? And, have you tried to take off the original heatsink and reseat with some proper care, and proper thermal compound. That's no small thing, i have read about quite a few people who've done with great improvements in cooling.

This post could've been full of smileys..

[electrodacus]

Wow the efficiency looks surprisingly high. Even too high.
Any idea how efficient are these things (AC-DC adapter and DC-DC psu) separately?

Specs for the DD-24AX V220-1 state 88%, couldn't find the ones for DD-24AX V220-2.

Did you also measure the AC voltage?
At these current levels the accuracy of this measuring device (UT203?) is playing a big role.

Yes the efficiency is high The 88% in spec is the worst case probably at 100% load on all lines this power supply is generating 12V first out of 18 to 28V input with a efficiency of more than 95% then from this 12V it will generate the 3.3V and 5V also with good efficiency but you need to add also the efficiency of the 12V line so taking current from the 5 and 3.3V line is less efficient. The brick is surprisingly efficient and with no load it takes no current most of the brick I have will take AC current even with no output load but this new bricks are more efficient.
It may be that the UT203 is not so accurate at this small loads But I compared measurements also with a different multimeter that uses shunt to measure voltage drop and I get the same results. I guess that all measurements are in max +- 10% tolerance.
I think AC measurement are the most accurate the problem with AC measurements is the device resolution I can only read with a step of 10mA AC this is 1.2W so I can be wrong with + - 0.6W.
This type of Power supply are the most efficient you can buy this were intended for medical devices but the project was canceled and I purchased this power supply at a good price.
I have a lot of them for sale on eBay.

[electrodacus]

I am very excited to read about your finds, awesome science and very well written and documented. 'No moving parts' - way to go, future here we come. I wonder about just one small (?) thing that is unclear to me, have you attached an additional heatsink to the existing NB heatsink ? If, so - why didn't you rather replace the original? And, have you tried to take off the original heatsink and reseat with some proper care, and proper thermal compound. That's no small thing, i have read about quite a few people who've done with great improvements in cooling.

This post could've been full of smileys..

Yes no moving parts is great I'm electrical engineer for me the moving parts are not reliable
Yes I attached an additional heatsink is an intel heatsink about 120g aluminium this add to the original one that is 49g I also replaced the thermal compound but passive cooling is a challenge at about 15W the CPU for example in my situation only takes less than 24W and I have a huge heatsink about 600g. For north bridge I did not found a bigger cooler I also have limited space because is a SFF case.
The best solution will be to have a low power mobile north bridge less than 10W will be great but I can not find something like this for LGA775.

[khaakon]

Yes I attached an additional heatsink is an intel heatsink about 120g aluminium this add to the original one that is 49g

If you attach another heatsink (on top) of the original, will it not obscure the convectional airflow from the first heatsink? Or will the mechanical contact between them(heat conduction) count for more? I also think some of your sentences could do with some more "." in between to make them easier to interpret. Love your work, I write this because i'm curious, and you mentioned struggling with the NB temp.

[electrodacus]

If you attach another heatsink (on top) of the original, will it not obscure the convectional airflow from the first heatsink? Or will the mechanical contact between them(heat conduction) count for more?

Yes it will obstruct the convectional airflow from the first heatsink but the additional heat sink will remove the heat from the first very easy because of direct contact ( conduction ).
Unfortunately is not enough to be able to run passive the 15.5W of the north bridge is close to 24W of the CPU and I have a much bigger heatsink for the CPU.

I also think some of your sentences could do with some more "." in between to make them easier to interpret. Love your work, I write this because i'm curious, and you mentioned struggling with the NB temp.

Sorry English is not my first language I do my best.
Not to mention that grammar is not my strong point "." even in my native language

[electrodacus]

I tested to see how the power supply will work with about 95W load this 95W was actually at wall AC so less from power supply output to do this load I overclocked the CPU at 3Ghz and added a 950mA LED macro lamp that I made on USB port so I get some additional load on 5V line.
I used the intel fan to cool the CPU is small and only about 900 rpm but was enough to cool the CPU the max temp was 62C on first core the other 3 cores under 60C.
And the power supply was fine with just very little air coming from intel fan the coils were the hottest components on the board but under 60C because I can put the finger on them without getting burned .



see larger image http://www.electrodacus.com/photo/ATX/Q8400S-3GhzL.jpg
and a screenshot http://www.electrodacus.com/photo/ATX/snapshot7.jpg

[Ksanderash]

Nice thread, electrodacus! I do use G31 too in my undervolted setup, but I have not a tool to measure AC so my experiments aren't so interesting:
viewtopic.php?p=480525

Recently I exchanged E5200 for E6300. I use it under 0.8500V idle, and 1.1250 load (stock 2.8GHz). Now my setup is even more silent than was before, cause the PSU fan has no additional heat to fight against, it stays on very low 350RPM.

[electrodacus]

Nice thread, electrodacus! I do use G31 too in my undervolted setup, but I have not a tool to measure AC so my experiments aren't so interesting:
viewtopic.php?p=480525

Recently I exchanged E5200 for E6300. I use it under 0.8500V idle, and 1.1250 load (stock 2.8GHz). Now my setup is even more silent than was before, cause the PSU fan has no additional heat to fight against, it stays on very low 350RPM.

Thanks.
Intel is crazy they have old E6300 65nm and new E6300 45nm

http://processorfinder.intel.com/List.a ... hKey=e6300

[Ksanderash]

Intel is crazy they have old E6300 65nm and new E6300 45nm

http://processorfinder.intel.com/List.a ... hKey=e6300

Yeah. I even have this old Core2Duo E6300 1.86GHz at work It is rather hot for its frequencies.

[NeilBlanchard]

Hi,

You have several fans in this system -- so it has moving parts!

[electrodacus]

Hi,

You have several fans in this system -- so it has moving parts!

The case fan is disconnected (so is not a moving part )
And the fan that is above the north bridge is moving at 5V because I was not able to keep the temp at a comfortable level for me. I used an additional heatsink glued on top of the northbridge heatsink but the temp is still a bit more than 60C without the fan.
This is the reason for this post I wanted help in finding a better motherboard that can be used completely passive But is hard to find a northbridge that uses less power than the G31 this is min 7.5W up to 15.5W at load according to spec and is really hard to cool 15.5W completely passive. I will probably need to wait for the core i3 platform if this will not work but the core i3 will probably not take less power than this system and the CPU will be less powerful only 2 core with HT.

[Ksanderash]

The North bridge is getting to hot I don't know the exact temp

I use MCHTemp but it is in Russian, sorry. Btw, an interesting fact: Intel's MCH internal temperature probe catches temperatures only from 50°C (even 66°C on some models) and higher. So I have constant <50°C readings in idle cause I did an undervolt mod down to 1.17V (266 FSB in pair with E6300) on my G31-M7.

[electrodacus]

The North bridge is getting to hot I don't know the exact temp

I use MCHTemp but it is in Russian, sorry. Btw, an interesting fact: Intel's MCH internal temperature probe catches temperatures only from 50°C (even 66°C on some models) and higher. So I have constant <50°C readings in idle cause I did an undervolt mod down to 1.17V (266 FSB in pair with E6300) on my G31-M7.

Interesting I did not know that can only read from 50°C+ I also use undervolt from bios normal voltage is 1.28V and I can reduce to with -0.05V,-0.10V,-0.15V but I do not see any difference maybe the undervolt is not working.

[Ksanderash]

electrodacus

maybe the undervolt is not working

Very maybe I receive minus 1.5W in idle effect. How about a hardware voltmod?

...

Have you tried to decrease the +5V rail voltage value in your system? Make it like +4.5V and you'll see that ~30W in idle is yet not the limit

[electrodacus]

electrodacus

maybe the undervolt is not working

Very maybe I receive minus 1.5W in idle effect. How about a hardware voltmod?

...

Have you tried to decrease the +5V rail voltage value in your system? Make it like +4.5V and you'll see that ~30W in idle is yet not the limit

I do not have the time now for the hardware voltmod maybe in the future if needed.
Decreasing the 5V will not help to much because the main load are using lower voltage (DC-DC converters) so is no advantage or to small for the trouble .

[Ksanderash]

Decreasing the 5V will not help to much because the main load are using lower voltage (DC-DC converters) so is no advantage or to small for the trouble .

Of course. I know this. But do not forget about the NB and SB having some number of internal circuits "sitting" directly on +5V and +3.3V. Cause I have no idea how could I get just now about 7.2W in idle (was 13.35W!) and 12.65W in load (was 18W!) by introducing in-series a Schottky diode into +5V rail. The 3.3V rail isn't so interesting, there are very few watts spend. So as the 12V one, but for another reason.

Can anyone with a G31/ICH7 board confirm my findings? Maybe my multimeter is wrong

...

OMG! +4.22V instead of +5V -- only 0.52A in idle, and that is ~2.5W!! But I have some wavy graphics distortions on screen... Got to lower DDR supply for such load current, and maybe that's the origin of this flickering.

...

The compie is absolutely stable! Prime95, Linpack verified, and the Braid game (I'm on internal graphics experimenting)

Everest readings (chassis -- PSU fan):

[electrodacus]

Decreasing the 5V will not help to much because the main load are using lower voltage (DC-DC converters) so is no advantage or to small for the trouble .

Of course. I know this. But do not forget about the NB and SB having some number of internal circuits "sitting" directly on +5V and +3.3V. Cause I have no idea how could I get just now about 7.2W in idle (was 13.35W!) and 12.65W in load (was 18W!) by introducing in-series a Schottky diode into +5V rail. The 3.3V rail isn't so interesting, there are very few watts spend. So as the 12V one, but for another reason.

Can anyone with a G31/ICH7 board confirm my findings? Maybe my multimeter is wrong

...

OMG! +4.22V instead of +5V -- only 0.52A in idle, and that is ~2.5W!! But I have some wavy graphics distortions on screen... Got to lower DDR supply for such load current, and maybe that's the origin of this flickering.

...

The compie is absolutely stable! Prime95, Linpack verified, and the Braid game (I'm on internal graphics experimenting)

Everest readings (chassis -- PSU fan):

I was curios so I tested with one of my power supply.
I used a different method to reduce the voltage on the 5V line I added a resistor in parallel with one of the two resistors use for the 5V feedback (resistor divider).
The initial voltage was 5.053V and I reduced to 4.462V and the total power consumption on my PC changed at idle from 31.92W to 31.54W and at load from 49.21W to 48.64W. So as I was expecting not a change worth mentioning.
I can confirm that it was stable at 4.462V but this is not a big surprise.
Maybe you can describe a bit more in detail how you reduce the voltage and where you measure the current.
There are 4 x 5V wires on the 20pin ATX and 5 x 5V wires on the 24pin ATX connector and there are also 5V lines for HDD and DVD than can go to the MB indirectly if necessary.
Can you measure the total system current?