Memory needs cooling, too

Mark Rose · 2017-12-14, 04:07

So I recently had a machine produce a bunch of bad results. The culprit? Insufficient RAM cooling. The memory is cheap stuff with no heat spreader, and went I installed another case fan (ironic), the fan cable was placed in such a way to block airflow over a couple of the chips. I moved the cable and now it has returned a good result.

Something to be mindful of in the future.

henryzz · 2017-12-14, 11:22

Quote:

Originally Posted by Mark Rose

So I recently had a machine produce a bunch of bad results. The culprit? Insufficient RAM cooling. The memory is cheap stuff with no heat spreader, and went I installed another case fan (ironic), the fan cable was placed in such a way to block airflow over a couple of the chips. I moved the cable and now it has returned a good result.

Something to be mindful of in the future.

If you are willing to add your own cooling to memory I would imagine that a large amount of cheaper kits would have overclocking room. I suspect that is the only difference between some ranges.

mackerel · 2017-12-14, 13:45

What spec ram is it? I'd be surprised if non-extreme ram would be a problem. On some of the higher spec (but not extreme) modules I have which can report temperature, they typically sit in the mid 30's under load. On a server ECC ram I've seen around 50-60C using infrared imaging which still isn't anything to be worried about. Unless there is a SSD-like mechanism going on where their controllers overheat easily, leading to throttling. Maybe it isn't the ram itself but some other logic that is getting too stressed.

preda · 2017-12-14, 14:05

Quote:

Originally Posted by Mark Rose

So I recently had a machine produce a bunch of bad results. The culprit? Insufficient RAM cooling. The memory is cheap stuff with no heat spreader, and went I installed another case fan (ironic), the fan cable was placed in such a way to block airflow over a couple of the chips. I moved the cable and now it has returned a good result.

Something to be mindful of in the future.

I recently had a situation that looked like RAM that was previously good becoming bad. But no, it was in fact a problem with the RAM contacts, because re-sitting the RAM sticks fixed it good.

Another approach is to use the new PRP with the Gerbicz error check instead of LL, and detect+correct any RAM errors early and reliably.

heliosh · 2017-12-14, 14:06

Higher temperatures lead to higher noise on the signal, which leads to higher error rate. Depending on how close you are running at the limit, temperature may make a difference or not.

RAM with heatspreader are not neccessarily a good thing. Sometimes they are just for design and mounted with glue (thermal insulator) rather than thermal compound.

Since I've overclocked my RAM (DDR3 1600 to 2000) I just took a thermal image. The RAM is around 60-65°C. The only thing I'm a bit worried is a bypass capacitor that is at ~80°C. Mprime seems to stress the RAM pretty hard.

Mark Rose · 2017-12-14, 21:23

It was a 32 GB 2x16 kit of G.Skill Aegis DDR-2133. Cheap stuff.

The RAM was reliable in that system for over a year.

I should break out the heat gun tonight to see what temperatures I can find. I don't have a thermal imaging camera, unfortunately.

Mark Rose · 2017-12-14, 21:24

Quote:

Originally Posted by preda

Another approach is to use the new PRP with the Gerbicz error check instead of LL, and detect+correct any RAM errors early and reliably.

I do this on a flaky machine I have, but the others are all running DCLL.

kladner · 2017-12-14, 22:20

My 2x8 GB, 2666 MHz HyperX, overclocked to 3200 MHz, voltage boosted from 1.2 to 1.22. Using 6 W, running at 48 C. Note that they are edge-on to one of the in-blowing radiator fans in the top of the case.

ewmayer · 2017-12-14, 22:36

I leave the room-facing side panel off my ATX case to help with ambient airflow. I wonder if one could reverse-face one of the case fans to blow air in toward the RAM if the latter is running too hot?

Mark Rose · 2017-12-15, 02:48

Quote:

Originally Posted by henryzz

If you are willing to add your own cooling to memory I would imagine that a large amount of cheaper kits would have overclocking room. I suspect that is the only difference between some ranges.

Naw, the CPU cooler fan pulls enough air over the RAM, now that the fan cable isn't covering a couple of the chips.

2017-12-14, 04:07	#1
Mark Rose "/X\(‘-‘)/X\" Jan 2013 https://pedan.tech/ 2⁴·199 Posts	Memory needs cooling, too So I recently had a machine produce a bunch of bad results. The culprit? Insufficient RAM cooling. The memory is cheap stuff with no heat spreader, and went I installed another case fan (ironic), the fan cable was placed in such a way to block airflow over a couple of the chips. I moved the cable and now it has returned a good result. Something to be mindful of in the future.

2017-12-14, 14:06	#5
heliosh Oct 2017 ++41 175₈ Posts	Higher temperatures lead to higher noise on the signal, which leads to higher error rate. Depending on how close you are running at the limit, temperature may make a difference or not. RAM with heatspreader are not neccessarily a good thing. Sometimes they are just for design and mounted with glue (thermal insulator) rather than thermal compound. Since I've overclocked my RAM (DDR3 1600 to 2000) I just took a thermal image. The RAM is around 60-65°C. The only thing I'm a bit worried is a bypass capacitor that is at ~80°C. Mprime seems to stress the RAM pretty hard. Attached Thumbnails Last fiddled with by heliosh on 2017-12-14 at 14:39

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Recommended Cooling Unit	lukerichards	Hardware	15	2018-01-30 20:51
GTX 980 Ti water cooling	wombatman	Hardware	2	2017-06-27 15:01
Laptop cooling	ATH	Hardware	33	2015-05-27 07:44
Air or water cooling?	schickel	Hardware	60	2011-12-09 17:48
Water Cooling.	nomadicus	Lounge	9	2003-08-08 15:28

2017-12-14, 13:45	#3
mackerel Feb 2016 UK 2⁶·7 Posts	What spec ram is it? I'd be surprised if non-extreme ram would be a problem. On some of the higher spec (but not extreme) modules I have which can report temperature, they typically sit in the mid 30's under load. On a server ECC ram I've seen around 50-60C using infrared imaging which still isn't anything to be worried about. Unless there is a SSD-like mechanism going on where their controllers overheat easily, leading to throttling. Maybe it isn't the ram itself but some other logic that is getting too stressed.

2017-12-14, 21:23	#6
Mark Rose "/X\(‘-‘)/X\" Jan 2013 https://pedan.tech/ 2⁴×199 Posts	It was a 32 GB 2x16 kit of G.Skill Aegis DDR-2133. Cheap stuff. The RAM was reliable in that system for over a year. I should break out the heat gun tonight to see what temperatures I can find. I don't have a thermal imaging camera, unfortunately.

2017-12-14, 22:20	#8
kladner "Kieren" Jul 2011 In My Own Galaxy! 2×3×1,693 Posts	My 2x8 GB, 2666 MHz HyperX, overclocked to 3200 MHz, voltage boosted from 1.2 to 1.22. Using 6 W, running at 48 C. Note that they are edge-on to one of the in-blowing radiator fans in the top of the case.

2017-12-14, 22:36	#9
ewmayer ∂²ω=0 Sep 2002 República de California 10110111101100₂ Posts	I leave the room-facing side panel off my ATX case to help with ambient airflow. I wonder if one could reverse-face one of the case fans to blow air in toward the RAM if the latter is running too hot?