Testing Thermal Performance
For thermal testing I decided to use a 5 mount method of logging temperatures at a fixed ambient measuring core, water, and air temperatures very carefully. Here are some of the specifics in my testing method:
- Intel i7 2600K Processor – Overclocked to 4800 MHz, 32nm, Vcore = 1.392 under load. Motherboard is an MSI P67A-GD65 with 8GB of Corsair Vengance memory. Everything else is on air at the moment. Test bench is a Danger Den Torture Rack
- 5 separate TIM applications and mounts, more if needed.
- Logging temperatures over a 45 minute test, 10 minute warmup is removed from the results, and the entire logged run is provided for review.
- 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 sensors (After radiator and pump, just before CPU block) Dallas DS18B20 Digital one-wire sensors, and CrystalFontz CFA-633
- 4 ea Air In Dallas DS18B20 Digital one-wire sensors, and CrystalFontz CFA-633
- 4 ea i7 2600K DTS sensors these were logged using Real Temp 3.67 .
- 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.
- Pump – Swiftech MCP-35X with reservoir. I think this pump represents the pumping power available to many users and gives a fair amount of strong pumping power and the PWM capability makes for easy variable pumping power levels
- Radiator – The Feser Company (TFC) 480 ER radiator with Yate loon D12SL12 slow speed fans at 12V. No longer in production, but a good strong quad radiator provide roughly a 3C water/air delta when under load with the fans I’m using. The small delta means warm up time is kept to a minimum. The system will normally stabilize in under 10 minutes so there is little waste in warm up and air water variability is kept to a minimum. Users with 240 rads with similar fans should expect results to be roughly 3 degrees higher than what I have recorded.
- TIM Material – I used Arctic Cooling MX-2 for this test in a cross application method. It is considered a paste that requires very little cure and commonly enough that I felt it represent general use fairly well.
- Hardware – If the system provided by the block ensures a common mounting pressure, I will use that. Otherwise I will use my own backplate and mounting system to make testing more consistent across blocks. In this test, there was not a special mounting system, so I used my standard.
- Prime 95 Load – I used Prime 95, 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. OCCT wasn’t capable of maintaining a full 100% load and so I feel Prime 95 is a more consistent loading routine and seems to work fine. It creates 8 100% load working threads and an extreme load.
- Stock IHS – I chose to leave the IHS alone this time since it would be more representative of what most people do.
INDIVIDUAL MOUNT RUNS
For each mount I like to show the actual logged run of the test to show the interaction between water, air, and core temps. Each air/water bump is a result of my force air furnace kicking on at the set ambient level. Water always lags a bit behind the sudden increase in air temperature, but overall the average of these results is ultimately used. Each sensor records a data point at every second, so it’s quite a few lines in excel logged over time. Here are those logged testing runs showing consistency in the test proceedure:
I am generally very pleased with the testing environment. I was able to average a standard deviation between ambient air testing results to a nice tight .17 C which is very good. This allows me to compare results without any correlation or correction, these are the straight numbers as recorded. I prefer the fixed ambient method if possible because not all heat is lost into the water and some processors are particularly susceptible to problems with varied ambients. The i7 seems to be less problematic with varied ambients, but since the tests were carried out in a fixed ambient there is no no need to correct anything or worries about ambient changes…all tested in the same condition here.
My biggest and problematic variable is the mount itself where my standard deviation between the five mounts was .60C. This means my mean number has a 68% confidence in accuracy to +-.6 degrees. For a 95% confidence in the result, two standard deviations or +-1.2C would be the range. More mounts would be required to improve this standard deviation in mounting. Some of this variability is due to the bowed base and rest is user error..:) While I’ve tried to be as consistent as possible, the bowed base allows the block to rock and potentially mount with more pressure to one side than the other. I attempted to ensure all four corners were of equal spring pressure by measuring thread depth and also by feel, but it’s a bit of a guessing game on these types of mounting systems.
That’s all fine, so how does it compare you might ask. This is my second block in this CPU block testing series, so I do have some data to compare. Rather than simply giving you a bar chart, I like to present the data with the variable mounts and results to again emphasize the importance of a good mount and what sort of variability you might expect. Since the data above is consistent in both core/air and core/water, I chose to present it relative to water temperature to encourage users to understand what their water temperature is. While I may be getting a 3C water/air delta, I suspect the average user with smaller radiator would see upwards of a 10C water air/delta if additional heat is added to the loop via GPU blocks. This leaves the data below still usable for someone to understand what they might expect. If you know what your water temperature is, the below is how much you add to get the core temperature average. Again, this is average of all four cores and I do have one core that runs noticeably cooler than the other three, so your mileage may vary..:)
NOTE: The Supreme used for comparison is the OLD EK Supreme V1, this is NOT the newer HF model.
Overall, I was impressed with how much the new Koolance did over the old supreme V1, particularly when considering the CPU-370 is much lower in restriction. While both are a similar microchannel/slot nozzle style, the CPU-370 is simply more optimized and provided my i7 2600K roughly a 3 degree mean improvement at 100% pumping power.
THERMAL PERFORMANCE VS FLOW RATE
As an added piece, I’m “attempting” my first flow rate effects piece. Rather than testing various flow rates at all mounts (A TON of work!), I simply kept the last mount in place and tested three additional pumping powers/flow rates just to feel out some information. My particular interest in this is seeing how the block reacts performance wise at extremely low flow rates. This is helpful information for both low pumping powers as well as for folks interested in dynamically throttling pump speeds via PWM pumps. Since the block was relatively low in restriction, the MCP-35X has plenty of power to explore a wide range of flow rates.
While the test data points are fairly scattered (only 4 points), it still gives you some ideas about flow rate effects:
I measured roughly a 2 degree loss in performance going from 100% pumping power beyond 2.o GPM to using 20% pumping power around .6GPM. This means the block’s micro-channel design is very capable of producing great results even with very very low flow rates or very small amounts of pumping power. So for those noiseless priority types like myself, you should have no problem dialing that pump speed down to whisper quiet speeds knowing the block will still perform very well there. Also because the block restriction is fairly low, there isn’t much need to consider multiple pumps in average systems.
Overall, I was very pleased to see the advancements in performance of the CPU-370 over the old EK Supreme V1. Koolance has managed to produce a block that has half as much restriction, it performs better by a few degrees, and is also a very strong performer all a variety of flow rates…excellent!!