Cool your DDR5

LazyGamer

FPS Junkie
Joined
Sep 5, 2020
Messages
2,404
Points
113
I've been on a bit of a bender - playing with Alder Lake, learning DDR5, finding out just how much there is to 'tweak' with these systems.

I've also stumbled across a problem - current DDR5 memory does not behave like the DDR4 we've been used to for the past five or six years.

You cannot set DDR5 to XMP and call it done.

There are about a half-dozen voltage rails that affect DDR5 stability, each of which vary in effect by motherboard (not just by brand!), and not all of which are set correctly when XMP is selected. And they also depend, of course, on the specific memory kit being used, and the quality of the CPU.

But that's not all!

You see, many of the higher-clocked DDR5 kits expect 1.35v or 1.40v. However, many of these kits also have inadequate heatsinks, meaning that running at high clockspeeds at 6000MTs+ (3000MHz+) will cause the DIMMs to heat up substantially. Many kits will idle around 35c in <25c ambient, and many of those same kits will scream to 70c under load.

Over 50c, random application crashes are likely to occur. May or may not be exhibited by stress test failures.
Over 55c, system crashes as well as stress tests failing more quickly.
Over 60c, expect more frequent application crashes as well as system crashes.
Over 65c, expect basically nothing to work for very long.



My current solution has been to lean a 40mm x 40mm Delta 10,000RPM screamer against the bottom of the sticks, sitting on top of the GPU. This is a stopgap solution, but it has resulted in keeping my kit below 50c.

I'm currently testing my DDR5 6400 C32 kit at its stock XMP.



If you're cooling your DDR5 and still having trouble, update your motherboard BIOS. Apparently motherboard manufacturers are still hard at work tuning their boards to work with the available DDR5 kits, and there have been quite a few instances of higher-end boards that have exhibited issues that warrant RMAs.
 
Not going against any of you findings, but I have found in my system using Alder Lake and DDR5 that it was completely stable after enabling XMP mode out of the box at 6000mhz. I attribute that to using Aorus memory with an Aorus motherboard mostly. I did update the bios initially, but did not have to flash the bios before booting the PC as a lot of other users stated needed to be done after much reading on the subject.
Also the Aorus memory I have has some beefy heatsinks on them and they run very cool. It looks like by the pictures that you are using GSkill memory, which I have read runs a tad on the warmer side especially the 6400mhz kits.
 
Not going against any of you findings, but I have found in my system using Alder Lake and DDR5 that it was completely stable after enabling XMP mode out of the box at 6000mhz. I attribute that to using Aorus memory with an Aorus motherboard mostly. I did update the bios initially, but did not have to flash the bios before booting the PC as a lot of other users stated needed to be done after much reading on the subject.
Also the Aorus memory I have has some beefy heatsinks on them and they run very cool. It looks like by the pictures that you are using GSkill memory, which I have read runs a tad on the warmer side especially the 6400mhz kits.
There's that - and I'm finding that with the RGB disabled, and with a Corsair dual-50mm RAM cooler - I'm seeing a bit more success with getting closer to 6400MTs.

I'll also say that with the same basic G.Skill kit, but spec'd at 5600MTs XMP, I didn't have any trouble running at XMP with the RGB on.

Clockspeeds and especially voltages play a big part of this. For the 6400MTs C32 kit I'm using, XMP calls for 1.4v - this means that the memory heats up really fast under stress testing.

And a note on testing - I'm using three different tests followed by some gaming. You'd figure that this wouldn't be necessary, but with how finicky DDR5 can be, I'm not ready to call anything stable that quick.



I also have to give props to Newegg, and almost certainly GamersNexus. I started on my DDR5 journey with Gigabyte's Z690 Aero D - a premium board that had every feature I really wanted, including decent memory support, 10Gbit networking, and especially Thunderbolt 4. A big part of that list is that I use systems for six to eight years in different roles.

Well, I ran across a problem that was very new to me - memory sticks dying. And not in a way that was catastrophic to the system. Since all the 'good' DDR5 kits are 2x16GB, for 32GB total, and 16GB itself is actually enough for most things users do, including gaming and photo editing, which I did while one of my sticks was not actually available.

I honestly didn't notice the first time it happened.

I tried all the logical troubleshooting actions - moving sticks around, booting off single sticks, running at stock, booting at higher voltages - and the good stick was still good, and the dead stick was still dead, and the system still behaved with the same quirks as with both sticks running.

This happened three times.

I've since returned the board and memory kits, the board specifically to Newegg beyond the return window who still took it back, all with notations that the parts were flawed and should not be resold as open box.

I'm now on an MSI Z690 Meg Ace - a board significantly more expensive than the Gigabyte board, and honestly with fewer features I care about (no 10Gbit networking) along with a black and gold scheme that while rather sharp on its own clashes with my white and black build - but as a board it's been rock solid.



Main thesis for the memory issues on the Gigabyte board has been Gigabyte's RGB Fusion software unnecessarily messing with the SPDs on G.Skills RGB kits. A bit goofy, but it reliably made modules unusable that weren't even able to run at their full XMP speeds stably. I haven't seen any sign of this hitting with my MSI board, and really, that's the only conclusion that I can make, and does amount to a guess.



I also don't want to sound like I'm calling out Gigabyte. I'm not. This is just my sample of one experience. I feel a bit personally burned, but I understand that these issues are part of the teething issues involved in a new generation of processers and especially a new generation of desktop memory.
 
Become a Patron!
Back
Top