I accidentally bought the wrong RAM...

[-]

DZCreeper@reddit

The brand names don't mean anything. The actual memory chips used on each stick is what matters.

Mismatched memory chips can stop XMP/DOCP from working, but when running at the slower JEDEC aka default speeds you can mix memory with no problems.

[-]

acewing905@reddit

I've had cases where ram from different brands that on paper should have the same specs outright refuse to work together

I don't understand the theory behind this but mixing and matching ram is always a gamble. Sometimes by the time you notice issues you might even be past return date so this is something people have to be careful about

[-]

DZCreeper@reddit

This is often because companies will change the memory chips used depending on production run prices. They will not change the model number when this occurs.

Generic bins such as 3200 CL16 and 3600 CL18 are most vulnerable to this tactic because so many different DDR4 chips are capable of such settings.

Motherboard topology can also be an issue, DDR4 is a mix of daisy chain and T-Topology, DDR5 is universally daisy chain to my knowledge. You can tell which topology is used based on the QVL, T-Topology will run fastest with 2 sticks per channel, daisy chain with 1 stick per channel.

[-]

acewing905@reddit

Unfortunately all I have are my own anecdotes, even though they may not make much sense
But basically, even before XMP became the standard, back in the DDR/DDR2 eras, I have come across RAM sticks that just did not do well together. Random BSODs and reboots, sometimes refusing to boot altogether
But the same sticks would be rock solid one at a time, and the same motherboard would be chugging away fine with a proper kit of two sticks
To this day I have no idea how this actually happens, but as a result I try to avoid mixing and matching RAM as much as I can, and stick to kits

[-]

RedBoxSquare@reddit

Sometimes I had problems with JEDEC speeds too. Some AM5 boards are picky about the RAM used, and they would take a really long time to post. Memory is like magic to me.

[-]

DZCreeper@reddit

Long initial times are normal on DDR5, the memory training takes longer than DDR4.

Also consumer DDR5 boards are physically optimized for only 1 stick per channel. You could run mismatched sticks in slots 2+4, but filling slots 1+3 even with matching sticks is not ideal.

If you really care I can write a followup comment explaining this stuff in detail.

[-]

ebullientmarshmallow@reddit

Please do. Is this just on AM5 boards? I have an MSI B760 Tomahawk wifi DDR5 - i5 13600K - 4x16 Kingston Fury Beast setup. It seems fine to me, but I came from a 12 yr old i5 computer. Any info you could provide on this would be very much appreciated.

[-]

DZCreeper@reddit

No, it is true for all motherboards.

There are 3 ways to hook up memory slots.

Direct aka 1DPC connection from each memory channel to 1 slot.
Daisy chain aka serial. Nearest slot is hooked up first, second slot is an extension.
T-topology aka parallel. Two slots are wired with equal length.

Electric fields propagate along the total conductor, areas of low resistance have the highest field strength but high resistance areas still add signal reflections equal to the path length. They also act as antenna.

This means direct is best for outright performance, as there is minimal signal reflections and noise created by unused traces.

Daisy chain is a compromise allowing for higher total memory capacity. You utilize the outside slots first because this minimizes the length of the non-terminated traces.

T-topology is optimized for total capacity, it assumes both slots are filled. If you only fill 1 slot it is just a worse version of daisy chain.

At low speeds the differences are less apparent, because there is more margin for reflection times. Higher voltages also help because the signal to noise ratio increases. This is why DDR4 and prior often had T-topology boards available, manufacturers could afford to optimize 2DPC setups and the average 1DPC consumer never ran their memory fast enough to notice the downside.

Higher memory speeds are also a driving force behind PBC layer counts. More layers means better isolation of memory traces.

[-]

ExaltedStudios@reddit

Taking a look at your motherboard's stats/technical specs page is the best way to figure this out for your own situation. It will usually list the maximum clock speeds you can get for all possible configurations of RAM.

I'm not going to pretend like I'm familiar enough to explain DDR4/DDR5 and AM4/AM5, just sharing what I do when shopping for ram when I already have a specific motherboard.

But,, long story short, from what I've found, 2 sticks is almost always better than one for consumer boards. When you start moving into server builds with EPYC and Threadripper, things start to change.

[-]

sidewaizsocks@reddit

I care! Drop a knowledge bomb on me!

[-]

halodude423@reddit

He's got DDR4, should be okay in most cases. DDR5 and specially AM5 is more picky.

[-]

aightmanboi@reddit

As long as it’s the same ram speed, it doesn’t matter.

[-]

RainbowDroidMan@reddit

Probably but since it’s brand new you might as well return it now while you can instead of having to find out.

[-]

-ThatGingerKid-@reddit (OP)

Thanks for the feedback! I just decided to return it and order the correct one.

[-]

Beauenheim@reddit

Even ordering the exact same kit (Matching latency, timings, all that) doesn't necessarily guarantee stability. Best practice is to buy a kit of 4 sticks, or better yet, a kit of 2 at higher capacity. For best chance of running at the best speed, they need to be paired together at the factory. K

its you buy online are matched pairs specifically, they aren't meant to be upgrade kits, rather, replacement kits. You've probably been told otherwise, and for a long time, especially in the ddr2 and ddr3 days, I feel like memory controllers used to be a lot more lenient to mixing.

All that being said, it could be fine, but I have personally had both good and bad experiences with mixing ram kits (even if again they are the same product).

[-]

WolvenSpectre2@reddit

Will it work? Yes.

Will it have the same timings? Maybe yes, Maybe no.

Will it work at the slowest timing? It will try, but mostly yes.

Will XMP/DOCA/EXPO work? HELL NO! If it does get down on your knees and thank Jebus. ESPECILALLY if you are on AM4/AM5.

Will you have to manually dial in your overclock to get the speeds that you paid for? No because you will not get 3200MT/s.

Will you have to manually dial in your overclocks to get the max stable speed you can? Yes.

What will help you tell if the speed you have dialed in are stable?

MemTestx86+
OCCT
Heavy Load.

[-]