Water Cooling 3800 CL14 16x2 Memory for Zen 3 - Results

OC2000

OC2000

OC2000

Associate
Joined
2 Oct 2009
Posts
789
Having been using the G.Skill Trident Neo 3800Mhz CL14 @1.5V for awhile now together with an Asus Strix 3090 OC it has become apparent the heat generated by the 3090 is impossible for the Ram to maintain a decent temperature without the need to underclock / under volt the 3090 significantly. This applies to any game I have played.

I am using the O11 D XL case with plenty of air being driven in via 6 fans and 4 fans placed to provide exhaust. I was continually seeing the Ram hitting close to 50C over a short period of game play with no signs of the heat stabilizing.

This will become an even bigger problem in the warmer months ahead. Ambient temp is around 14-16c in this room currently

I purchased the parts necessary to water cool the Ram and thought I would share my results.

To begin with Ill share my base line results.

This is 100% stable and works fine in any application I use, however when the 3090 is doing its thing the temps rocket up. I have simulated this with OCCT and Prime 95.

You can see with OCCT it takes less than 10 minutes to get the Ram to 50C. This is an open case test bench by the way, not the O11D which would be far worse.

vOCxHkd.png


You will see here it took less than 10 minutes to reach 50C. While simulated, the same pattern happens when playing any GPU intensive game.

k7mmGvt.png


wfQAb8a.png


Next we will go through the setup
 
Last edited:
To test the effectiveness I’ve set up a test system in an old core X9 of mine.

The main parts I’ll be using are:
- Asus Crosshair VIII Hero Dark with mono block
- 1080Ti with water block
- HW Labs 480 SRC2 with 4x 1800RPM fans
- mcp35x2 pump
- AX1500i PSU
- G Skill trident neo 3800 CL14 16x2gb kit.

- aftermarket GELID 15Wm/k 0.5mm thermal pads
- kryonaught thermal Paste

removing the heat sinks was easy. Run OCCT for 10 minutes and the heat sinks peel away with no effort leaving no residue. I had a heat gun at the ready, but it wasn’t needed due to the heat these give off.

oYrXI7Yh.jpg


qQNQru3h.jpg


I added some thermal paste between the Ram plate connections

MBNCqVth.jpg


trMSonnh.jpg


While I have time on my hands I’ll test the following.

Ram without water block
Ram with water block no water
Ram with Thermal Pads under Water block
Ram at 1.5v with water
Ram at 1.6v with Water

I’ll then try optimize my timings more

more to come
 
Last edited:
I'll summarise the results at the end of this post, but will go through my findings now with some images and explanations.

The first test I did was using the aftermarket heatsinks vs the G.Skill Stock heat sinks.

dVYlKNzh.jpg


uVCJSbD.png


After 15 Minutes of OCCT, the ram topped out at 47.9C

Just the aftermarket heatsinks alone (using 12 W/mk thermal pads) kept the Ram under 50. By this time it almost stabilised in temperature and had only risen a few .0x C between the 10 minute and 15 minute mark.

The Second Test I did was adding a water block on to it and testing the the water block temperature for heat transfer.



RHeCkrmh.jpg


RmG6ulK.png


After 15 Minutes with OCCT the Ram temp was 45.3. This time both Ram sticks were closer in temperature too. The water block Reach 34C

I decided to then try adding some thermal padding between the water block and ram sticks to see if heat would transfer more efficiently.

JHul3Lfh.jpg


6UhUfb2.png


The Water block Reached 32C (cooler by 2C) and Ram was hotter at 46.2C, making thermal pads not worth doing.

Next Test Was Ram with the water block at 1.5V.

Before the testing the results I checked the change in water flow.

CgsZcShh.jpg


vs

4NqxmZrh.jpg


3.83l/min vs 2.76l/min

How accurate this is subject to air bubbles etc and flow generally increases after a period of time, but there was roughly a 1 l/min loss in flow rate adding a Ram water block. Whats worth pointing out that after doing this, adding a block onto the back of the 3090 backplate resulted in no loss of flow rate.

Test Results.

DsMuUBeh.jpg


Ib5L3eJh.jpg


68UWTtB.png


After 15 Minutes of OCCT at 1.5V the ram was sitting comfortably at 27.8C. The water was at 16C giving it a Delta of 11.8C

The Last test was running the Ram at 1.6V

ETk57dr.png


Very little change. The Ram rose from 27.8 to 28.3C

To Summarise the results

Stock Ram - 50c within 10 minutes

After market Ram heatsink without water block - 47.9C

Ram with water block no water - 45.3C

Ram with Thermal Pads under Water block 46.2C

Ram at 1.5v with water 27.8C

Ram at 1.6v with Water 28.3C

Flow rate reduced by approx 1l /min (maybe due to air bubbles)

When testing GPU intensive applications which increased the water to about 27C The Ram maintained a temp around 29-30C

Next up will be optimising the ram settings.
 
Last edited:
My goal was to try get a stable 4000 mhz / 2000 FCLK, however no matter what settings I used even at 1.6V I still got WHEA errors. I think the BIOS or AGESA still needs to mature. Apart from the WHEA errors though the system still booted and seemed stable.

Using Asus Crosshair Dark Hero BIOS version 3003 I was able to achieve the below results at 1.54V. I then updated to the latest Bios with AGESA 1.2.0.0.0 (3204) which caused 69000+ WHEA errors in 2 minutes. The same resulted in the original stock timings. I had to revert back to the stable 3003 BIOS based on AGESA 1.1.9.0.0

IehKtYZ.png


I decided to then try my old Gigabyte Aours Extreme rev 1.0 X570 motherboard with the latest F33a Bios based on AGESA 1.2.0.0.0 which Ran the following Ram settings without any issue. Unfortunately even 2000 FCLK on this BIOS resulted in fewer WHEA errors, but still a no go.

For some reason AGESA 1.2.0.0.0 has increased the latency in Ram too.

The below was running at an All core 4850 / 4800 at 1.38V too.

Final Settings @1.54V

Dbikqh1.png


Ram never exceeded 29.4C and ran the full OCCT test without errors.

I have now shelved the Asus Crosshair Dark Hero and will be returning to the Gigabyte Aourus extreme permanently. This board also scored higher in every CPU test I had tried with the Asus one and ran a lot cooler too.

Ill be posting pics of the final setup in my Project log in the project log section of this forum and will report on the ram temps on this thread after a long gaming session that used to top the ram out at close to 50C as well as the flow changes from my single D5 pump.
 
Last edited:
I'm wondering why no one ever put out a single rad mini AIO watercooling system just for RAM, like fix smallish pump by the radiator, and you are good. And heatsinks would be able to fit majority if not all DIMMS on the market. I know thermaltake has CPU/RAM combo cooler (which is quite expensive), but why not separate? Unless watercooling RAM in normal environment does not give any performance/OC improvements
 
I’ve got all the results, but got stuck doing the water flow tests for the graphics card and then having to re apply my old water block as the Aqua computer one was rubbish.
I’ll post it all tomorrow morning
 
Got a chance today to play a couple of hours of Fortnite with my kids.

my settings were 1440p no DLSS mostly epic settings at a sustained 240 FPS using the Samsung odyssey 32G7

my Strix 3090 was water cooled with the EK block with the VRM thermal pads removed from the back plate to eliminate coil whine and an MP5works serial version.

it was undervolted to [email protected] constant.

Ram was water cooled and set to the specs shown in the post above as final settings.

Ryzen 5950X running at stock as I haven’t worked on that yet.

total System Draw ranged between 450-600W

room temp was 20C. Flow Rate 3.8 liters per minute and water reached 29-30c. This was roughly the same in both pre water cooling Ram and post.

pre water cooling the Ram and with the parallel mp5works my GPU reach around 45C and both Ram reached 48C after an hours play. The Ram was running at looser timings @1.5v

U0GhU1T.jpg


with the new setup I took more readings.

Max Water temp -30C
Max CPU temp - 66C
Max GPU temp - 39C
Max GPU mem temp - 58C
Max GPU VRM temp - 45C
Max system Ram temp - 35C

0 WHEA errors vs the 20 or so I got before after 2 hours of Fortnite.

while Fortnite isn’t the most stressful of tests, it’s the only game I have installed at the moment, and something I could directly compare to from previous readings. it will be interesting to see what Cyber Punk will do especially when the GPU is Overclocked to 2100+.

Overall very happy with these results and glad I finally ventured into Ram water cooling. I would definitely recommend it.
 
@OC2000, you're a legend! I was trying to find info on watercooling ram after catching a bit of a Buildzoid video (https://youtu.be/7593hzYPeYs) where he said how a Thermaltake AIO that cooled CPU and ram would be no good for ram because the water would be too hot in an AIO with their weak pumps, but I was wondering how it would work in a loop with relatively cool water, and now I know! Thank you!

I think he also mentioned in that video that a tiny loop just for ram could be interesting, actual heat output of ram is quite low, 20W or so. Could be a cool project if a suitably small pump for a 120 radiator could be found.

So from your last post, above, your flow rate returned to 3.8L/min with the ram block installed? And it's a single D5 you're using?

You've now got me thinking about the EK nickel ram waterblocks. I am a touch worried how my EK D5 pump will handle that plus CPU, GPU, and 3 radiators though.
 
TT158 said:
@OC2000, you're a legend! I was trying to find info on watercooling ram after catching a bit of a Buildzoid video (https://youtu.be/7593hzYPeYs) where he said how a Thermaltake AIO that cooled CPU and ram would be no good for ram because the water would be too hot in an AIO with their weak pumps, but I was wondering how it would work in a loop with relatively cool water, and now I know! Thank you!

I think he also mentioned in that video that a tiny loop just for ram could be interesting, actual heat output of ram is quite low, 20W or so. Could be a cool project if a suitably small pump for a 120 radiator could be found.

So from your last post, above, your flow rate returned to 3.8L/min with the ram block installed? And it's a single D5 you're using?

You've now got me thinking about the EK nickel ram waterblocks. I am a touch worried how my EK D5 pump will handle that plus CPU, GPU, and 3 radiators though.
Click to expand...

To maintain over 2l per min on that system I had to increase my D5 from 3600 rpm to 4000 and then when gaming I set it so that it runs full speed when gpu temp goes over 30c which brings it to over 3l a min. You can’t hear it due to the music etc. Without the Ram water block and MP5works in serial I got just over 4l per min with a single D5 pump.

By contrast my test setup I used an MPC35x which is basically 2x DDC pumps which could do 10l per min without the Ram and MP5works and 8.95l with at full power.

As long as you can maintain over 2l or 0.6 gallons per minute you are fine. I’ve run dual loops before and while it can’t hurt to add a second D5 in if you can fit it for quieter operation or redundancy, it’s not 100% necessary
 
Thanks for the info.

The EK D5 I have is actually barely audible at max speed, even when I was leak testing, I don't mind running it at full speed.

I've just pulled the trigger on the EK stuff. Wish I had 2 dimms instead of 4, but oh well. The temp probe I'd stuck between the dimms has been reading quite toasty after installing my current loop (which admittedly is not an ideal setup). Looking forward to the results in the new case with an extra rad the ram block :)

Thanks again, OC. Love your work!
 
You must log in or register to reply here.
Back
Top Bottom