DDR5 active cooling.

[User5566]

Hi,

In general I have pretty good airflow in my case, but still I'm getting ddr5 temps of +55C during heavy use and +70C during stress testing. And that's below their xmp profile speed.

I've asked about it before and was told not to worry, because ram should be fine at that temperatures, but I also found lots of online sources claiming dram becomes far less stable above 60C and best temps for it are below 45C (my ddr5 idles at 48C). I even repeated the stress test with no gpu (with built in igpu) and the temps are the same.

So I'm looking into active cooling. This is g.skill ddr5-6400 48gb*2 kit with hynix chips. I'd also like to OC the ram a bit more to their rated 6400MT with 1:1 uclk:mclk. If I do that without active cooling they'll get to +75C very quickly during a stress test and error there. I haven't even thought about raising voltage further (from 1.365v) at these temps.

Alphacool is selling individual waterblocks for ram sticks for €20, but including fittings, extra tubing etc the total cost for 2 sticks (to add to existing loop) is more like €75eur + postage. Also I'm being shown in hwinfo these sticks are pulling 5W each. Seems like a huge overkill to watercool these and spend close to a €100 doing it. It sure would be neat, but I'd like to get other people's opinions on how much of a difference sticking a 80mm noctua fan on top would do(on a 3d printed bracket) ? Assuming the fan runs around 1k rpm only.

Would that keep these temps below 60C in a stress test or is watercooling them a way to go?

 

[EarthDog]

I run my ddr5-7200 (6800 xmp) kit at 1.45v with temps peaking in the 70s.... all good here for what I do.

Maybe I'm off, but, to match those clocks 1:1, you deal with the SOC voltage which shouldn't raise temps on the sticks as SOC is on the CPU. I never tried at that high, so not sure.

Air over the sticks would be plenty to knock temps down...but to your goldilocks of 60C... not sure as there are so many factors involved. But surely I'd spend $20 on a fan and print a bracket (or buy an existing kit) before I watercooled ram, period.

 

[Woomack]

I've asked about it before and was told not to worry, because ram should be fine at that temperatures, but I also found lots of online sources claiming dram becomes far less stable above 60C and best temps for it are below 45C (my ddr5 idles at 48C). I even repeated the stress test with no gpu (with built in igpu) and the temps are the same.

Most people who post that info don't really test the hardware and only repost BS that others said. A lot of posts like that are on OCN or TPU (including reviews). I don't even reply to anyone anymore, as it's a waste of time.
Some read tips made by extreme overclockers and take them as a rule for regular home PCs at ambient temps.
Temperatures matter, but there are many variables that affect stability, not only temperatures.

DDR5 has a tolerance of 85°C+ (I don't remember exactly, but it should be 100°C or something), but at JEDEC specs, not overclocked. With higher frequency, stability drops at higher temperatures. Current kits are designed for 5600, so 6000 is hard to call overclocking. Up to 1.45V shouldn't be problems on air, unless the RAM manufacturer fails something or you run it heavily loaded for long hours, on a 2-slot mobo in a tight SFF PC with no forced airflow.

At 6400 1:1 with high-density RAM you are asking for troubles. It barely ever works, that's why 2x32GB and 2x48GB AMD memory kits have EXPO at 5600/6000.

I see you are searching for issues that don't exist. Start to think about liquid RAM cooling when you run your sticks at 1.5V+, or try to push it for something like 6400 CL26 or 8000 CL32.
High-density modules dislike high voltages, so you will be forced to keep lower voltages anyway.

 

[User5566]

Most people who post that info don't really test the hardware and only repost BS that others said. A lot of posts like that are on OCN or TPU (including reviews). I don't even reply to anyone anymore, as it's a waste of time.
Some read tips made by extreme overclockers and take them as a rule for regular home PCs at ambient temps.
Temperatures matter, but there are many variables that affect stability, not only temperatures.

Agreed, I've run gddr6 at 95C for long enough in GPUs.

DDR5 has a tolerance of 85°C+ (I don't remember exactly, but it should be 100°C or something), but at JEDEC specs, not overclocked. With higher frequency, stability drops at higher temperatures.

OK.

Current kits are designed for 5600, so 6000 is hard to call overclocking. Up to 1.45V shouldn't be problems on air, unless the RAM manufacturer fails something or you run it heavily loaded for long hours, on a 2-slot mobo in a tight SFF PC with no forced airflow.

This above is about 16gb modules right?. I have 48gb modules.

At 6400 1:1 with high-density RAM you are asking for troubles. It barely ever works, that's why 2x32GB and 2x48GB AMD memory kits have EXPO at 5600/6000.

So you're essentially saying no matter the temps running these high density modules at 6400 cl32 at 1:1 is a pipe dream, right? So why are they sold with 6400mt expo?is it just bullshit? Or are you saying my instability at 6400mt 1:1 is a cpu (soc) problem?

I see you are searching for issues that don't exist. Start to think about liquid RAM cooling when you run your sticks at 1.5V+, or try to push it for something like 6400 CL26 or 8000 CL32

How am I searching for "issues that do not exist"?

The fact my high density ram crashes when it gets above 75C at certain frequency does exist. I'm not imagining it. I'm interested in if it can be run like this in a stress test if cooled properly.

Low density sticks are not what I'm interested in.

High-density modules dislike high voltages, so you will be forced to keep lower voltages anyway.

At the moment I'm running at 1.365. My plan is not to exceed 1.4 as I read on the ddr5 o guide on this site these hynix modules don't like above 1.4v.

Coming back to the subject of RAM cooling. I might do a test with just a desk fan later today. I have a feeling these will work fine at 6400mt 1:1 if kept below 50C. But that's just a feeling.

 

[Woomack]

This above is about 16gb modules right?. I have 48gb modules.

No, this is about 48GB modules. Most 2x48GB kits are designed for 4800-5600, and new ICs are actually designed for 5600 SPD/JEDEC at CL46 1.1V.

So you're essentially saying no matter the temps running these high density modules at 6400 cl32 at 1:1 is a pipe dream, right? So why are they sold with 6400mt expo?is it just bullshit? Or are you saying my instability at 6400mt 1:1 is a cpu (soc) problem?

I can't see the G.Skill 2x48GB 6400MT/s kit designed for AMD listed anywhere. It's not on the G.Skill website. There are only XMP memory kits up to 6800MT/s. The highest 2x48GB memory kit validated on AMD motherboards is 5600MT/s. The highest 2x32GB is 6000MT/s.

How am I searching for "issues that do not exist"?

The fact my high density ram crashes when it gets above 75C at certain frequency does exist. I'm not imagining it. I'm interested in if it can be run like this in a stress test if cooled properly.

Low density sticks are not what I'm interested in.

Put a high RPM fan over RAM, and it will cause a significant temperature drop. It's the easiest way to check if it's worth cooling it more or not.

At the moment I'm running at 1.365. My plan is not to exceed 1.4 as I read on the ddr5 o guide on this site these hynix modules don't like above 1.4v.

Coming back to the subject of RAM cooling. I might do a test with just a desk fan later today. I have a feeling these will work fine at 6400mt 1:1 if kept below 50C. But that's just a feeling.

If you mean the guide about DDR5 Overclocking on the front page, it's about the first Hynix M-die, and the article is not even finished as there was no proper data to do all the tests. New Hynix M-die in 24/48GB modules are something else - need lower voltages and overclock better.

 

[User5566]

No, this is about 48GB modules. Most 2x48GB kits are designed for 4800-5600, and new ICs are actually designed for 5600 SPD/JEDEC at CL46 1.1V.

They may be designed for 2000MT,but if g.skill binned them and says they are capable at working at 6400MT,cl32 at 1.35V and ASUS has them on their QVL for my AM5 MB runnining at 6400MT,cl32. What does it tell us? That they possibly can run at these speeds. And my kit has been happily running at 6200. 6400 too if it is kept under 70C.

I'm not quite sure what is the point you're making here. I haven't seen any test of 2x48gb kit by you or anyone else at these speeds.

I can't see the G.Skill 2x48GB 6400MT/s kit designed for AMD listed anywhere. It's not on the G.Skill website. There are only XMP memory kits up to 6800MT/s. The highest 2x48GB memory kit validated on AMD motherboards is 5600MT/s. The highest 2x32GB is 6000MT/s.

I said so before. My kit is validated at 6400MT by Asus, with AMD and my MB. See below.

ProArt X670E-CREATOR WIFI｜Motherboards｜ASUS Global

Built for creators, ProArt X670E-Creator WiFi features 16+2 power stages (70A), PCIe® 5.0, DDR5 support, USB4®, 10 Gb & 2.5 Gb LAN, WiFi 6E and four M.2 slots

www.asus.com

Put a high RPM fan over RAM, and it will cause a significant temperature drop. It's the easiest way to check if it's worth cooling it more or not.

Perhaps the fan I tried (an 80 mm noctua nf-a8) wasn't high speed enough, but it did pretty much nothing useful, as well as a desk fan used the same way. The RAM was entire 4C cooler.

Perhaps the temperature pads in these G.SKILL modules are not that great and/or the heatsinks are crap.

If you mean the guide about DDR5 Overclocking on the front page, it's about the first Hynix M-die, and the article is not even finished as there was no proper data to do all the tests. New Hynix M-die in 24/48GB modules are something else - need lower voltages and overclock better.

I'd love to read anything about overclocking high density modules on AMD and anything about people successfully cooling ram modules under 50C.

These modules are really only using about 5W of power. Then the LEDs are maybe adding 1W more. We should be able to cool them passively very easily. It speaks volumes about the crappy radiator design that favours aesthetics over function that a device 16x the size of a cpu consuming 5W can't be kept under 50C.

It is also possible the problem is not with the radiators, but in packaging of these chips, but to test this theory water cooled setup is necessary. I might get the Alphacool one.

 

[EarthDog]

It speaks volumes about the crappy radiator design that favours aesthetics over function that a device 16x the size of a cpu consuming 5W can't be kept under 50C.

Because wattage does not equal temperature. There are many variables involved.

I had a mining ASIC with 5W IC's that would melt your finger to it, lol. It needed a CPU-sized heatsink to keep it within operating params, lol.

Also, IIRC, the SPD hub/PMIC/inductors are what get toasty on DDR5, more so than the ICs. So w/e block you're buying for this endeavor needs to cool those bits too or you're throwing even more money down the tubes.

 

[MaddMutt]

I don’t have a ddr5 system BUT I have a lot of ddr3/ddr4 systems. These require removing the hs from the memory chips which voids their warranty. You also have to worry about ripping a memory IC off the stick, making it worthless. I flamed Corsair many years ago (on here) because of their warranty. I had a set of ddr3-1600 that i OCed to 2000+ until they died. I requested a RMA but was TOLD that i voided the warranty by running it outside of it’s specs in speed and voltage. IE.. No one will accept a RMA if the item is physically damaged or has burnt components.

As Woomack and others will tell you.. OUTSIDE of benching for pure speed/points on HWBot.org.. OCing memory to the max will give nothing (~1%) increase in games. You will see a BIGGER increase by upgrading the video card or cpu.

 

[magellan]

@User5566
What's the use-case for the two 3090's in your box? HPC? Cryptomining? I'll bet you can use your rig as a space heater during the winter.