Some theorycraft about low power and Fanless

zir_blazer

Golden Member
Jun 6, 2013
1,260
573
136
When I want to purchase Hardware, I really don't care about the nominal specs, but its actual potential, and how I can adapt it to serve a purpose. For example, when I purchased my current DDR3 modules, I was looking for the cheapest modules with a specific IC (Hynix MFR) which I was told were the best ones at that moment. If I were to buy a Processor, I would buy the cheaper/slower part that includes a specific feature (Be it being a Quad Core, having Hyper Threading, VT-d, TSX, whatever), or the full possible feature set of that silicon die, and so on. What I don't like, is to pay the premium for just some MHzs worth of binning. Actually, I like to toy around figuring out my place on the Frequency/Voltage curve, that is something that I'm used to find and decide myself.
In this generation, since my Supermicro X10SAT being a conservative Workstation-class doesn't allows me to touch any value in the BIOS to run it out-of-spec, I'm out of luck for undervolting my Xeon. And since most Software tools are Windows based, I can't even try them since I use Windows in a VM. But if I were to build a new computer, I will take any flexibility I can if possible, which I absolutely loss in this platform.

While I was thinking of making this Thread for some days, a similar one appeared. If I want to pick that Thread as a basis for what I said above, you will see that most of the difference between the low power parts and the standard or Unlocked ones, is not only that they have a lower TDP, but also lower Frequencies across all Power States.
Since usually on non-Unlocked parts you are limited to choose as max Multiplier the one from the nominal Frequency, with the only possible overclock to force the BIOS's MultiCore Enchanment to be able to use the max Turbo Multiplier, you have an effective artificial cap on these low power parts. You would not have such cap on the standard or Unlocked parts, and since these days there are no proof that they're different or better binned for low power or overclocking ability (Remember Mobile Athlons XP or Socket 939 Opterons?), in that case, the best option would be to pick a standard or Unlocked Processor, which should be more widely available, then underclock and undervolt it to levels similar to the low power Processors. Otherwise, if tomorrow you need higher performance or don't care about low power, the low power parts would be unable to scale up, while the Unlocked part has the freedom to scale in either way.
That pretty much sums the background of this Thread at: I like to pick my parts based on how much silicon I will be getting, then decide at what settings I want to run it.

Normally, manufacturers includes automatic power management with power states and fans settings, but I think that I am more smart than the stock config since I know my needs, and besides, they should be rather conservative to begin with, while I can try to tweak them playing the silicon lottery and testing what its the lower Voltage that I need for a determinated Frequency and still be rock solid, a thing which I already did on my previous Athlon II X4.
So far, I was thinking that I would like to see two or three modes: A totally silent mode where all Fans are actually off (Effectively Fanless) at rather low Frequency values but that still are confortable for Web browsing and light applications, a middle mode with Fans spinning at low speed, and the full blow mode at either max nominal or overclocked settings. In that previous machine, I could do something close to that, since I could open K10stat to manually change the AIIX4 Frequency and Voltage values from within Windows to my best found values, then reducing the Fan speed to the minimum via Speedfan, and achieve nearly the exact results I wanted.


The problem is that it seems that while you can make the Fans spin at the minimum speed, it seems that its not possible to actually power them off. Yes, I know that Fans at slow speeds are barely audible, but I would prefer to not having moving parts at all if possible since that's my aim for the lowest Power State.
Also, for a Fanless system, disconnecting the Processor Fan seems to not be a good idea at all, since many Motherboards seem to assume that if the Processor Fan is not connected, the user is a moron that didn't installed the Heatsink and risk to kill the Processor. Truth be told, I recall some guy that managed to run his S939 Opteron at 800 MHz and with a strong undervolt, without even the Heatsink at all. Since current Processors are much more power efficient, I believe than that should still be possible, assuming you can POST at the desired low power values.
So in a ideal system that follows my idea, the POST should be always done in safe Fanless-capable settings, then manually change them to the desired, faster settings after booting the OS.

Conceptually, that is what I would like to achieve. I always prefer to enforce manual settings than letting automatic ones be in charge when I know what I am doing. On the Athlon II generation I was happy with the results, but I couldn't test anything in this one.
I'm rather confident that considering that both manufacturing processes and architectural designs are all aiming for low power and performance per Watt, and based on some Ultrabook Haswell models, it should be possible to achieve a Quad Core @ 2 GHz with GPU at around 20W power consumption. Fanless with the stock Heatsink should be possible at lower values, but it should still kick ass for light usage.


So, anyone that wants to merge their low power HTPC and high power gaming or work system has any similar through, or tried to implement something similar?
 
Last edited:

PPB

Golden Member
Jul 5, 2013
1,118
168
106
You can always aim for making your metal enclosure your very passive heatsink. The case itself is the largest piece of a high conductive material that is at the same time exposed to the lowest ambient temperature. This is the principle behind Streacom's cases. They even admit fairly potent CPUs considering the max TDP they allow you go for.
 

zir_blazer

Golden Member
Jun 6, 2013
1,260
573
136
You can always aim for making your metal enclosure your very passive heatsink. The case itself is the largest piece of a high conductive material that is at the same time exposed to the lowest ambient temperature. This is the principle behind Streacom's cases. They even admit fairly potent CPUs considering the max TDP they allow you go for.
But that means spending on a specific part instead of adapting a mainstream one. Is would still be a good idea if I was good with modding projects.
 

superstition

Platinum Member
Feb 2, 2008
2,219
221
101
You're better off having two systems connected with a KVM. One would be a passively cooled system and the other would be one that is designed around the highest performance you can get within your comfortable (multi-hour) noise maximum.

Another thing that can help immensely is long cabling. I have a computer in my closet. The doors cut the noise quite a lot. I also have another computer in an adjoining bedroom and run cabling through the wall. That cuts the noise even more. The drawback to the long cabling, though, is the input lag from the USB extension cable. More ideal would be to have a soundproofed wall so the cable run would be short.
 

know of fence

Senior member
May 28, 2009
555
2
71
Even though passive cooling is backwards, inefficient and dangerous, there are threshold wattages where the heat increase is low enough that a fan can't significantly improve the situation.
I believe that is why Thermal Design Power envelopes are so important and why engineers talk about them all the time. Given a certain die size and heat sink, you simply can't go further than 4.5 W (or whatever) in thin mobile devices. For desktop CPUs those thresholds are rarely discussed. However from the broad power spectrum of GPU card designs we can roughly see wattages at which the inclusion of fans or heat-pipes becomes preferable. Some video cards consume around 10 W at idle and actually turn off their fans. Graphics cards mostly hang with their heat sinks upside down, which prevents dust buildup - the second big issue with passive cooling.

Passive cooling is extremely limiting, it can't prevent hard drive hum or electrical buzzing either.
 

Flapdrol1337

Golden Member
May 21, 2014
1,677
93
91
I used to control my 2000 rpm fans with speedfan (could control the voltage on the mobo headers). If the voltage is low enough the fans would stop spinning, and since they're brushless DC they don't use power if they stop due to low voltage.

Slow fans probably spin up with lower voltage, so you might not be able to really stop those.
 

superstition

Platinum Member
Feb 2, 2008
2,219
221
101
Even though passive cooling is backwards, inefficient and dangerous...
That statement is true if the item being cooled passively exceeds the capability of the passive cooling solution. Otherwise it's not.
Passive cooling is extremely limiting...
Not if you're trying to build a music PC for recording or some other system that needs to not emit noise.
it can't prevent hard drive hum or electrical buzzing either.
I doubt that anyone who goes to the trouble of building a passive system in 2015 is going to use a mechanical hard drive instead of ssds. Electrical noise is something that can be controlled.
 

Blastman

Golden Member
Oct 21, 1999
1,758
0
76
On my Gigabyte M/B, with ET6 (EasyTune6) I can adjust the fan speed vs temperature settings to the point where the fan will shut off while the CPU is idling. Minimum speed of the fan (I use voltage settings) is 500rpm -- anything lower and the fan stops. This is with the stock Intel HSF.

The ET6 settings allow quite a large variance such that with a larger aftermarket HSF, I could probably get a CPU fan to not turn on at all except in the most extreme heavy running CPU situations. This wold be essentially a passive setup with a failsafe in case the CPU temp gets too high.
 

know of fence

Senior member
May 28, 2009
555
2
71
That statement is true if the item being cooled passively exceeds the capability of the passive cooling solution.

The great strength of passive cooling is that there is some headroom in which heat transfer gets more efficient with increasing temperature difference, just like any heat engine. The flip side is that to be efficient you need one end of the heatsink to be much hotter than the other. Indeed the theoretical efficiency limit for Carnot cycle heat engines is determined by temperature difference alone. If you think about it the analogy holds true, that chunk of metal does more than transfer heat it actually expands air and moves it around like an engine. So moving air around with a fan immediately appears to be much more preferable to moving it around with convection. I suspect that this is because heated air molecules fly off in all directions including the wrong ones...

Already we have CPUs that increase voltage based on temperature to maintain stability (VID states) and there is Turbo (Temperature dependant up and down-clocking). So by running say 20° C hotter than active cooling, you actually run less efficient and slower, throttling earlier, even without ever exceeding specified TPD. This increase may not justify 3 fan radiators and water pumps, but certainly a 1 W heat pipe cooler.
 

zir_blazer

Golden Member
Jun 6, 2013
1,260
573
136
Forgot about this Thread.


You're better off having two systems connected with a KVM. One would be a passively cooled system and the other would be one that is designed around the highest performance you can get within your comfortable (multi-hour) noise maximum.
This is against what I want to do: An all-in-one, multipurpose system. Since it is possible for a high end, power hungry system to have multiple low power states where it could compete with the power consumption, cooling requeriments and noise of a low power system (And actually beat them in performance, since big cores like Haswells give more performance than Atoms at same power consumption levels, reason why the high end ULV has big cores instead of superclocked small ones), there is no need for such a dedicated low power system on the first place.

For reference, a current power user trend is having all-in-one systems based on a Hypervisor, then throw multiple VMs on top of it, using PCI/VGA Passthrough to put non-shareable resources at the VM you need them. So you can have for example, a Windows VM with an assigned Video Card to play games, a VM with another OS to act as File Server with its dedicated SATA/RAID Controller with multiple Hard Diskes using the ZFS File System, a secundary Windows VM with an USB Controller and another Video Card to make it a multiseat computer, and even host VMs intended to be remotely accessible in case you want someone to do his work there, like a personal cloud.
What I want to do, is throw zero noise on top of that. The problem is that the current trend in order to save power consumption (and noise), is usually entering states like Suspend or Hibernation, while what I want is instead a fully functional low power mode that relies on low Frequency and Voltage, instead of simply a state that is quick to enter and exit but useless while there. There aren't technical limitations for this to not be possible, just that it requires more tinker that usual because manufacturers didn't care about it functional low power states, but how many parts you can power off without powering off the entire system.


I used to control my 2000 rpm fans with speedfan (could control the voltage on the mobo headers). If the voltage is low enough the fans would stop spinning, and since they're brushless DC they don't use power if they stop due to low voltage.

Slow fans probably spin up with lower voltage, so you might not be able to really stop those.


On my Gigabyte M/B, with ET6 (EasyTune6) I can adjust the fan speed vs temperature settings to the point where the fan will shut off while the CPU is idling. Minimum speed of the fan (I use voltage settings) is 500rpm -- anything lower and the fan stops. This is with the stock Intel HSF.

The ET6 settings allow quite a large variance such that with a larger aftermarket HSF, I could probably get a CPU fan to not turn on at all except in the most extreme heavy running CPU situations. This wold be essentially a passive setup with a failsafe in case the CPU temp gets too high.
These are interesing, however, I can't try it.

Another reason why I can't experiment with my current Xeon, is that since my Windows is in fact a VM with VGA Passthrough like the one I previously linked, I can't use all the widely know Windows applications to do low level control of my system, and I didn't looked around what Linux has to offer me in Hardware control tools.
 

VirtualLarry

No Lifer
Aug 25, 2001
56,587
10,225
126
zir_blazer, DFI made a "Twin" S775 high-end mobo, that was two systems in one. It took a 775 CPU for the main rig, and it also had a small low-power Atom-based rig also on board.

It was pretty innovative for the time.
 

zir_blazer

Golden Member
Jun 6, 2013
1,260
573
136
zir_blazer, DFI made a "Twin" S775 high-end mobo, that was two systems in one. It took a 775 CPU for the main rig, and it also had a small low-power Atom-based rig also on board.

It was pretty innovative for the time.
I just googled it. It should be the DFI P45-ION-T2A2. That thing is absolutely bizarre, AsRock level bizarre (Or maybe better, if both platforms could be used at the same time). Still, that is not my goal. The whole point about the virtualization trend is to consolidate systems into one by sharing resources, not to have multiple systems with KVM switchs and the like.

Anyways, that is not related to where I'm pointing at with this Thread. Most of your computer parts don't work at a single, fixed state. Since you can make that your shining Core i7 4790K work at either default 4-4.4 GHz, overclock it higher, or underclock and undervolt it to oblivion to compete with low power parts, you don't need multiple systems, since you can do everything in a single one if you tweak it.
As there are a whole bunch of tasks that can be done on fanless type computers, it means that with a proper setup, you should be able to do things that doesn't require a lot of CPU or GPU horsepower (Internet browsing, File Server, etc) without even spinning up a Fan, yet still have the chance to ramp up performance if you decide you want to play a heavy game. You should be able to achieve that level of flexibility with a SINGLE computer, not multiple.
 

Auric

Diamond Member
Oct 11, 1999
9,591
2
71
There's quite a minimum speed variance between motherboards and Intel coolers from various suppliers -and even the same fan on different systems. Wish I could get 500 but is currently 1078 RPM, and bypassing PWM is not an option.
 

sm625

Diamond Member
May 6, 2011
8,172
137
106
You could always buy an arduino board and a MAX6675 type K thermocouple module and an SSR and write a program that controls a solid state relay which can turn off your fan when the temperature is below 60C and turns it back on when the temperature is above 70C.

http://www.instructables.com/id/Arduino-and-Thermocouple-K-MAX6675/

If your cpu is idling it really does not need a fan. Even a 4790K, as long as it have a decent cooler. You can even force idle speeds by setting the windows max processor state. Then in theory you could have a completely silent high end pc that can switch on its own fans and become a beast when you need it to be.

You could also buy a super cheap (50 cent) SOIC SSR and wire it to a digital I/O line from your printer port, and write a simple program that reads your CPU temperature and toggles that digital I/O line when the temperature is low enough. The SSR would be placed in series with your CPU fan power. As long as your cpu has a decent cooler and is forced low enough in frequency, the SSR should stay off, keeping the computer silent until you actually need more performance.

I'm not really sure how practical any of this is though, since a nice 120mm fan is already damn near silent when your cpu is at idle.
 

monkeydelmagico

Diamond Member
Nov 16, 2011
3,961
145
106
The problem is that it seems that while you can make the Fans spin at the minimum speed, it seems that its not possible to actually power them off.

Smart Bios control and profiles are your friend. Undervolt/clock. I can turn every fan in my system off if desired and still perform everyday functions like email, web, and edit documents/spreadsheets.