Is it bad to have an odd number (usually 3) of modules of RAM?

While configuring various laptops on various website, I realised that very few have the option to have 24 GB of RAM, the majority of websites offer either 16 or 32 GB.

Is there something not ideal with having 3 modules of RAM, as opposed to 2 or 4? Why most resellers won't give you the option to choose 3?


EDIT: WRONG TITLE, MY BAD: it should say "Is it bad to have an ODD number (usually 3) of modules of RAM?", not "even number"

 

Because 3 modules of RAM means one large module is in dual channel and the other stick is in single channel. Odd configurations like that are not really beneficial to the system, though aside from battlefield 4 (the only program I know of that will literally REFUSE TO WORK RIGHT unless you have properly paired dual channel modules of RAM in your system) I don't know of any real downsides... maybe stressing the IMC a bit?

It's also quite rare you'll find a single stick selling on websites like amazon etc.

As for why not the option? Take your pick.

 

Even when all 3 modules are identical?

 

Yes. Dual channel is something that requires pairs of modules. So you need a multiple of two for all sticks to run in dual channel mode.


Sent from my iPhone using Tapatalk

 

I have a 4GB stick and a 8GB stick.. Both should be DDR3-1600 CL!1 but one is Hynix and one is crucial.. Will it still run in dual channel mode?

 

Your machine is running at what is known as "asymmetric" or "asynchronous" dual-channel mode. Better than plain "single-channel" and worse than full-blown "dual-channel".

Pretty old article, but covers the essential issues decently:

Synchronous vs. asynchronous Dual Channel - You Want 4 GB RAM on Your Notebook?

 

What is the performance difference really?
I just ordered a laptop with only 1x8gb stick of RAM (instead of 2x4gb matched pair) so I'm curious if the performance difference is noticeable?

According to these tests:
http://www.gamersnexus.net/guides/1349-ram-how-dual-channel-works-vs-single-channel/Page-3

The single RAM module vs dual channel "matched pair" is a 17% advantage only when doing pure computation tasks; when running games and other common applications the difference is slight, or not significant.

Edit to add:
And older test from Tom's Hardware (2007) also says in conclusion there is not much difference:

http://www.tomshardware.com/reviews/PARALLEL-PROCESSING,1705-15.html

 

PlaneRider404, you won't notice performance difference. Sooner or later you'll add another 8GB stick to your laptop, so you made a wise decision.

 

Yeah. Probably not noticeable everyday. But probably makes a difference encoding or other highly computational stuff like distributed computing. Probably some games will perform better but not big difference. Ram is just too damn fast.

 

If BF4 is your game, you WILL notice extra stutter and get far less frames than you currently get unless you introduce proper dual channel configurations. Otherwise, most other games seem to be okay with it.

 

I don't think any workload you're likely to come across will notice, as dual vs single channel RAM is purely for bandwidth as far as I know, and you likely won't be maxing that out.

Synthetic Benchmarks - Parallel Processing, Part 2: RAM and HDD

Quick googling finds this. Essentially 0 difference in their benchmarks, which makes perfect sense, as the bottleneck with RAM is latency, not bandwidth.

 

I wouldn't worry much about it. As pervious posts indicated,there isn't really much of a performance difference between even and odd numbers of DIMM modules.

Also, 3 isn't an even number

 

It will run in Flex Mode.

If you calculate the maximum theoretical bandwidth of dual or flex mode, you will find that it will exceed the CPU's FSB/QPI.

Avoid single channel as that is a good way to bottleneck the CPU.

 

Damn! Oversight! I'll change the thread title now. I wonder how many people I confused with that title..

Sigh, it doesn't let me change the thread title

 

I've fixed that for you.

You might find this table that I prepared over 3 years ago to be of interest as I checked the effect of different RAM configurations. I think the conclusions are still relevant as there have been no major changes to either the RAM or the memory controller.

John

 

Thanks John.

I'm looking at the table... I'll be honest, I'm not sure I understand 100% of it.
But I noticed something. The penultimate entry on the first column.. doesn't that amount to 12GB, as opposed to the stated 6?
Or maybe it was supposed to be 2 x 1GB + 1 x 4GB (that would amount to 6).

 

You've passed the test to check if you read the table. That entry should read 1 x 2GB + 1 x 4GB = 6GB.

Ignoring WEI (included because people often refer to it) and table shows that the best results are with full dual channel (ie two modules of the same capacity) and the worst are with single channel (one module) with the mixed pairs of RAM in between. However, while the memory bandwidth tests clear show that single channel is well behind the others, the spread of results for the combined benchmarks in the last 4 columns is far less because of the impact of other components in the overall system.

I work on the principle that any pairing of two SODIMMs is better than only one module (I'm currently running with 8GB + 4GB because I found that 8GB wasn't enough on some occasions but I never need the full 16GB which just pushes up the size of the hibernation file and time needed to hibernate / resume).

John

 

Naturally dual channel will be superior to single channel, but bandwidth in your memory is not the bottleneck.

And nothing in terms of RAM is going to change that - the biggest improvement to performance has not been about increasing bandwidth, which is incredibly cheap (you can "buy" bandwidth), but instead the issue is latency.

All of those CPU improvements you hear about, all of those architectural differences, are about reducing latency, because that has a far more deterministic effect on workloads.

Why? Because RAM is terribly slow, and it doesn't get faster at the same rate as CPUs = in other words, the gap between CPU and RAM increases.

So imagine you've got a CPU and RAM. Your RAM can send 1 message to the CPU at a time in single channel mode, and 2 messages in dual channel mode. Woah, 2x speedup, right? Except every time your CPU goes to RAM it takes about a 50-500x penalty. So you fit twice as many messages, but you're still sitting there for a billion clock cycles waiting for that.

That's why I don't think that bandwidth in your RAM is really going to matter very much. It can't make up for the latency.

The only workloads that might be affected are ones where you can actually saturate the entire bus to the CPU. Any workload that *does* get affected is probably doing it wrong anyways, because they should either be spending way more time in the CPU or way more time in the GPU.

RAM really is great for one thing - not sucking as much as Flash or disk drivers. It's a few orders of magnitude faster than that, so for buffers or caching file access it's amazing. Once you have enough RAM to buffer/ cache the files you're most likely to access, you'll start seeing very few returns.

That said, of course, dual channel is better. Just don't imagine that you'll be seeing very large differences.

If anyone has something showing otherwise, showing an actual workload or perhaps a game jumping up in frames, I'd be very interested in seeing it.