Changed H100 all-in-one to a custom water loop i7 3970X @4.6 GHz temperature and power test results before and after

[Tam3n]

My old Corsair H100 all-in-one liquid cooler was reaching it's limits with the I7 3970X running at 4.6 GHz, so I decided it's time to cool my cpu proper

 

I've had my R9 290s under water with 480 mm radiator for a couple of years (did that one pretty quick lol), but the cpu has always been in it's own loop with H100. I had I7 3820 before for a long time (ran it @4,3 GHz), but recently I swapped it to an old 3970X from ebay. This thing is power hungry on a complete different level. While the H100 certainly could cool it, I decided to finally add the cpu to the existing loop (and about time, I had 4x 10/16mm fittings lying in a box reserved for the additional radiator and cpu block to come for almost two years). So I also added a 360 mm radiator to the loop while I was at it. Sandy Bridge Extreme series processor still have soldered ihs and I read online the H100 flow rate really wasn't much to write home about (you could feel the inlet side of the radiator be much hotter than the outlet), so I was expecting some results.

 

For reference, here's an old pic of the PC with H100:

Spoiler

 

And here's now (I'm still leak testing so the side panel is off - I just completed putting it together and ran the tests):

Spoiler

 

So here are relevant setup details:

 

Old cooling setup:

I7 3970X @ 4.6 GHz, Vcore ~ 1.37 V (offset mode, so voltage and clocks go down in idle)

CPU cooling: Corsair H100, push/pull with 4x noctua fans

Tim: GC- Extreme

GPUs in their separate loop

 

New:

I7 3970X @ 4.6 GHz, Vcore ~ 1.37 V (offset mode, so voltage and clocks go down in idle)

Radiators: 360mm (3x noctua fans pull) + 480 mm (4x Alphacool NB-eLoop 1200rpm push)

Pump: EK-5D Vario (X- RES 100)

CPU block: Alphacool NexXxoS XP³ Light - Acetal

Tim: GC- Extreme

(GPU blocks: EK-FC R9-290X - Acetal+Nickel (Rev.2.0))

Loop in series: pump -> GPU2 -> GPU1 -> 360mm -> 480mm -> CPU -> pump, tubing ~ 3m

*added a Noctua fan to blow at the back side of the cpu socket (to provide more cooling to the backside vrms on the motherboard)

 

Rest:

Spoiler

Motherboard: Asus P9X79

RAM: Kingston HyperX 32 GB 1866 MHz DDR3 @ 1.65 V (two different 4x4GB kits together)

GPUs: 2x Gigabyte R9 290OC (Corsair Link software wakes the gpu2 from sleep, so both are idle in these tests)

Storage: Kingston HyperX 256 GB, 3x Seagate 1 TB HDD

PSU: Corsair AX860i

Case: Fractal Design R3 (not meant for this level of water cooling :D)

Case fans: 2x 120 mm low speed intake front, 1x 120 mm semi-intake back, 1x 140 mm low speed intake bottom, *1x 120 mm intake backside, exhaust thru the 360 rad on the top

OS: Windows 10, 64 bit

 

 

I ran three tests on both setups:

1. Idle at desktop (system has been heated up and running at least 30 min before the test)

2. Blender (2.78.0 latest build) BMW 3D rendering benchmark scene (realistic 100% load on the cpu)

3. Prime95 (28.10) AVX small FFT (maximum CPU heat and power)

 

In all cases the room ambient temperature was 21 C and all test were ran with fans and pump at full speed (to keep things simple for this). Crapload of data was logged with Corsair AX860i PSU and Corsair Link 4 software to spread sheets. Focus will be on CPU core temperatures and system power.

 

Below are the temperature and power graphs from these tests. Results with H100 are first, and CL (= custom loop) follow in each test scenario.

 

1. System idle (mostly):

Spoiler

 

 

H100 CPU cores avg: 30 C

Custom loop CPU cores avg: 27 C

PSU reported idle power is about 150 Watts in both cases (system was doing something during the jumps)

 

2. Blender BMW rendering benchmark scene:

Spoiler

 

 

 

 

H100 CPU cores avg reaches: 72.5 C

Custom loop CPU cores avg reaches: 49 C

PSU reported power is about 400 Watts in both cases, but it starts to slightly increase with H100 to around 420 W as components heat up more.

 

3. Prime95 AVX small FFT:

Spoiler

H100 CPU cores avg reaches: 83.5 C, hottest core: 88 C

Custom loop CPU cores avg reaches: 57.5 C, hottest core: 62 C

Things to note: H100 setup severely throttles at around 310 secs (either processor Tcase max 66.8 C is reached, or motherboard VRMs overheat) and test is soon stopped. The torture test is run longer on the custom loop while no throttling occurs.

We start to see how heat affects efficiency, H100 system power: 450 Watts, custom loop: 420 Watts

 

As a side note, the cpu seems to be pulling quite nice amps through the additional CPU power (8-pin pcie, marked yellow in the graph):

Spoiler

 

So there you have it, and I'm extremely pleased with the results!

 

And one last thing, I noticed that the CPU block contact surface was not entirely flat, but it was slightly raised in the center and edges. I considered milling it it flat, but the results are looking good, so I'll let it be

 

I hope someone found this info interesting and if you want to share your experiences, please feel free to post below!