Where does the 'Two Sticks' are better than 'Four Sticks' idea come from?

DarkSideA8

Gawd
Joined
Apr 13, 2005
Messages
989
I've heard this touted throughout my perusing of tech forums:
  • Two sticks are better than four.
  • You want two 8gb sticks, not four 4gb sticks to get to 16gb of RAM
etc. etc.

I've always assumed the people asserting this knew what they were talking about. Then I ran across these:
https://www.techspot.com/article/1971-more-ram-modules-better-for-gaming/
https://www.techspot.com/article/2140-ryzen-5000-memory-performance/

Both articles claim that filling up all 4 slots is better than running only 2 in the primary.

So if TechSpot is correct - where does the 'old saw' come from, and how long has 4 been better than 2?

Edit - sincere thanks to the respondents below. For a summary of what I discovered as the answer see post #5, below
 
Last edited:
2 dimms for extreme overclocking
4 dimms if you are using single rank dimms on Ryzen 5000 and are CPU bound in games
That is what I remember reading last year

here is a little clip about ram slots

Hey - thanks for that vid - it was great!

Also it led me to this, which explains T-Topology and daisy chaining



Still watching it - but I now need research my Mobo to see what Gigabyte did on my board

Edit - this video also (finally) explains why I hear single rank is better than dual rank for overclocking
 
Last edited:
Okay - so to answer my own question (for anyone skimming):

The old saw about 2 being better than 4 is largely related to how the architecture of the board feeds information from the RAM slots to and from the memory controllers on the cpu - mostly about the overall efficiency of different designs.

The best / most efficient design is one dimm per channel, especially with single rank RAM. Dual density RAM can add some additional workload to the MC on the CPU - so theoretically a single density 8gb stick will hit higher speeds than a dual density 8gb stick. Important to know - but we are really looking at Mobo architecture and how it affects the efficiency of the system, thus overall, dedicated OC boards will have two memory slots (1DPC).

When you start moving to 4 DIMM designs (2DPC), the different architectures of how to get each of two DIMMs to talk to a single MC introduces some latency, and in a way that is unique to each architecture.

Daisy Chain is simple, and cheap but can introduce noise even if you only have two sticks (one per MC) making it less efficient than one dimm per channel boards - and with four will run at the lowest frequency of whichever stick is struggling the most. In the DC design, one Dimm per channel will be the 'preferred' slot while the other has more latency... Thus filling out all 4 slots guarantees that the timings will limited to the slowest slot.

T-Topology is more complex and has longer paths than Daisy Chain, but it's architecture works better for 4 dimms (not really a 'preferred slot' design) - giving this architecture a better shot at higher frequency timings with 4 than DC can get. Even so, with only 2 slots filled the path length of a T-T design is less optimal than a 1DPC and still longer than a DC board.

So - all of that above answers 'where does the old saw come from' - but it doesn't explain why TechSpot got a different result through their testing. While not specifically addressed, I'm guessing the mobo(s) used were T-T. The newer Ryzen chips' MCs may also handle 4 DIMMs better than previous CPUs.
 
Last edited:
I remember it first coming back from Socket 939 days when using 4 sticks got issues. Not enough voltage or something. As 939 was massively popular back then (like Ryzen now) I guess a lot of folks just went off 4 sticks. Though I have run 4 sticks fine in SKT775 etc. boards. Before then you stuck in as many sticks of whatever you had around in the old SDR days. I think I had a old VL-Bus 486 board that had 8 1MB sticks in it.

However, I happen to have a setup that actually encourages 4+ sticks for quad channel (x99).
 
I remember it first coming back from Socket 939 days when using 4 sticks got issues. Not enough voltage or something. As 939 was massively popular back then (like Ryzen now) I guess a lot of folks just went off 4 sticks. Though I have run 4 sticks fine in SKT775 etc. boards. Before then you stuck in as many sticks of whatever you had around in the old SDR days. I think I had a old VL-Bus 486 board that had 8 1MB sticks in it.

However, I happen to have a setup that actually encourages 4+ sticks for quad channel (x99).
1366 socket also had massive issues with 6 sticks of ram. Some platforms also don't seem to benefit from 2 sticks per channel or dual rank either like x299 or 3647 lol.
 
I used 4 sticks 2x G.Skill and 2x Team back in my Q6600, ran overclocked for 7 years, except the G.Skill was downgraded because the Team sticks had slower timings, nevertheless, it worked fine.
.
Now I have 4 Dominator Platinum sticks, 32GB total and I'm yet to have issues with my system, and I've cracked my 10700KF to 5.2GHz. So.............God knows where that "MYTH" came from.
.
And there's the pic with the 4 sticks............it works fantastic. Then again, my system is not known for being one of the most problematic platforms to have issues with RAM in general. I tend to go that way all the time.
I ran my Q6600 OC'd as well with 2 different sets of ram kits for years with no issues.
gtx285.jpg

have 4 sticks in my new build too, but same make and model
IMG_1648.JPEG
 
The reason two sticks are generally better is because most motherboards only support a dual channel configuration, and generally the ideal configuration is one stick per channel. So if you have a dual channel configuration, usually having one stick on each channel should allow you to reach the best speed. It really depends on how they design the board, though. Most boards should be able to run four sticks, especially if you stick to the motherboard's recommended RAM speed and buy all four sticks as part of a single kit so you know all the RAM is the same type. The one thing you REALLY don't want to do is run just a single stick of RAM on a board that can do dual channel. Like just putting a single 16GB stick in to get 16GB of RAM. That would be worse than four sticks of RAM for sure, because then you drop down to single channel.

The game changes a lot if you have, say, quad channel memory. If you have four channels, then anything less than four sticks of RAM is sub-optimal, and you will probably have eight RAM slots on the board anyway. So that advice is very conditional on the assumption that you only have two channels, the board is designed in a way that two slots are favored, and you might be overclocking the RAM. If any of those assumptions don't hold true, then things change.
 
The articles finding four sticks faster than two were actually re-discovering the benefits of dual ranking. With dual ranks per channel, you can interleave refresh cycles and get an improvement. You can see the identical effect by using dual ranked sticks (16GB sticks with 8gbit memory chips); with two sticks per channel you are automatically getting dual ranking. Electrically, two sticks per channel is a bit harder to drive, as you have additional capacitances and small timing variations to deal with. Depending on how robust the IMC is, and depending on the luck of the sample, you may or may not be able to reach the same top speeds with four sticks (2 per channel) as opposed to a pair.
 
Hey - thanks for that vid - it was great!

Also it led me to this, which explains T-Topology and daisy chaining



Still watching it - but I now need research my Mobo to see what Gigabyte did on my board

Edit - this video also (finally) explains why I hear single rank is better than dual rank for overclocking


If you consult the motherboard comparison spreadsheet, there's a good breakdown of which boards are T-topology vs. Daisy chain.
 
current rig has 4x32gb 3600mhz havent had any issues im not in pursuit of the bleeding edge here either but in gaming its working perfectly fine no issues or huge lag spikes. or weird oddity's. i tried watching understanding the GN video on this but still didnt quite get it as i dont see any issues.
 
current rig has 4x32gb 3600mhz havent had any issues im not in pursuit of the bleeding edge here either but in gaming its working perfectly fine no issues or huge lag spikes. or weird oddity's. i tried watching understanding the GN video on this but still didnt quite get it as i dont see any issues.
It's not about 'issues' in the sense that it causes a conflict that might prevent you from using the computer, rather the architecture choices that allow two MCs on the chip to talk to 4 sticks come with a relative efficiency cost compared to just having 2.

In other words - the 2 dimm solution can be moderately faster than either 4 dimm solution, but they should all work (just not equally efficiently as each other)
 
ahh see that makes more sence lol, though i dont see any point in downsizing to say 2x32gb sticks as i can use the ram i have to fullest extent, would it really be that noticeable in say video rendering/coding occasional gaming? or will this kill the imc on cpu over time since it's more stress on the imc of the ccx's?
 
ahh see that makes more sence lol, though i dont see any point in downsizing to say 2x32gb sticks as i can use the ram i have to fullest extent, would it really be that noticeable in say video rendering/coding occasional gaming? or will this kill the imc on cpu over time since it's more stress on the imc of the ccx's?

No, you won't "kill" anything. It's more just whether or not the IMC is stable or not. With Ryzen, it's really up to the silicon lottery at times.

It's not about 'issues' in the sense that it causes a conflict that might prevent you from using the computer, rather the architecture choices that allow two MCs on the chip to talk to 4 sticks come with a relative efficiency cost compared to just having 2.

In other words - the 2 dimm solution can be moderately faster than either 4 dimm solution, but they should all work (just not equally efficiently as each other)

It's not so much the number of DIMMs but the number of ranks on the DIMMs. 4x dual rank DIMMs is going to be harder to run than 4x single rank. And I remember seeing that 4 ranks in any configuration actually gave more performance than 2 ranks. So 2 x dual rank or 4 x single rank DIMMs would be better than 2 x single rank.
 
ok so this setup is pretty much overkill and i can stop worrying im grinding the imcs to dust making it run D.O.C.P 3600mhz cl 14 across 4 dimms.
 
And I remember seeing that 4 ranks in any configuration actually gave more performance than 2 ranks. So 2 x dual rank or 4 x single rank DIMMs would be better than 2 x single rank
Well, see that there is the point of this whole thread - the two techspot articles I link to in the OP are the first I've heard that 4 could be better than 2...and sadly they don't get down deep enough in the weeds to explain the discrepancy between their findings and the very detailed explanation of 2 > 4 in the video I linked in post #4.
 
FWIW I'm running 4 sticks in my 3900X/X570 system.
Two are 16GB G Skill dual rank 3000
Two are 16GB ADATA Single Rank 3200 W RGB

Timings are different for both sets, and I just defaulted to the 3200 XMP timings. With ram at 1.4v and a little boost to Vsoc (+5mv) everythings been perfectly stable for months. I'm not pushing things too hard, and actually have my processor noticeably undervolted, but stability is key for me. At normal voltage my processor seems to be fine, its just an extra 200mhz for 10c higher temps and that's not worth it IMO.

I was using 2x16GB and 2x8GB until last christmas when I realized that 48GB wasn't cutting it for what I use my computer for.

This is noticeably different from Ryzen 1 or Ryzen 3 on my X370 board which was hours upon hours of clusterfuck to get running correctly.
 
Back
Top