Single channel dual rank vs dual channel single rank memory

[bubnu]

While trying to put together a relative low end itx system (rx 570 + r3/i3) i got this feeling that 16gb of ram wont be enough for the next 4-5 years (also might have to do some 3d stuff which might req a better gpu and cpu).

No matter what brand I'll go for (intel/amd) i will want upgradability(gonna swap cpu and/or gpu if needed). As I said, I might need to do some 3d stuff and i dont know if 16 gb will be enough for that.

And that led to dilemma: Will a DUAL RANK memory stick perform as well as a DUAL CHANNEL ? what are the pros/cons of each config?

Also I'm gaming on 1080p.

[jj9987]

Dual rank can perform ever so slightly better (difference of up to 5% according to quick google) over single rank memory, but it all depends on the application and use (it could be the other way too!). Realistically, it doesn't matter if it is dual rank or single rank. So it doesn't really matter, if you go single rank or dual rank or one of both.

 

Dual channel is another story though. This brings almost twice as much bandwidth to memory as you get two concurrent channels. Meaning you need at least two sticks and they need to be in correct slots (tho most ITX boards only have two slots, so that's fine).

[Mira Yurizaki]

Memory ranking does not affect bandwidth performance. Think of multichannel memory as RAID 0 for memory.

[bubnu]

Interesting, Will i lose any significant performance by going with an i3 9100f(boosts to 4.2) and a dual rank single chanel 16 gb stick?

Also do you think 16 gb will be sufficient for the next 4-5 years? In gaming I really think it will be fine, but i got no knowledge of 3d programs and what they require(gpu play a role there as well<?>).

[porina]

Get a mobo with 4 slots. Get dual channel ram now. You still have upgrade room for another dual channel kit later on. Wait, just saw, ITX system, implying only 2 slots? Personally I'd still go dual channel 16GB now. Worry about upgrading beyond 16GB at the time it becomes necessary. You might decide to upgrade the system as a whole by then. Unless you have specific demanding tasks, 16GB is sufficient for pretty much any normal uses.

 

15 minutes ago, Mira Yurizaki said:

Memory ranking does not affect bandwidth performance. Think of multichannel memory as RAID 0 for memory.

It does, but in a subtle way. If you have single vs dual rank modules of otherwise the same settings, the theoretical peak bandwidth doesn't change, but you get more practical bandwidth out of dual rank. For normal users, it doesn't matter. For some compute uses, it makes a huge difference. I still don't understand it myself yet, but it seems to be something to do with allowing switching between reads and writes faster that doesn't show up in synthetic benchmarks looking at them separately.

[Mira Yurizaki]

1 minute ago, porina said:

It does, but in a subtle way. If you have single vs dual rank modules of otherwise the same settings, the theoretical peak bandwidth doesn't change, but you get more practical bandwidth out of dual rank. For normal users, it doesn't matter. For some compute uses, it makes a huge difference. I still don't understand it myself yet, but it seems to be something to do with allowing switching between reads and writes faster that doesn't show up in synthetic benchmarks looking at them separately.

That sounds like it has more to do with latency than bandwidth. The bandwidth, at least the "instantaneous" one, isn't going to change.

[porina]

Just now, Mira Yurizaki said:

That sounds like it has more to do with latency than bandwidth. The bandwidth, at least the "instantaneous" one, isn't going to change.

If you run synthetic benchmarks like aida64, there is a small increase for dual rank over single, and actually measured latency also increases. So I see it as a practical bandwidth increase. They synthetic bandwidth results are usually not at all close to the theoretical bandwidth unless you manually tune every last timing to the minimum possible workable value.

[uberas]

32 minutes ago, porina said:

If you run synthetic benchmarks like aida64, there is a small increase for dual rank over single, and actually measured latency also increases. So I see it as a practical bandwidth increase. They synthetic bandwidth results are usually not at all close to the theoretical bandwidth unless you manually tune every last timing to the minimum possible workable value.

AIDA64 reads for dual rank 3000CL16 read/write/copy/latency are better than single rank 3000CL14 in my scenario so it does make difference (see attachments - sorry for quality)

 

 

copy single.bmp

latency single.bmp

read single.bmp

write single.bmp

[kostasmous]

Heey how are you? i wanna ask you a question about my ram Crucial ballistix ( BL16G36C16U4B ) . ive bought a single stick ram ( the one ive mentioned) and i wanna ask you if i but the EXACT SAME one model , so im gonna have these two single stick rams ( AND NOT A FACTORY KIT ) will i be able to dual channel them on ur mobo?? Thanks a loot!! ps: my ram is SINGLE RANK