Read/Write on larger RAM-faster? slower?

[yoadknux]

Hey guys,

Quick theoretical question - Does RAM performance get better or worse if you have an equal amount of memory but with smaller sticks? Quad channel aside, does 4x4 perform better or worse than 2x8?

The smaller questions that lead to this question are:
1. Say you have an application that requires 2GB of RAM. You have either 1x4 or 1x8 - Both more than enough for the application. Will the application run quicker on the 1x4, the 1x8, or doesn't matter?
2. Now let's say you have 2x4 or 2x8 and an application that requires 6GB. Both are enough but obviously the 2x4 is now almost completely filled (while the 2x8 is not). Does the application now run slower? Does RAM become clogged when the individual sticks are filled?

Would love some insight! Thanks!

 

[mackerel]

Ram quantity is enough or not enough, with an indirect effect into OS using excess as cache but this is far less important in the age of SSDs than it was with HDs.

Ram configuration does make a difference.

Assuming a dual channel system with 4 slots, so you can fill 2 or 4 slots and run dual channel in either case. 4x4 may run faster than 2x8 of same specification, depending on the module. Rank per channel makes a difference to both bandwidth and latency. 2 rank per channel typically gives higher bandwidth at the cost of slightly higher latency. Thus with 4x4 you know you have 2 rank per channel and are operating in that state. 8GB modules may be single or dual rank, so that is less clear. The trend is for newer/faster/lower latency modules to go single rank. So 4x4 single rank modules are generally faster than 2x8 single rank, but pretty much same as 2x8 dual rank modules. Of course, the exact sub timings can have an impact here.

When building a system, I prioritise in order:
1, use the available channels e.g. quad if supported, otherwise dual
2a, use the fastest speed modules that doesn't cost silly money
2b, if possible get dual rank modules over single rank ones, where I figure dual rank is worth roughly one speed grade in use, which doesn't show up in synthetics
3, does the module look nice
4, latency is relatively unimportant

However, that is for my uses, and different uses might benefit more from latency than bandwidth.

Side note: I don't have data to see what happens if you go above 2 rank per channel, as you could go to 4 rank if you throw enough ram at it... I never needed high capacity so this hasn't been a consideration.

 

[yoadknux]

Ram quantity is enough or not enough, with an indirect effect into OS using excess as cache but this is far less important in the age of SSDs than it was with HDs.

Ram configuration does make a difference.

Assuming a dual channel system with 4 slots, so you can fill 2 or 4 slots and run dual channel in either case. 4x4 may run faster than 2x8 of same specification, depending on the module. Rank per channel makes a difference to both bandwidth and latency. 2 rank per channel typically gives higher bandwidth at the cost of slightly higher latency. Thus with 4x4 you know you have 2 rank per channel and are operating in that state. 8GB modules may be single or dual rank, so that is less clear. The trend is for newer/faster/lower latency modules to go single rank. So 4x4 single rank modules are generally faster than 2x8 single rank, but pretty much same as 2x8 dual rank modules. Of course, the exact sub timings can have an impact here.

When building a system, I prioritise in order:
1, use the available channels e.g. quad if supported, otherwise dual
2a, use the fastest speed modules that doesn't cost silly money
2b, if possible get dual rank modules over single rank ones, where I figure dual rank is worth roughly one speed grade in use, which doesn't show up in synthetics
3, does the module look nice
4, latency is relatively unimportant

However, that is for my uses, and different uses might benefit more from latency than bandwidth.

Side note: I don't have data to see what happens if you go above 2 rank per channel, as you could go to 4 rank if you throw enough ram at it... I never needed high capacity so this hasn't been a consideration.

Hello,

Thanks for the great comment! I will now respond to what you wrote and then afterwards present the situation and my needs.

What is "rank per channel"? Is single/dual rank a function of the module, or the configuration? How can I know my current rank per module? By bandwidth, do you mean the operating frequency (1600MHz, 1866, 2133, 2400 etc)? In terms of performance, do rank/bandwidth/timings contribute together?

As for my situation specifically - It is a different situation than what I presented in my topic. I have a X79 system, so I have DDR3 quad channel. I am looking for the best gaming performance - I am playing at 1080p (144Hz) with 1080 GTX. I understand memory probably doesn't make a big difference, but after overclocking the GPU and CPU, I wanted to optimize my RAM as well.

I have two different sets: 4x4 CL10/2133 or 4x8 CL12/2133. Since 16GB is more than enough for gaming, I always thought the 4x4 setup is faster because of the better timings. However, I also wondered whether using relatively small sticks (4GB) might do something to performance. I imagine that my 8GB sticks never run out of capacity (individually), However, 4GB might - Does it become slower?

Would love to hear your opinion!

 

[EarthDog]

Its the timings that make it faster. But, unless its a one off title, ram speed makes very little difference in gaming.

When you run out of ram, it writes to the pagefile on your OS drive. Obviously, a hdd/ssd are exponentially slower than system ram. I highly doubt you are using 16gb though. 20+ Chrome tabs and playing pubg or bf4, i dont break 11gb used...

 

[mackerel]

Rank is a module characteristic. CPU-z sometimes reports it, otherwise you can try thaiphoon burner free. What I wrote above assumed DDR4. As far as we know, 4GB modules are all single rank. 8GB modules can be single or dual rank. 16GB modules... no idea, I don't need the capacity to buy them to try out. I'd assume they're probably dual rank, but can't say 100%. When you insert them into a system, you can put more than one module per channel, and thus, you can end up with more ranks per channel than a single module.

With DDR3, things are a bit different. I have 4GB modules that are dual rank and single rank.

Bandwidth is a mixture of things, dominated by the speed of the modules and how many channels you run. Then rank has an influence, and other timings.

In your specific scenario, as you have both kits already, pick some games that are important to you and bench it. The thing about bandwidth is it only really matters when you don't have enough of it. If you have fast enough ram, getting more bandwidth through rank might not make a difference any more. In that case, timings might have more value.

The system should access the ram in parallel in order to realise the bandwidth. Kinda like raid 0. Thus the used ram is spread out evenly on all sticks. You're not going to fill one stick up while others sit idle.

 

[yoadknux]

The system should access the ram in parallel in order to realise the bandwidth. Kinda like raid 0. Thus the used ram is spread out evenly on all sticks. You're not going to fill one stick up while others sit idle.

Good to know! So from that point of view, it doesn't matter how small my sticks are, as longer as the RAM as a whole doesn't get filled.

I checked the 8GB sticks last night and they were all dual rank.

I think I will download some gaming benchmarks and compare the 4x4 to the 4x8. My gut tells me there won't be a huge difference between CL10 and CL12 but I could be wrong. I'll post the results later!