martinsliquidlab.org

Testing Thermal Performance

THERMAL TESTING
I’m using a slightly revised test standard for my 3930K platform which is pretty much the same as previous with the exception of utilizing the same mounting hardware and mounting pressure.  I am also now using the latest batch 2 MX-2 compound which is much thinner in consistency and spreads easier than my old batch.  In addition I am now testing each block over 2 mounts vertically, 2 mounts horizontally, then the 5 mount whichever way is determined best by the first 4 mounts.  This way you can see what if any differences there are between mounting orientations.  This setup is generally focused on absolute and extreme performance levels using stronger than average pumping power, stronger than average radiator performance, and maximum flow CPU block only type performance.  The only thing average is the thermal paste which isn’t going to perform as well as the liquid metals, but it is one of the better thermal pastes that I have personally used.

THERMAL TESTING SPECS

• ASrock X79 Extreme 7
• i7-3930K processor, overclocked to 4.5Ghz at 1.344-1.352Vcore, Should be between 175W to 250W of heat. I have the newer C2 revision which could go much higher, but I’m taking it easy and 4.5 still puts out plenty of heat.
• 16GB Corsair Vengeance RAM
• EVGA 570GTX Video Card
• Swiftech MCP-35X Pump @100% with reservoir
• 1/2″ ID x 3/4″ OD Danger Den tubing, Koolance Spring Clamps
• TFC 480ER Radiator with Yate Loon D12SM12 fans at full speed
• Wal-mart brand distilled water (<5uS) + PTnuke
• Fixed Mounting Hardware – XSPC Raystorm M4 Studs, Large Metal Washer, Small M4 Washer, Small M4 Black Nylon Washer, Danger Den M6 7-8lb Spring, Danger Den M6 Brass M4 Oversized Thumbnut.  Factory ASrock LGA2011 built in retention/backplate.
• Arctic Cooling MX-2 Feb2012 Batch 2, 20g Tube, large + method
• The block is seated over the processor very carefully and each thumbnut is tightened equally with opposing corners back and forth until the spring is fully compressed by indication of stud depth.  Then each nut is carefully tightened just until full spring contact is made.
• The system is then launched into windows, followed by activating the Crystalfontz CFA-633 which is logging two Dallas Digital one wire probes for water and four each for air.  Packet debugger is turned off in CFA-633. Then Real Temp GT 3.70 is launched and recording 6 each of the core temperatures at 1 second intervals.
• Prime 95, Custom Small 8K FTT is started to load 12 worker threads at 100%.
• The test is typically run for 1 hour, the data is then combined in excel trimming the first 10 minutes of warm up time, and plotted to ensure air/water/loading was done correctly.  Finally the average of 30minutes minimum is averaged to produce the summary results.

5 MOUNTS SUMMARY

Those are the 5 mounts of data. The more minimal bow did result in fairly consistently good results on the larger core processor netting a lower than average .32C standard deviation.

Pages: 1 2 3 4 5 6 7 8