2x16GB sticks or 4x8GB sticks on Ryzen Gen3

newls1

Supreme [H]ardness
Joined
Sep 8, 2003
Messages
4,607
Trying to figure out if its preferred to get 2x16GB 3600mhz kit or 4x8GB 3600MHz kit for my upcoming 3950x build? Not gonna OC mem much, just want to apply XMP profile and go with it, hence why im interested in the CL16 NEO kits. Which mem sticks would make the Zen2 mem controller happiest?
 
Hello, not to hijack this thread but rather not create a new post:

I was watching some youtube videos and its still not clear to me but more a surprise that the ram i get is also important for the CPU to perform at its best.
I am still unsure which Motherboard to get BUT for now, my question is around the DDR4 ram i must get.
Can somebody tell me if this is very good ram, i dont care if it looks sexy or has more dropped suspension , i just need to know if it will make the Ryzen 3700x and 3900x perforn at their best....

I was about this ram, and want to know if it is the right ram?

G.Skill Trident Z Neo 32GB 16GBx2 DDR4 3600 RGB RAM F4-3600C18D-32GTZN

I dont know, but if i coukd even just use 1 stick of 16gb i would be happy enough.
I would prefer not to get 1 millin sticks of 4 or 8gb ram it is just strange to do this.
If i had the money i would get 64gb, but i am happy to get either 16gb or 32gb

Please let me know, thanks
 
Buy pairs, you will thank me later. What the hell do you need 64gig for? E-peen?
 
I have 2x16GB dual rank and it tightens to 3600Mhz Cas 16. I wouldn't pay extra for single rank unless you needed some other feature (RGB, etc.).
 
arent all 16GB sticks dual rank? is there NO draw backs going to 2 x16gb sticks vs 4 8gb sticks then?
 
If we're talking latest gen Ryzen, most motherboards support higher xmp frequencies in dual channel mode with two slots, slower when all four slots are filled. You can look in the manuals to see for your self (most mobo manufacturers have the manuals online so you can look at them before you buy the mobo).

So if at all possible, I'd go for 2x over 4x pieces of ram. All of my current rigs are dual channel (3900x, 2700x, 1700 and 2400G).
 
Last edited:
there is no such thing as "quad ch" on x570, and if you mean running 4 dimms (IE: populate all dimm slots on a 4 dimm slot board) still doesnt give you quad channel mem. What my questions was referring too is this: 16gb dimms are atleast to my knowledge ALL dual rank dimms, so running 2x16gb dual rank dimms is it harder, or "less" likely to achieve a slightly better OC vs running 4x8gb single rank dimms.. Does the Zen2 mem controller prefer one over the other?
 
Depending on the MB topology..


with a 32GB target -

Dual slot - best for signaling. forced 2x16
T-topology MB prefers 4x8
Daisy Chain MB prefers 2x16, one per bank
 
Depending on the MB topology..


with a 32GB target -

Dual slot - best for signaling. forced 2x16
T-topology MB prefers 4x8
Daisy Chain MB prefers 2x16, one per bank



I found that link (very nice btw) which shows some of the cheaper ASRock boards are T-Topology. Reddit seemed to indicate that the Taichi was sampled in T-topology, but the retail board is daisy chain. Asus seems like it is daisy chain throughout the X570 lineup.
 
so like I said before. 2x16

I mean, it depends on the board, but you'd figure if you're looking at an expensive CPU you aren't going to throw it in a cheap MB. High end boards seem to all be daisy chain with the exception of the mDTX Impact which is direct (same conclusion though). T-topology was much more prevalent with X470, etc.
 
I run 4x8GB 3600 at 3800 on mine. I do need to get some faster, timing wise, RAM but I will just pick that up later if I decide to bump to a 3950X or not. What I have is more than fine for Facebook.
 
I'd say 2x16, otherwise you might end up in a weird situation like me where I have 48GB total from having 2x16 and 2x8.
I can only run mine at 2800 (the sticks are all rated for 3200) but, frankly, with a gen 1 ryzen and a gen 1 ryzen board I've never been able to run any combination of sticks at 3200 stable. Going from 32GB to 48GB was only a minor downgrade in stable speed (from 2933 to 2800) and was a huge improvement in loading speed.

I use the extra 16 gigs as a RAMcache. Half for my OS drive and half for my gaming drive (which is further accelerated by an NVME tier 2 cache).
 
Last edited:
single rank 2x16

While possible (32GB DDR4 dual rank unbuffered now available) these do not exist in the market, and likely never will.

Every good 16GB dimm for Ryzen is dual rank, and b-die. (but I repeat myself)
 
On an Gigabyte X470 Ultra Gaming (the cheap 120 buck board) under BIOS 42g (IIRC) I have 4 Sticks of 3866 running at 3600 (32 GB: 4 X 8 GB) every time I try to push it up to 3733 it automatically kicks the ram up to 3866 after post and the CPU fabric takes a frequency hit. Honestly, on order to get the BIOS to actually run the RAM at 3600 I had to set the BIOS to 3533... There is something funky about the BIOS actually locking values correctly at the speeds selected if your RAM is capable of higher speeds (even without selecting the XMP). Not sure if anyone has seen this as well. Probably because I'm running a 3600 on a bullshit motherboard ;)
 
While possible (32GB DDR4 dual rank unbuffered now available) these do not exist in the market, and likely never will.

Every good 16GB dimm for Ryzen is dual rank, and b-die. (but I repeat myself)
thats exactly what I thought above... some people think 1 stick in each channel means "single rank" and that isnt the case
 
My thought has always been the bigger the DIMM size the better. You can always go up but you can't go down so I would go with 2x16. I know 32 GB seems like alot, and maybe in 2-3 years DDR5 will be the standard but from my experience, it usually takes 2-3 years after a release before it becomes standard and so hopefully you'll be using you're memory kit for another 4-5 years and by then 32 GB maybe the standard but you'll want 64 GB for high performance computing which is what I assume that's your market segment based on the CPU and amount of memory.
 
While possible (32GB DDR4 dual rank unbuffered now available) these do not exist in the market, and likely never will.

Every good 16GB dimm for Ryzen is dual rank, and b-die. (but I repeat myself)

Good is relative. I mean Hynix CJR can run at 3600Mhz and Cas 16...At half the price of B-die. That's $120+ extra for other components in one's budget. I've had multiple sticks of B-die, but it's not the be all end all of RAM.
 
Good is relative. I mean Hynix CJR can run at 3600Mhz and Cas 16...At half the price of B-die. That's $120+ extra for other components in one's budget. I've had multiple sticks of B-die, but it's not the be all end all of RAM.
do you happen to know of any 2x16gb sticks of 3600 cas 16 CJR kits I could buy?
 
do you happen to know of any 2x16gb sticks of 3600 cas 16 CJR kits I could buy?

I have the Ripjaws V set of 3600 Cas 19 which is CJR. I use the timings from the Ryzen memory calculator and it puts it at cas 16 at stock voltage. You might get tighter subtimings with b die, but I don't know that you would notice anywhere other than your wallet.
 
do you know if the gskill neo's 2x16gb cas 16 sticks are cjr?
 
do you know if the gskill neo's 2x16gb cas 16 sticks are cjr?

I don't off the top of my head. I'm sure someone here has run them through taiphoon burner though. Or a review online should say what they are.
 
Buy pairs, you will thank me later. What the hell do you need 64gig for? E-peen?

I use 64GB to use half of it for cache on my gaming drive, and another 8 of it to load the paghefile into. I love the load times when revisiting areas again and I have noticed it can really smooth out games that otherwise have microstutter. Makes for a much better gaming experience especially to someone like me who is extremely sensitive to stutter and low framerate
 
I use 64GB to use half of it for cache on my gaming drive, and another 8 of it to load the paghefile into. I love the load times when revisiting areas again and I have noticed it can really smooth out games that otherwise have microstutter. Makes for a much better gaming experience especially to someone like me who is extremely sensitive to stutter and low framerate

ok. Once my games load, they are done (FPS). Also, pagefile is literally pointless in my case. idk. Whatever works for u. Guess it's ur game specific scenario.
 
Going 4x8GB would depend on your cooling and how tall the RAM is. Usually, tall RAM does not work when having 4 modules due to the heat-sink, unless you're doing water cooling.

Personally, I went for 4x modules because I wanted to populate all 4 DIMMs since I'm using a Loop and RGB on memory modules is not an eyesore for me, but it's dimmed to the lowest possible.

But it's up to you and the cooling solution you use, if 4 modules is possible? Go for that one, if because of a heat-sink? Then 2 modules.

Not sure what memory you went with but if you had Vengence Pro RGB, they have dummy modules with RGB to use as fillers,
https://www.amazon.com/CORSAIR-VENG...nics&sprefix=corsair+,electronics,152&sr=1-21
 
As an Amazon Associate, HardForum may earn from qualifying purchases.
I am using the Trident Z 2 x 16gb sticks. I will use the other 2 for something else. They seem to work fine.

f4-3200c16q-64gtzkw
 
There are no single rank 2 x 16gb kits that exist. They're all dual rank

Yes and no. I havent seen any unbuffered non-ECC DDR4 in single rank 16GB sticks, but there are ECC registered chips that are single rank 16GB from both Samsung and Hynix. Actually, Samsung even has a 32GB single rank registered dimm at DDR-2666 speeds.
Samsung also has unbuffered non-ECC LPDDR4 chips in 16gb capacity which is what you need for single rank 16GB sticks. Only LPDDR4 is meant for mobile so do not get used in 8 chip configurations.
So there are chips and kits that exist just fine, it just isnt in desktop user format yet. Only a matter of time though
 
Last edited:
I'm not really aware of how single vs dual rank works. But from my understanding, dual rank memory has to make 2 pases to acces data while single rank only one, increasing lantency (kind of like single vs dual vs quad channe mem I think).

So, based on that, I would go for 4x8gb if possible.
 
I'm not really aware of how single vs dual rank works. But from my understanding, dual rank memory has to make 2 pases to acces data while single rank only one, increasing lantency (kind of like single vs dual vs quad channe mem I think).

So, based on that, I would go for 4x8gb if possible.
4x8 single rank sticks is the same as 2x16 dual rank sticks, only with added latency skewing between the slots which makes things harder. (edit: unless you have T-Topology trace layouts and then latency skew should be the same for both dimm slots on a channel)

dual rank is usually more strain on a memory controller, but also is faster at the same MHz speed as single rank sticks. This is because interleaving can better take advantage of dual rank modules. Quad rank is usually too much and cannot take even more advantage of interleaving because you can still always only access a single rank at a time, and by the time you get around to the 4th one, the first one or two sets of chips are already available for access anyway so you dont really gain anything while just making the memory controller work harder and usually need slower speed.
 
dual rank is usually more strain on a memory controller, but also is faster at the same MHz speed as single rank sticks.

It is not "more of a strain." If anything of it to be is less of a strain. This is due to the fact that dual rank can have a page within one of the ranks(banks) and therefore more page hits. Dual rank always performs better, and is what you actually want. But if you were to run particular tests it may seem "less." Interleaved dual ranks will always provide the best performance.

Strain comes with voltage, installed banks (more installed more interference), and out of spec configurations. Going out of spec is entirely a different argument.



Quad rank is usually too much and cannot take even more advantage of interleaving because you can still always only access a single rank at a time


Incorrect, the whitepapers specify for Intel's IMC, that the memory controller in dual-channel symmetrical mode (interleaved) can do such with pages. AMD's IMC is similar:



nj4NQHj


This is why multi channel does better. Interleaved always performs better and having all banks installed can be better. They are designed to perform as so, and will operate as long as all configuration are held in occordance to specifications. Note, this does not include other factors of the IMC's abilities, e.g. Intel's Fast Memory Access, or what AMD IMC's abilities are. It is when you use a higher than rated speed of the controller spec, and voltage that "it is an issue." If you have a little Google Fu, you will find a-many-a-rabbit-holes about this: https://www.tomshardware.com/reviews/amd-ryzen-3000-best-memory-timings,6310-2.html. The trick is not to think of it as "bandwidth" but as pages retrieved from memory.

Note: I know we are talking about Ryzens, but used Intel's as an example since the documents were quicker to find.
 
Last edited by a moderator:
so question... why are my 4 sticks of 16gb dual rank modules operate @ 1T command rate, but when i run 4 sticks of 8gb single rank modules they have to operate @ 2T?
 
It is not "more of a strain." If anything of it to be is less of a strain. This is due to the fact that dual rank can have a page within one of the ranks(banks) and therefore more page hits. Dual rank always performs better, and is what you actually want. But if you were to run particular tests it may seem "less." Interleaved dual ranks will always provide the best performance.

Strain comes with voltage, installed banks (more installed more interference), and out of spec configurations. Going out of spec is entirely a different argument.

So, your saying I am incorrect, but then you say that you agree with what I said? Not trying to be a dick here but its confusing when you say one thing and then repeat the same information I did as correct.





Incorrect, the whitepapers specify for Intel's IMC, that the memory controller in dual-channel symmetrical mode (interleaved) can do such with pages. AMD's IMC is similar:



This is why multi channel does better. Interleaved always performs better and having all banks installed can be better. They are designed to perform as so, and will operate as long as all configuration are held in occordance to specifications. Note, this does not include other factors of the IMC's abilities, e.g. Intel's Fast Memory Access, or what AMD IMC's abilities are. It is when you use a higher than rated speed of the controller spec, and voltage that "it is an issue." If you have a little Google Fu, you will find a-many-a-rabbit-holes about this: https://www.tomshardware.com/reviews/amd-ryzen-3000-best-memory-timings,6310-2.html. The trick is not to think of it as "bandwidth" but as pages retrieved from memory.

Note: I know we are talking about Ryzens, but used Intel's as an example since the documents were quicker to find.


You are talking about something different here. I was talking about how installing quad rank or higher on 90% of motherboards out there the specs say you must lower speed or loosen timings, thus being lower performance. nothing about the technical aspects of internal memory controller and chip configurations. As you said, if you OC it out of spec and it runs then that is a different matter, but in the majority of boards out there the spec says speed must be lowered when installing quad rank and higher worth of sticks.
 
Back
Top