Cooling the Cube

[rossi1959]

Thank's for the encouraging support Ersterhernd, In all honesty I do think that you have reached the practical limit and would probably only need to get a little more airflow through the meshed bottom plate to get your temps down to make it totally reliable at full tilt, where you would not have to worry how the temps are at all, Sleppeks build I would think especially with the graphics card would need the airflow increased a lot to keep that system on a leash, I think even though the low wattage i7's produce little heat or even the same as the i3 in mild use and idle states, they do start to churn out the heat under load. as does the i3,i5 or any cpu for that matter. It will be interesting when I get the i5 in there and see it under load with my modded heatsink , that will be the telling factor even before I install the graphics card as to where I go to from here, I hopefully will be doing this tomorrow with a screen shot of the tests. Rossi.

 

[ersterhernd]

Rossi I think you've aptly coined the key phrase as 'practical limits' in your last post. The stress tests we perform on these builds, Prime95 especially, go well beyond what could be defined as any kind of practical limit. In all my years of building PC's, I've never used prime95 as a measure of heat, but rather overclocked stability. A prudent PC or Hackintosh builder makes careful predictions and planning BEFORE the project starts, and later confirms his projections with software like prime95 to test the upper thresholds of temp and stability.

A 'practical limit' is more clearly defined by real world applications (video proccesing, gaming) that would likely be used at some point. Turning Prime95 onto its max heat torture test, cranking up the furnace in the house, closing the office door and walking away are simply condiitions that would never likely exist.

I'm far more interested in seeing the efforts like yours rewarded with successful operation under realistic environments than those testing the stress thresholds of the hardware. It'll be interesting to see where your build ends up.



Cheers.

 

[minihack]

I am not sure if this is helpful - but thought I'd mention it in case anyone is brave enough to try!

One of the reasons for relatively hot CPUs in the Ivy Bridge line is apparently an inadequate thermal junction to the heat spreader (internally). A non fan related improvement can be made by removing the heat spreader, re-pasting and reassembling. There are lots of guides on doing it - here is one: http://www.youtube.com/watch?v=XXs0I5kuoX4

Personally I would be too scared to do this one.

 

[rossi1959]

I am not sure if this is helpful - but thought I'd mention it in case anyone is brave enough to try!

One of the reasons for relatively hot CPUs in the Ivy Bridge line is apparently an inadequate thermal junction to the heat spreader (internally). A non fan related improvement can be made by removing the heat spreader, re-pasting and reassembling. There are lots of guides on doing it - here is one: http://www.youtube.com/watch?v=XXs0I5kuoX4

Personally I would be too scared to do this one.

I'm with you on that one Minihack, in my post #19 I was pointing to that procedure, but I don't think I would go to those measure's, looks like a very costly mistake could be made.(probably not too different in the fatal mistakes department to what I am trying to achieve at this moment)

 

[rossi1959]

Rossi I think you've aptly coined the key phrase as 'practical limits' in your last post. The stress tests we perform on these builds, Prime95 especially, go well beyond what could be defined as any kind of practical limit. In all my years of building PC's, I've never used prime95 as a measure of heat, but rather overclocked stability. A prudent PC or Hackintosh builder makes careful predictions and planning BEFORE the project starts, and later confirms his projections with software like prime95 to test the upper thresholds of temp and stability.

A 'practical limit' is more clearly defined by real world applications (video proccesing, gaming) that would likely be used at some point. Turning Prime95 onto its max heat torture test, cranking up the furnace in the house, closing the office door and walking away are simply condiitions that would never likely exist.

I'm far more interested in seeing the efforts like yours rewarded with successful operation under realistic environments than those testing the stress thresholds of the hardware. It'll be interesting to see where your build ends up.



Cheers.

My aim with this mod is primarily to see what power limits can be reached in the cube, and have the cube as an every day easy to live with PC (High Hopes), I could have played it safe and left it with the i3 running the HD4000 graphics, and it would have served me well in that configuration for what I use it for, but I've always pushed the limits and like a challenge. I just wish it were a bit less costly as I have just purchased 2x Noctua 92mm fans to put in the blower at the base of the cube and a new Noctua CPU cooler to replace the custom hybrid cooler I am using at the moment, I am doing this to try and eliminate the PSU heat more effectivly and the Noctua's will help in that area, I will also remove the GT 640's fan and see if the air flow from the Noctua blowers is sufficient to keep the temps down, as the cards heatsink will be central in the cubes tunnel. I will try to get the i5 in and see what the temp rise from it will be, and I hope to have the Fans and graphics card next week to complete all the parts, I am thinking that the G4 in a G5 mod I have will be ideal for this as it will accept a 180mm fan installed into the back sucking air through the cube which would be backwards to the normal airflow, will keep you all up to date as much as possible.
Cheers.
Rossi.

 

[ersterhernd]

Personally I would be too scared to do this one.

Oh yeah. Me too...

 

[minihack]

I'm with you on that one Minihack, in my post #19 I was pointing to that procedure,

Sorry Rossi - I missed that post. Ooops.

The way the guy in the You Tube video treats his CPU looks also to be just a little bit too brutal. Looks like a nice way to lose money when you get that procedure wrong!

 

[rossi1959]

Sorry Rossi - I missed that post. Ooops.

The way the guy in the You Tube video treats his CPU looks also to be just a little bit too brutal. Looks like a nice way to lose money when you get that procedure wrong!

I came across that gem studying the temperature characteristics and the limits of the i series CPU's for my project, and thought I am spending enough time and money without having to resort to those dubious tactics. Rossi.

 

[rossi1959]

After a fatal crash of the OS, to which I am not sure what did it, I couldn't boot, and used -v and saw that it stopped at "Unrecognised CPU", I googled and found it was quite a common error among not only Mac but Win as well, I found that using the flag "CPU=1"(without quotes) got me into the OS(once) where I ran multi beast,rebooted and needed -v -x and PCIRootUID=1 to boot because CPU=1 didn't work this time, ran the beast again and the - v boot stopped at a line with reference to EXFAT32, and that's where it stopped at every time I tried to boot back in, so I tried different flags and all different combinations of flags, but the real kicker was that the CCC clone drive wouldn't boot either, so I reset the bios because the date & time settings had changed and were wrong and still no go, so I reflashed the bios with the very latest file and still no boot into either drive, so after quite some hours of very frustrating boots and reboots, I threw in the towel and got out the SL disk, reimaged the drive and downloaded the already purchased 10.8.3 ML, loaded unibeast and here I am again.

I have just received the new Gigabyte GT640 graphics card and pcie ext cable, and so while I have the hardware from the cube on the bench awaiting the new fans for the cube I thought I would do some testing to see if the HD4000 is anywhere near the GT640, and no it's not.
As you will see below geekbench has the HD4000 ahead of the same system with the GT640 added in (go figure), but cinebench clearly shows the difference with the GT640's FR well ahead (especially when they are running so low down the scale,it's way ahead), I will be installing an i5 3750k for the next test which will show some very much improved performance but will create a bit more heat and I hope to do this with just a 300 watt PSU, this following link is a good example of the i5's grunt, but keep in mind that their test system has a 670 GC in it as well. http://apcmag.com/ivy-bridge-cpu-buyers-guide.htm

Kiwi has stated in another thread that his Cube build will incorporate the new Haswell CPU and if it is an i7, then all this will be in vain because Haswell is rumoured to be twice as powerful in the graphics department and is also putting out more heat in the process, but would still be under the power consumption and heat production than the contraption I am building, the interesting part in the link refers to the Cuda cores, and the GT 640 has 2 gig of only DDR3 Ram and 384 Cuda cores, the link states that the card will need 896mb of ram for the Mercury Playback Engine to work in Adobe Premiere. Oh well back to the drawing board and see what the i5 brings and see if I need to go up in the PSU.
Rossi.

 

[ersterhernd]

Wow Rossi that's quite the crash.

My i7/HD4000 combo pushed out 25+ FPS in Cinebench, about 50% faster than the i3. HD4000 performance is clearly tied to processing power. I'll benchmark it with the Heaven Benchmark next. I looped that program continuously yesterday and the core temps stayed cool. I think Heaven @ 1920x1080 is probably a better real world comparison between onboard & dedicated GPU's.

Haswell is certainly going to be one to watch.

Good Luck with your continued testing.


Ersterhernd