also…
why not 4x16?
Steve said in his video he does not have single rank 16GB sticks and that primarily it was more about number of sticks than overall size as the games that benefit fit in around 7800MB worth of the 16GB minimum they use so it was not a size restraint but just adding more sticks gave more performance.
If the best is 2x single rank 16GB sticks and 4x 8GB sticks have shown to be better than 2x 8GB sticks. The very best may well end up being 4x single rank 16GB sticks. But Wendell has only demonstrated a 2 slot itx board and Steve does not have the sticks to try it… Yet.
Is there now faster ECC available for Ryzen? I haven’t found any yet…
Afaik the fastest is still 3200.
1 Like
Kind of surprised this is not a bigger discussion…
Let me point out that we need a bigger data set 
More benching needs to be done.
The more I look into this the more confused I am.
It seems like the answer is to get either a 4x8gb DDR4-3200 CL14 kit or a 2x16gb DDR4-3600 CL16 kit between what GamersNexus, Wendal, and Hardware Unboxed are saying. I think the weird part is figuring out of the 2x16gb sticks are single or dual rank but it’s unclear how to know if you have dual or single rank.
Some where else I read that 2x16gb is always dual rank and then someone else said it depends on how new the 2x16gb kits are. There also isn’t a discussion around quad rank which I’m guessing isn’t really meant for normal desktop machines but does this play a factor also for Zen 3?
Generally a 4x8 3600 CL 16 with decent timings is cheap enough and performs well enough that it will not matter much. Tweak tuning for small gains and higher cost is were this lies.
The 3200 as much about compatibility even tho it has solid performance but far more expensive.
Same thing with 3800 and up ram…you rapidly start losing the performance for an ever increasing cost.
1 Like
I wonder how well 4x16GB would do
2 Likes
As @SapphironZA mentioned I’m also pretty sure that Steve mispoke and actually meant dual rank. Buildzoid’s video @Hako linked also points in that direction.
This again points to dual rank being best:
2x8GB SR < 4x8GB SR < 2x16GB DR < 4x16GB DR
The sweet spot is 2x16GB because of dual rank and dual channel.
I was kind of waiting on Wendell’s video on it. I have not personally ran 16g sticks. Little curious about it.
should be out soon, sent Steve all the stuff
should be a good one. The answer, like all answers, is IT DEPENDS 
6 Likes
Isn’t “Go with max supported sticks for platform” usually best strategy? I’m planning 2x32GB, and they will probably be DR because current chip manufacturing limitations?
No… mostly because larger capacities stress the memory controller more thus reducing ability to run at much higher frequencies . If I have that correct ? Some of the limitations are very much platform dependent . What cpu /motherboard / os / use case
I agree on “platform dependent” part, its pretty obvious and always been the case, since 8086.
But I heard that “stress controller” stuff few times already. What’s that even mean, and how it would work?
From what I understand 2x32GB DR is electrically equal to 4x16GB SR, except memory modules are further away from each other, because traces.
So how this “stressing” would lower clocks more on two modules instead of four? And I still need at least 64GB at the end of the day…
I guess stress is not a hard enough term but damage is.
Since your memory is ultimate middleman any errors or problems tend to lead non functional. data loss
One of these computer scientists floating around here could explain it better I can.
I am literally just the average joe 
Thanks for info but, to summarize: stuff happens in silicon, smaller process, more problems. Nevertheless, it will be happening, no matter what. And, those articles from wiki state as much, nothing more.
So far it’s literally like saying “Don’t put bigger GPU card into your rig, or you will wear PSU more quickly”.
However, I would assume that if company is making DRAM sticks with controller chip on it, then stands to reason that they would plan for required load for it. And they are probably very well aware of those problems you are speaking of. No matter how many chips they put in them.
Also I found some papers that are more relevant to the topic:
https://ijpsat.ijsht-journals.org/index.php/ijpsat/article/download/1323/682
But I failed to find study that shows direct impact of stick size on controller lifespan.
I suppose there might be something to it, if controller would be separate from memory chips, and we would add them manually. But last time I did that was in 286 era AFAIR. I’m not sure if FP modules later had ICU or not.
Sorry, but for now I have to put this between urban legends and handwaving, because I still don’t see any concrete proof of reason why you would always prioritize smaller sticks if you have the option.
1 Like
Let me get this straight. Running ram and the memory controllers outside its design is hand waving and snake oil in regards to their lifetime and stability and has no negative repercussions what so ever. How many quad channel dims or even servers do you see running at 4000 plus ?
Second link is certainly more readable. We are talking about overclocking ram beyond is design intent not just population and capacity.
I believe you have a point. Is it one of those misconceptions that has risen because it works/and or easy. I will have to dig more. Is it one of those things that everyone has been doing for so long that it became excepted fact/method.
Seems like it was reply to my post? I never said or even suggested that. Of course it will have impact.
What I’m trying to say is, that engineers making specific ram stick will make sure its memory controller is within advertised speeds (at least hope so).
I never claimed anything about overclocking beyond spec. Silicon degrading bit more while doing that is pretty obvious.
But I never came across name for it, so thanks for pointing out technical term for that.
Yeah that more akin to what I suggested.
Still my original point was just this:
“When in doubt, use max supported stick size for platform. Usually you will be better off that way”
So you could come up with edge cases, when having smaller sticks will be beneficial. However I believe you will be very hard pressed to prove its common fact. Hence my jabs at “stressing memory controller with more chips” statement, since I couldn’t even find any reasonable definition for it.
And again, to be clear, I don’t claim anything when it comes to overclocking. Still I would like to see tested what others already suggested:
4 SR sticks vs 2 DR sticks (if equal, my point stands)
2 DR sticks vs 4 DR sticks (irrelevant for my point, but I’m curious)
1 Like