Thermal Paste and Fan Upgrade Improvements

[Egad]

I wanted to share my results with doing some rig maintenance in hopes that it may be of values to others now or in some nebulous future when some is searching around.  

 

When I first built my rig I was admittedly pretty lazy, just used the various stock fans, manufacturer paste applications, etc.  Setup was:

 

• Case: Noctis 450
• Front Intake Fans: 3x120mm of whatever NZXT fans came with case
• Rear Exhaust: 1x140mm whatever NZXT fan came with the case
• CPU Cooler/Top Exhaust:  H110i GTX (280mm AIO) with stock Corsair static pressure fans/pre applied Corsair paste cooling a 5960X OCed to Core: 4.5 GHz (1.282 volts), Cache: 4.25 GHz (1.24 volts)
• GPUs: 2x EVGA 980Ti with ACX 2.0 coolers running at EVGA provided speeds

 

Long story short nothing thermal throttled, although in games that did thread the Corsair fans started to sound like my computer was taxiing for takeoff.  Since I wear over the ear headphones will gaming though, meh.

 

However as time when on I found more of an urge to OC the 980Tis beyond what EVGA had done and I found once I had them up to 1511 MHz on the Core and 3954 MHz on the memory, I'd hit 85 C much faster than I wanted to do.  Since I kind of had the urge to tinker I decided to get new fans and repaste everything, but to do so incrementally and do a bit of measurement.  I used GTA V for everything because it and Witcher 3 are the most demanding games I play and Witcher 3 did ship with a benchmark.  My third most demanding game would be Borderlands 2 and yeah I was doing just find there before the overclocking.  A note I built the computer in August of 2016 so the paste was ~26 months old.  

 

One other item of note:

 

When running GTA V I'd have EVGA Precision, Intel XTU, CPU-Z and Task Manager open on the second monitor.  Prior to the memory OC of the'd see 8.4 to 8.6 GB of system RAM be used, varied from bench to bench).  After the OC, with the same drivers, I'd see 12.4 to 12.7 GB of RAM used during the bench.  Given nothing else on the system had changed I found that interesting, perhaps the faster GPU was keeping the system cpu/memory controller busier so it didn't have as much time to garbage collect.  

 

Round 0

Ambient Temp: 20C

• 5960X Stable Temp running Intel XTU Stress Test and Furmark running (100% load on all 16 logical processors): 88C 
• 980Ti results:  Rapidly throttle down to 1488 MHz and hit 85 C midway through the fighter flight part of the GTA V benchmark  
• Subjective Notes:  Furmark + XTU Stress was load, even through over the ear good quality headphones, you could really feel the warm air coming out of the top of the case

 

Round 1, Replacing Thermal Paste

I repasted both of the 980Tis and the CPU with Arctic Silver MX-4.

Ambient Temp: 20C

• 5960X Stable Temp with XTU Stress Test and Furmark: 84C
• 980Ti results: Never went above 78C, GUP clock stayed above 1500 MHz
• Subjective Notes:  Still pretty loud, while I probably could have messed with the GPU fan curves a bit, the Corsair fans are the main noise culprit and thus it's not really worth it
• Practical Notes: 980Ti SLI means GTA V at 4K ultra settings posts a min fps of 58.3 for the benchmark and an average FPS of 78

Or in summary, I paid 5.99 for the MX-4, plus the cost of some rubbing alcohol and wipes to drop 7C off my GPU temps.  I'm glad I did this over spending on some kind of conversion kit to let me mount AIOs onto my GPUs.

 

Round 2, New Fans

• Front Fans: 3x120 Notcua NF-S12 PWM (120mm airflow fans)
• Rear Exhaust: 1x Noctau NF-A14 PWM Chromax  (140 static pressure)
• Top Exhaust/AOI: 1x NF-A14 Industrial, 1x NF A-14 Chromax (140 static pressure)

Note: Two different A-14s ordered, to get some brown anti vibration pads

 

Ambient Temp: 23.33C (my wife came home and demanded to know why the thermostat was set so low)

• 5960X Stable Temp with XTU Stress Test and Furmark: 85C
• 980Ti results: Never went above 78C, GPU clock stayed above 1500 MHz, over repeated runs I was able to get stable at 1522 MHz GPU clock at 78C
• Subjective Notes:  Much quieter at load.  Significantly more air seems to exhaust out the back, whereas originally it was pretty hot air coming out the top and more or less room temp air coming out the back, now I feel warm air coming out the rear exhaust and roughly the same temp air coming out the top.  No noticeable temp difference.  
• Practical Notes: Considering the ambient temp bump up, the new fans did contributed to temps, although their main contribution was with regard to noise.  It seems the NZXT and Corsair fans were moving air, they just liked announcing to everyone on the first floor how hard they were working.  The biggest wins are the removal of those Corsair fans and the fact my GPU fans aren't cycling as hard since the GPU fans are now the nosiest in the build.

 

Bonus Round

Round 2, but with only the middle front intake fan on (the one blowing on the GPUs)

Ambient Temp: 23.33C

 

• 5960X Stable Temp with XTU Stress Test and Furmark: 85C
• 980Ti results: Never went above 79C, GUP clock stayed above 1500 MHz
• Notes: This more or less replicates Luke's work that one intake in the front is typically enough.  I would note though though that my GPU fans were working harder when I watched the fan speed reports on Precision during the test and the noise increase was noticeable.  

 

Final Thoughts:

 

Replace manufacture thermal paste, especially old stuff.  I want to credit EVGA for having a decent application on both 980Tis (especially compared to some OEM jobs I've seen where it looks like a sex crazed maniac ejaculated thermal paste all over the card), so it was either quality or age that was playing the largest role in the temps.

(Paste as applied by EVGA)

 

The results on the fans are nice, but not worth tying up significant money in unless your case fans are much worse or much louder than mine.  I'm happy with the result and scratched my itch to open up the case and tinker with my case.  Although I probably could have gone with just one front 140mm fan.  The third 120mm fan is pretty much a waste since it's so low down all it does is blow air over the ugly nest of cables in my case's basement.  The midlevel fan pretty much aims directly at both GPUs and is doing the lions share of the work for cooling them.  

 

Hopefully this helps some other folks, now back to looking at full loop hard tube cooling and telling myself "No, you don't need that."
 

 

 

[Damascus]

27 minutes ago, Egad said:

"No, you don't need that."

You don't, but if you want to get into it on the cheap I'm your friendly neighborhood parts dealer (I have 2 DDC, 1 D5, 2 240mm and 1 360mm, 14 3/8, 5/8 fittings and 3 reservoirs )

[markr54632]

Okay, so problem number 1 is your choice of thermal paste. Artic mx4 is fine for ihs to heatsink. It is not okay to use for bare die applications. It is actually the worst performer. Temps will be okay for 1-2 days. Within a week you will be back to the start.

 

As for cheap quiet fans I like the arctic fans.

 

[Egad]

7 hours ago, markr54632 said:

Okay, so problem number 1 is your choice of thermal paste. Artic mx4 is fine for ihs to heatsink. It is not okay to use for bare die applications. It is actually the worst performer. Temps will be okay for 1-2 days. Within a week you will be back to the start.

 

As for cheap quiet fans I like the arctic fans.

 

I know people talk about different time frames for MX-4 to degrade, be it weeks, months, etc but I've repasted five GPUs with MX-4 and never personally seen that behavior.  I have an 18 month old repaste job on a Titan Black in my Plex server that gets hit for GPU assisted transcodes daily and it's still showing the same temp improvements from when I pasted it, so I expect similar behavior out of the 980Tis, we'll see though (and it got worked under continuous load for two weeks while I transcoded most of my library).  I mean I did the 980Tis five days ago already and I've seen no temp change under load in games, never breaking 79 and in the 60s for less demanding titles.

 

End of the day I could carefully apply some liquid metal or grab a tube of Gelid for ~2.5x what I paid for the MX-4 or frankly if the MX-4 degrades after X months in it's about a dozen screws and five minutes of work to reapply during a regular dusting of the rig internals.  

[Cyracus]

liquid metal isn't advised for gpus because exposed electrical stuff around the die, granted I've heard some suggestions of using clear nail polish or something like that to coat all that to make liquid metal less risky