Special thanks to Sidewinder, Swiftech, Danger Den, D-Tek, Enzotech, OCZ, XSPC, TFC, and NCIX, and Rick Cain/Niksub1 for providing the samples.
Thermal Test Specifications (for us testing geeks)
For thermal testing I decided to try and follow a 5 mount method of logging temperatures. Here are some of the specifics in my testing method:
- Real World Full System Testing – Over time I developed a radiator testing bench that I’ve decided to incorporate into my CPU block testing setup. This gives me a full 8 air inlet sensors, 4 air outlet sensors, and two water sensors to much more precisely monitor everything. My testing occurs as close to the real world as possible, I just have a crap load of sensors on everything. I am testing on a processor, with a loading program, in a real world computer case, with a fixed pump, and a regular radiator setup. All of this helps include the little odd things that actually occur in a regular system.
- Intel Core 2 Quad Q6600 Kentsfield Processor – Overclocked to 3600 MHz, 65nm, Vcore = 1.472 under load. Motherboard is a DFI Lanparty X48 LT with 4GB of Corsair Dominator memory with fan module. The northbridge and southbridge chips are both watercooled. Video card is an EVGA 8800GTX, also watercooled. Case is a Thermaltake Armor, position is horizontal for easy block mounting case cover left off.
- 5 separate TIM applications and mounts averaged – This is not common, but extremely important. It’s not uncommon at all to see mounting variations as high as 2 degrees or more, so with only one mount, that error is 2 degrees. When you mount 5 times and average those results, your standard deviation is significantly lowered and the overall testing confidence improved. In addition multiple mounts serve as a means to validate data, because each test is carried out again and again, chances are if some variable is affecting results, it will show.
- Logging temperatures – After several iterations of a new testing method I finally landed on logging of temperatures for 1 hour. I then can simply start up the loading routine, and trigger on my two logging programs and walk away for an hour. I then come back and remove the first 10minutes for warmup time. The remaining 50minutes is then left to average out temperatures which are recorded every second over 18 active sensors plus 4 core temperatures. Logging is essential for higher resolution measurements. Our DTS core sensors are only resolved to 1C, however after logging that resolution out for 50 minutes of testing, that can be reduced significantly. In addition my ambient temperatures are held constant by an A/C system thermostat that actually makes the temperature swig up and down in a sawtooth like fashion over a 2C limit. Logging this sawtooth occurrence over a long period of time also levels out the ambient to a nicely resolved level of accuracy.
- Temperature Probes Deployed – I kept my sensors fairly basic, but I did run a few extra’s just for interesting information. This includes a sensor for:
- 2 ea Water Out sensors (After radiator and before CPU block) Dallas DS18B20 Digital one-wire sensors, and CrystalFontz CFA-633
- 8 ea Air In Bottom (Bottom inlet side of radiator) Dallas DS18B20 Digital one-wire sensors, and CrystalFontz CFA-633
- 4 ea Air Out Top (Top out side of radiator) Dallas DS18B20 Digital one-wire sensors, and CrystalFontz CFA-633
- 4 ea Q6600 DTS sensor these were logged using Core Temp Beta Version 0.94. I chose this particular version because it stacks logged data cores in separate column which makes integrating into my crystalfontz data much easier.
- Crystal Fontz logging is accomplished through the use of their Cyrstalfonts 633 WinTest b1.9. Only special settings are turning off all packet debugger check boxes to avoid paging the processor.
- The Dallas DS18B20 Digital one-wire sensors that were used as noted above have a specified absolute accuracy of .5C with a .2C accuracy between 20 -30C temperature range. They also have resolution down to .0625C which is very good, and because they are digital they are not affected by the wiring or length of wire like thermocouples are.
- The CrystalFontz CFA-633 is an LCD with up to 32 channels of monitoring and logging capability. It logs temperatures of each channel on a one second interval, so over a 30 minute test, I’ll have about 1,800 entries noting time and temperature of each channel. These are then averaged for a fairly accurate number.
- Pump – Laing DDC3.2 with XSPC Reservoir top. I think this pump represents the pumping power available to many users and gives a fair amount of strong pumping power. This top rated near the top in my pump top testing and is very powerful.
- Radiator – The Feser Company (TFC) 480 ER radiator with Yate loon D12SL12 medium speed fans at 12V with TFC shrouds in pull condition. This is an extremely powerful radiator and was purposely chosen because the smaller water/ambient deltas reach equilibrium fast and pressure drop for this radiator is minimal. This provided me very short warmup periods and ensured maximum pumping power for the CPU blocks being tested.
- TIM Material – While I really like the TIM consultants TC Grease 0098, I found the thicker consistency wasn’t helping with mounting consistency and required user effort to seat the block down. For that reason I chose Artic Cooling MX-2, it’s been very popular in the forums and preliminary testing showed to perform well, it’s noted as non-curing, and more importantly is a thickness/consistency that more accurately represents most thermal compounds and easily applied and removed. I felt this was important to maintaining a higher level of repeatability. There was no cure time allowed for the compound (even though it’s noted as a non curing compound). Testing was started immediately following block installation. TIM installation method is the thin line method.
- Hardware – These results only tested the “As Shipped” method that use the hardware included and provides users some evaluation and performance of what you get straight from the box unless noted with a backplate. During supreme testing round 2 I found that adding a backplate and using the same mounting pressure didn’t result in any notable difference. So in effort of protecting my valuable motherboard, I used a Scythe backplate during those tests and they are indicated by a “Scythe” backplate notation on the results.
- Prime 95 Load – I used Prime 97, torture test, Custom, Min FTT 8K, Max FTT 8K, Run FFTs in place checked ON. This is an easy to use and constently loading program. It provided the most consistent loading I could find for quad cores.
- Lapped IHS – My Q6600 has been lapped flat down to 1200 grit to ensure a true and flat surface. The stock intel IHS can be very irregular, some are convex, some are concave, some are wavy, and some are fairly flat. Lapping a processor voids it’s warranty, but it ensures a nice flat surface for optimal heat transfer. My particular processor has been lapped. A complete stock IHS may benefit more from a bowed block than my samples because it is flat.
- Fixed Ambient Temperature – My recent roundup of block testing brought forth that regardless of measuring water temperature you will in fact still incur testing error of the processor core temperatures if you allow ambient temperatures to fluctuate. I found on the order of .2 t .3C error for every degree in ambient difference, although it was too variable to pin down and appropriately correct for. In the end I decided on buying a window A/C unit that holds all testing to the same temperature. Ambient do go up and down as the thermostat kicks on and off, but the overall average temperature over a long logged test has been holding to less than .5C which is very good. I am running a Hair 8,000 BTU thermostatically controlled A/C unit in “Cool fan on at all times” to keep temperatures at 22C plus or minus about .5C on average over the length of the test.
As Tested Resulting Flow Rate
Pressure Drop – Restriction
Apogee GT – Bottoming screws ensures accurate/even pressure, Back-plate
Apogee GTZ – Bottoming screws ensures accurate/even pressure, Back-plate
Danger Den TDX – Standard springs/open thumb-nut
Danger Den MC-TDX – Standard springs/open thumb-nut
D-Tek Fuzion V1 – Standard springs/open thumb-nut, Back-plate
D-Tek Fuzion V2 – Bottoming screws ensures accurate/even pressure, Backplate
EK Supreme – Standard springs/open thumb-nut
Enzotech Saphire RevA – Bottoming nut ensures accurate/even pressure, Backplate
OCZ Hydroflow – Bottoming nut ensures accurate/even pressure, Back-plate
XSPC Delta V3 – Standard springs/open thumb-nut
Apogee GT – Machined Copper Base, Cast Delrin Top
Apogee GTZ – Machined Copper Base, Machined Delrin Top
Danger Den TDX – Cast Copper Base, Laser Cut Acrylic or Brass Top
Danger Den MC-TDX – Cast Copper Base, Laser Cut Acrylic or Brass Top
D-Tek Fuzion V1 – Cast Copper Base, Cast Delrin Top
D-Tek Fuzion V2 – Cast Copper Base, Cast Delrin Top
EK Supreme – Machined Copper Base, Machined Delrin or Acrylic Top
Enzotech Saphire RevA – Forged Copper Base, Forged Copper Top, Plastic Cover
OCZ Hydroflow – Machined Copper Base, Nylon Top
XSPC Delta V3 – Machined Copper Base, Machined Delrin Top
In the end for me, they all passed. They all cooled my Q6600 more than enough to run the overclock I was running. During the winter when not testing and running a wood stove, I can see up to 15C fluctuations in house ambient temperatures so 4C is really tight in that context. In addition with the mounting variations and limited DTS sensor resolution, the thermal results are still only good to roughly 1-2C so I can’t help but feel many of the results are simply too close to conclude much of anything but that they did cool the processor down well enough. I was surprised how close the results are thermally, obviously we are coming to a barrier in thermal efficiency. More importantly these results only apply to this one processor, it would be different for dual core, or larger cores. Generally I think blocks that have very small center nozzles would do better from smaller processors and those with long spread-out type nozzles or distributed flow paths would do better with larger cores. Most people will likely see more gain by doubling up on their radiator capacity than changing out of CPU blocks. If the average radiator water/ambient delta is 10C, doubling the rads would be worth 5C which is greater than the difference between any block here. The differences are just pretty small…and frankly hard to even measure.
Also note the mounting mechanism differences. The old spring/open nut styles are not going to be as easy to use and ensure even mounting pressure as those that have some sort of bottoming nut or other design item that prevents the user from tightening one side or corner more than the others. Furthermore, some come with back-plates while others do not, you better plan on buying one for those that do not, it’s essential in protecting your motherboard and a hidden cost.
Finally bowing and stepping do help with the last few bits of efficiency, although they also have some negatives. They work by increasing mounting pressure per unit area for the critical surface directly above the die. However they work against mounting consistency and are more difficult to mount because of this. Finally some may or may not be optimized or designed to cover larger or smaller cores. I like a small amount of bowing myself, it seems to work like camber in a bridge if nothing else and counter the flex of the block in high pressure applications.
Just one set of tests, under one set of conditions, using a lapped Q6600 processor. I make mistakes like anyone, your results may differ especially if it’s any other type of processor.