Is mixing ram really that bad?
Posted by JuniorPro69@reddit | buildapc | View on Reddit | 64 comments
Hi all! I currently have 2x8 ram modules (Corsair 3200) and am thinking about upgrading, but since the prices are still high I'm considering adding 2 more sticks.
I have Ryzen 5600G on AM4 socket. May it really cause freezes during gaming?
I've found info that it is not recommend, and at the same time I've found info that the only harm it can cause it's jus they'll run at the slowest speed.
And is the speed really that necessary for gaming? What is the difference between 2400 and 3200?
jimmyjammy6262@reddit
Difference is measured in nanoseconds, don't let the little knowledge is a dangerous thing brigade tell you otherwise!
lazyhustlermusic@reddit
A difference of 10 nanoseconds can literally be twice the latency
Deadward84@reddit
There are 1,000,000 nanoseconds in 1 millisecond, and 1000 milliseconds in 1 second. If a 10ns difference doubles latency, you could multiply that 10ns by tens of thousands and still see no perceivable impact.
Modern technology is more akin to magic than anything else. There are chips today smaller than a fingernail that are orders of magnitude smaller and faster than what we had 15 years ago, to say nothing of 100 years ago. We are able to measure it and express it, but it largely defies human comprehension.
lazyhustlermusic@reddit
Not really, if your memory operations take twice as long then you can only do half of them in the same timeframe.
That’s like putting your hand on the heatsink and complaining 75C is too hot. Your perception/assumption of time scale has nothing do with actual operating parameters.
Deadward84@reddit
Speaking of things that have nothing to do with the topic at hand, I'm not sure why you would compare time to heat. Ones tolerance to heat is subjective. Time is not. The point is that at the nanosecond scale, if your memory operation takes 10ns to complete or it takes 90,000ns to complete, it's still sub 1ms. Imperceptible to human senses.
feed_me_haribo@reddit
This is just so incredibly wrong. It's not about a single read taking less than a millisecond, it's about making 3 billion of them in 1 second.
Deadward84@reddit
Jesus Christ. Ok, look. A single read/write isn't going to take 90,000ns. It was an exaggeration. Hyperbole. Just like insisting that a 10ns difference DOUBLING latency is going to be noticeable to the average person. The real world difference between 3200 and 2400 is \~8ns. Eight billionths of a second. From 3200 to 2133 is \~12.2ns. Twelve billionths of a second. You. Will. Not. Notice.
lazyhustlermusic@reddit
Except that there’s plenty of observational data that says otherwise. Why is your only train of thought ‘I can’t react that fast, so a computer can’t either’?
Deadward84@reddit
Why is your only train of thought to assume I'm basing this on personal anecdotes? What observational data do you have? Because I'm ready to link a bunch of science journals and stuff.
lazyhustlermusic@reddit
lol what is this irrelevant garbage supposed to be exactly? You’ve completely gone off the rails.
Deadward84@reddit
Not at all. From the start I've argued that a person cannot notice a perceivable difference in nanosecond scale latency. Have you been arguing this entire time without realizing that?
lazyhustlermusic@reddit
I’ve highlighted your brain dead take in a variety of replies. You’re too hung up on ‘human can’t nanosecond’ whereas there’s a measurable performance difference in computational environments that you can certainly see and feel.
When presented with objective numerical data, you simply deflect back to ‘human can’t nanosecond’. I even made a heat related comparison for you that you also couldn’t grasp.
Best of luck.
Deadward84@reddit
"We are able to measure it and express it, but it largely defies human comprehension."
Literally in my first comment. At no point did I argue there wasn't a measurable performance difference. You've consistently failed to grasp the very basic truth that in the context of this post, where OP asks the difference between 2400 and 3200, the nanosecond differences in this instance, while measurable, would be imperceivable to the average person.
Measurable =/= Perceivable
You claim I deflect when presented with objective numerical data, but it's clear that from the beginning you have misunderstood both my words and my intent. You seem to think that your entirely reductive 'human can't nanosecond' take on my argument somehow contradicts what you said. It doesn't. And again, I never claimed it did. These things need not be mutually exclusive, and despite engaging with you in good faith, your responses devolved into ad hominem attacks. You couldn't even manage an "agree to disagree"
Have the day you deserve.
coolboy856@reddit
Absolutely correct, this however is not something you've been saying :-)
lazyhustlermusic@reddit
Why would it need to be perceptible to human senses? I simply gave you another example where human perception is irrelevant and you doubled down on it lol.
It’s weird that you don’t understand basic double/half mathematics though, but that’s why you aren’t an engineer.
Deadward84@reddit
Interesting that you're starting to make assumptions about me and not the argument. I understand what you're saying just fine, but you're not understanding that in discussing it with regards to the OP, they're not going to be able to tell the difference between 10ns and 100,000ns. If you count how long something takes, what number do you start with? Do you start with 1ns? 1ms? Of course not. You start with 1. Your argument and your examples are irrelevant in regards to this topic because OP and the vast majority of people who post here and in similar subs don't understand the technical aspects. If OP has 2x8GB of RAM and performing some operation on the computer takes 1 second, and then they add another 2x8GB and that same operation still takes 1 second, to them it's effectively the same, even if their latency is 10x what it was before.
lazyhustlermusic@reddit
‘Your assumptions were correct, but I’m going to be upset anyway’.
The computer functions more poorly with double the memory latency, and its measurable. Your opinion of how long a nanosecond feels, is irrelevant to anything.
Your analogy is funny and the description of someone who is unfamiliar with any sort of measurable performance metric.
Deadward84@reddit
I never said anything to the contrary, and I have nothing to be upset about. There's no such thing as an opinion on how long a nanosecond feels. Humans cannot physically perceive a nanosecond.
lazyhustlermusic@reddit
That continues to be irrelevant.
Consider the time difference between CL30 and CL36 ddr5 6000 is literally 2ns
Deadward84@reddit
And what am I supposed to be considering with this time difference?
lazyhustlermusic@reddit
I mean even that is up to a 5% uplift in some gaming titles.
Ever wonder why x3d chips have such a higher performance ceiling in latency sensitive workloads? Largely due to avoiding the dram round trip penalty of about 60ns.
Hooray you learned something today.
Deadward84@reddit
This is what one would call moving the goalposts. Don't pat yourself on the back too hard, you didn't teach me anything new. I just asked to find out why you thought it was relevant.
lazyhustlermusic@reddit
Your tism is flaring up again.
Ddr4 is also measured in nanoseconds, much like all sdram, since you didn’t know that either.
A full learning experience!
Deadward84@reddit
Ah, I see we've gone all in on ad hominem. Of course DDR4 still counts in ns, but touting the IPC uplift of Zen 4 or 5 X3D parts in regards to gaming performance had nothing to do with the initial argument. Though if you want, I'll humor you. If OP doubles the amount of RAM he has from 2x8GB to 4x8GB, and considering they are running a 5600G and a 3060 Ti, what is the expected impact if they have to drop the RAM speed from 3200 to 2666? Keeping in mind you're limited to PCIe 3.0 on both the CPU and GPU.
Not that it matters at all, but what was your desired reaction in calling me autistic? Am I supposed to be angry? Offended? That if I was autistic I'd cry and feel bad and not respond? If I did, would that make you feel good?
lazyhustlermusic@reddit
lol you continue to highlight your true lack of comprehension from silly misunderstood examples.
laffer1@reddit
If a cpu is overheating, it throttles. It will take more time for the same tasks. You get less boost clocks, etc.
It can relate to time. 75c is in spec for many modern CPUs but a problem for some old ones.
Deadward84@reddit
You're getting off track. He said putting a hand on a 75C CPU and complaining it's hot. What's hot to me isn't hot for you. I said perceivable difference in regards to time. Heat of anything isn't relevant to perception of time in this.
lazyhustlermusic@reddit
167F will burn anyone. Your opinion of whether it’s too hot is irrelevant to the PC which operates fine in the same parameters. That’s the entire point.
You seem determined that ‘hero derp human can’t react or feel nanoseconds’ but computer performance is 100% absolutely tied to those values.
laffer1@reddit
It’s still in the same realm. How one measures the heat is irrelevant. You asked how time and heat relate. I explained.
coolboy856@reddit
Brother what? Half is half regardless if you scale it to a trillion vs. 2 trillion?
In a situation where a single instruction takes 10ns, you are claiming that 500,000ns per instruction will result in "no perceivable impact".
0.002% the IPC is definitely going to be perceivable to anything capable of perceiving things
Deadward84@reddit
Can you perceive a billionth of a second?
coolboy856@reddit
Yeah that's not how anything works, buddy
Xiij@reddit
If all memory reads take twice as long, it will scale all the way up to the user level, and you will notice your tasks taking twice as long.
For example, if you are a video editor, a render that used to take 1 hour, will take 2 hours instead.
Twice means twice, regardless of scale.
netsx@reddit
Nanoseconds to a human is nothing, but to a machine that is a LOT.
feed_me_haribo@reddit
Unfortunately FCLK and MCLK run in hundreds of picoseconds.
justice7@reddit
It will just run at the slower ram speed. As long as it works it should be fine.
USSHammond@reddit
Running at a slower jedec speed is absolutely NOT guaranteed, it may work it may not even boot at all.
Intrepid_Bobcat_2931@reddit
You're right. Best case it runs at the slower speed. But can very well not work.
RecalcitrantBeagle@reddit
Best case it all runs at 3200mhz, actually - I've manually clocked cheap green 2667 DDR4 pulled out of prebuilts up to 3200mhz with a 5600G and had it run rock solid. That's not at all guaranteed, but I think people overestimate how finnicky it is sometimes, though for good reason - recently, DDR5 is very unforgiving, and previously the early Ryzen 1000/2000 could have trouble just running a kit at its rated speed. The Ryzen 5000 series, though, landed in a spot of having a relatively mature memory controller before changing standards.
RecalcitrantBeagle@reddit
So, it's never 100% for sure, but in your case, it'll probably be fine - the Ryzen 5000 series had pretty robust memory controllers for DDR4. Generally, you'll be completely fine running it at the lower speed (if you have 2x8GB 2667 and 2x8GB 3200, for instance, you'll likely be able to run it just fine at 4x8GB 2667.) Having said that, nothing is guaranteed, but I've had very good luck with manually setting it to the higher speed, or an in-between (say, 4x8GB at 3000mhz) for a Ryzen 5600G specifically.
Note that this does not carry over for AM5, there I've had issues with just getting a 6400mhz kit to run at its rated speeds with a 9900X. DDR
USSHammond@reddit
Mixing kits in any shape or form is a bad idea and can lead to full system boot failure, especially on an AMD platform which has a weaker memory controller
Worldly_Device_@reddit
Yeah can confirm. Mixed some 3200mhz and 2600mhz ram once and I had random blue screens and sometimes didn't boot. Bought another stick of the 3200 one and the issues stopped. I thought maybe the 2600 stick was faulty, but been using it on another PC for years and it's been just fine
thekins33@reddit
Can confirm and shits the bed and won't boot sometimes but I'm too lazy to open the PC and remove the slower ram...
theRealtechnofuzz@reddit
It will just run at a slower speed. DDR4 is less finnicky than DDR5, so you should be ok. Alot of people think mismatched ram will cause endless issues, but they really blow it out of proportion. You have a 5600g, so you really want the fastest ram you can find because it has a much better imc than a 5600 or 5600x. Mismatched ram shouldn't be an issue, but if you can get 2x16gb ddr4-4000 that would be ideal.
Scrogdor@reddit
I ran different ram on my old x370 board 5800x3D. Did fine, two 16’s and 2 8’s.
Not all boards are CPUs will be stable though. May fail to boot. If it does you likely know what the problem is.
Jahoyhoy@reddit
YES. YES IT IS.
Elitefuture@reddit
There's a small chance it might not work at all, but you could always return it.
As for the speed, 2400 vs 3200 is a pretty big difference. Also you may need to do manual tunings to get the proper optimal speed. Note that it'll need to take the slowest of each subtiming as well.
And to top it off, the 5600G uses ram for its vram(assuming you're using the igpu and not a dedicated GPU). So the slower ram will be more noticeable.
Just gotta weigh the pros and cons. If you're regularly needing more than 16gb of ram, then these cons are worth it.
Also, can you not get a faster used kit of ram? Or can you buy a used 2x16gb then sell your current ram?
JuniorPro69@reddit (OP)
I use 3060 ti, I've bought 5600G just to save up on GPU.
You say 2400 vs 3200 is a pretty big difference, but what impact you really get in gaming? Will it lag or freeze?
Elitefuture@reddit
It won't lag or freeze, it'll just run slower.
Like a 10%-20% fps loss in games where ram wasn't an issue: https://www.youtube.com/watch?v=t5soCeuGNx0
https://youtu.be/I0zGeb9M8zM?si=6WBP--Bv2bEAUKl8&t=322
In games where 16gb wasn't enough, then you'll get an improvement.
Also note that the 5600g has a lot less cache than the 5600x, so it calls the ram more often, so the performance loss might actually be a bit bigger.
Uhmattbravo@reddit
Not really bad, but not ideal either. 16gb isn't that bad for that processor, and if you're running the integrated graphics instead of a discreet graphics card, you'll want that RAM to be as fast as possible.
JuniorPro69@reddit (OP)
I use discreet. I've bought 5600G on a GPU price boom
GABE_EDD@reddit
Ideally you want to pair the exact same speeds and timings, i.e. 3200MT/s 16-18-18-36 should be paired with another kit of 3200MT/s 16-18-18-36. If you don't pair it with another kit with the exact same speeds and timings you may end up having to run the kits at slower than their rated speeds and timings to avoid stability issues. Overall, it's a very slightly performance penalty running your RAM slightly slower, like single digit percentage gaming performance penalty typically.
Also, if you don't reach \~15GB or more RAM usage while gaming then you don't have a use case for more RAM. A lot of games won't use that much RAM, especially if it's the only thing really open (no google chrome, discord, etc.) Monitor your Memory usage in Task Manager to determine if you even need a RAM capacity upgrade in the first place.
JuniorPro69@reddit (OP)
That's the issue I play heavily modded games so the ram usage is high
laffer1@reddit
I’ve seen 3 games use 32gb+. It’s not common yet but can happen.
Electronic_K1rk3r@reddit
yes it is
just either buy another 2x8 of the same brand, model and frequency
or swap for 2x16
going with different model sticks can either work with the ram speed being the one of the slowest stick or not work at all, best to be safe and go for same model
netsx@reddit
It could work out (probably). But as others have said, not guaranteed to work.
If you have a motherboard that allows setting memory timings and DRAM voltage, you could literally overclock the 2400 kit to something higher, while downclocking the 3200. Both pairs have to run the same settings. I'd recommend active cooling, for the sticks (case fans blowing directly on them is usually enough). My system wouldn't even boot on 2 of sticks (Hynix DJR) at XMP 3600 "Mhz". But now i have 4 of the same of them running at even better timings than XMP. All four of the sticks are 2400 "Mhz" when not in XMP mode.
If you don't have a motherboard that allows those settings, it would be more of gamble if you could even make the combination work. So ensure you do have a motherboard that lets you tweak it.
MagicPistol@reddit
It may work but cause crashes. It may not boot up at all. It may work just fine but at lowest speeds.
You never know, and that's why it's not recommended.
B-R0ck@reddit
Just means you will be limited to the speed of the lowest stick you have.
mgp901@reddit
I have mixed kits or RAM. Just get the same speed, capacity, and CL rating to what you already have. You'll have to manually set the unmatched timings to the bigger number. Worst case scenario, just pick 1 speed lower in expo profile than what you have, ~3150 MHz.
ecktt@reddit
It used to be.
But SPD+JEDEC has made things a lot better.
When you get into RAM overclocking, EXPO and XMP, it IS bad.
Creative_atom0406@reddit
I had a similar situation with a Ryzen 5 2400G and I recommend that you don't mix them. I bought several sticks and all of them didn't work. It worked at the end but I had to match the speed, timings and even the rank (how many physical chips there are on the stick), coz the memory controller in the CPU couldn't handle ram with more chips.Just skip all the headaches and get a set.
MoistWelder8712@reddit
If your intention is to stay on this system for longer, then honestly sell your kit and get a 2x16gb kit instead. 4 sticks work, but it's always worse. Even if all 4 sticks have the excact same speed/timings it's harder on the memory controller, you're likely forced to run them at a lower speed than XMP and playing around with timings can be a pain in the ass.
Get a used 2x16gb Kit instead, sell your sticks and save you any possible troubles imo if you plan on staying on that system for longer.
thekins33@reddit
The difference is and will be a problem source I've done it before and have it on my current pc sometimes the PC won't start I have to power cycle it to get it to work I have 5600 ram and 6000 ram installed. It's running at like 5000 or something slower not sure anyways yes it's a problem. Just get all the same brand and speed ram.
lazyhustlermusic@reddit
At least match the speed, if the timing tables don’t match up it gets funky. A lot of people blow it out of proportion though.