I really liked my air-cooled i920 setup: Promalitech Megahalems with a pair of push/pull Scythe Slipstream 1200s. I wanted a relatively quiet air setup that I could be fairly aggressive with overclocking, aiming to hit 4.2GHz. The conventional wisdom is that the Megahalems is the one of the best HS, especially out of the box, and that push/pull is the optimal fan setup and that 1200 rpm is right in the neighbourhood of the optimum speed for a Megahalems fan. It’s a no-brainer, right?
It seemed to work out fine. The dummy OC had great temps. I got up to 3.77GHz on standard voltages, and had what I thought was a stable 4.2 OC with temps in the 80s. But I had these 38mm Delta fans someone had given me, and a Scythe fan controller, and some extra time, and a new Newegg order that I could toss a couple of 1900rpm Slipstreams onto to actually test push vs push/pull and a fan known for high static pressure against one that wasn’t so highly regarded.
Test setup:
OS: Windows 7 Ultimate RTM
CPU: Intel i920 D0
CPU Cooler: Prolimatech Megahalems
TIM: MX2
RAM: OCZ Platinum 3 x 2GB 1066MHz
MOBO: EVGA x58 SLI
GPU: EVGA 275 GTX SLI
PSU: Corsair 850TX
Case: Cooler Master HAF 922
Case Fans: Cooler Master 200 x 25 110 CFM (front and side panel, both intake)
Scythe Slipstream 1200rpm (front, intake; rear and 2x top, exhaust)
Ambient temperature: 18C/20C
Fan Controller: Scythe Master Ace.
Test fans: Scythe Slipstream 1900rpm; Delta AFB1212LE 2000rpm
I set out to test the Delta fan first, going with just a push fan plus a similar fan in the exhaust port right behind the heatsink. I’d been having trouble with VREG temps and I thought that if I could get away with a single fan pushing then I could get a small fan in front of the VREG heatsink. More on that later.
I intended to run the Deltas at 2000rpm, 1600rpm and 1200 rpm to match the available fan speeds of the slipstreams, but neither would get over 1650 rpm. I tested the Slipstreams at 1200rpm, 1600rpm and 1900rpm in push and push/pull. I was going to test each set at stock CPU clock, dummy OC 3.3GHz, 3.77GHz and 4.2GH with both IntelBurnTest and Prime95, but I found that the last OC wasn’t stable. The first test setup, Deltas at 1600rpm, put temps in the mid-80s before Prime95 quit responding. I also thought that no-one is really interested in the out-of-box speed of the famed i920, so I saved myself some time and just looked at the middle two CPU clocks. I ran 15 iterations of IBT and followed up with Prime95 Blend for ten minutes
Besides IntelBurnTest and Prime95, I used CPUID’s HWMonitor to track max temps, RealTemp to validate the temps from HWMonitor real-time, and CPUZ64 and EVGA E-LEET to check that the CPU clocks were valid. I recorded the four core temps and the VREG temp for each test.
I started with the lower CPU clock speed and with the faster fan RPM’s and worked my way down, assuming that as I went along temps would increase, and I didn’t want the previous tests to thermally influence the follow-on tests. I also reset the min and max in HWMonitor for each test.
Before I get into the specific results, there were a couple of anomalies I’d like the mention. The first was the biggest – the 3.77GHz OC actually ran cooler than the dummy OC did, in some cases by 10C if we take into account the ambient temperature delta between the two test days. I compared the two overclocks, and the only difference other than QPI settings that I found is that the VCORE in the dummy OC is bumped up by 75mV. I have no idea how that little voltage bump could be the difference. I’ll be running at the 3.77GHz from now on, though.
The other anomaly was the behaviour of the temperature readings between push and push/pull. For some reason the push test had lower core temps and higher VREG temps than the push/pull tests at the same settings. It wasn’t until the very last test, the 1200rpm push test, at about midnight on the first night that I figured it out. I had run Prime95 and IBT on all the setups and was pretty well burnt out, staring into the grill on the side of the case as the test ran. “Hey”, my groggy self thought, “I can see that the CPU is running all right, because I can’t read the label as it spins around”. About five seconds later groggy self replies, “Label? You shouldn’t see the label, that’s the side the fan blows towards”. Groggy self face-palms at the realization that the push fan is mounted backwards and had been mounted backwards for the push tests, and all of the push/pull tests, too. Seven years of college, er, four hours of testing down the drain. At least the Delta tests were valid.
I reversed the fan and ran all the tests again for the 3.77GHz push and push/pull setups, using just IBT this time. The following night I ran the tests for the 3.3GHz dummy OC. The previous anomaly held, in that the higher clocked CPU was cooler than the dummy.
Anyway, onto the results. I was surprised at how little of a difference the temps were between the varying setups. This admittedly non-scientific experiment seemed to overturn convention wisdom, in that push/pull wasn’t necessarily better than push, that 1200 rpm was the sweet spot for the Megahalems and that a high static pressure fan would greatly out-perform a plain old case fan.
The data:
