Has anyone done any benchmarks or have any experience with the performance degradation that comes from running 4x8GB sticks vs 2x16GB sticks? I bought 2x8GB sticks and it's just not enough. So I'm debating buying another 2x8GB vs buying 2x16GB and trying to dump the 2x8GB. Am I over thinking this? The 2x8GB sticks are DDR-3200 CL 14 (F4-3200C14D-16GTZ).
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Performance degradation with 4 sticks instead of 2
- Thread starter ngaugler
- Start date
- Joined
- May 7, 2005
for me there was zero difference. i mean literally my aida64 cache/memory tests were almost identical with 2x8gb vs 4x8gb vs 2x16gb. running 2x16gb 3200c14 the aida64 results were virtually the same as the 2x8gb and 4x8gb running at 3200c14. of course chipset and imc can help alot in ocing but running stock speeds theres just not much to be gained. go with the extra 2x8 kit and call it a day(itll oc better). with intel, chasing the memory dragon is a waste unless your shooting for crazy high ocs! im sure youve seen Woomacks insane ocing threads.
- Joined
- Jan 12, 2012
Performance degradation in what way? What are you doing that 16 g of memory isn't enough?
- Thread Starter
- #4
No overclocking at this time. I think the most I'd go is 1.4V maybe CL 13 based on Woomack's tests. I did notice that even without overclocking the MaxxMEM results do not come close to Woomack's default MaxxMEM results. I presume it's just the Gigabyte Aorus Z390 Ultra not performing as well as his ASRocks?
Theories on the interwebs are that since the channels are only dual, when you go from two sticks to four sticks there's overhead on the extra workload for the memory controller which causes some degradation in performance over just two sticks. I haven't found any benchmarks or tests to back it up though, hence why I said 'theories' and 'interwebs'.
As for what I am doing, I'm trying to disable Pagefile in Windows and applications keep crashing while gaming. I've used RAMMap to free the Standby List but it's a cat and mouse game that I can never win. I have no idea what my wife was doing the other day but she had the system completely maxed out even though no applications were running and the Pagefile was just trashing my NVMe. I'd rather just have it disabled completely to keep the NVMe from excess activity.
- Thread Starter
- #6
Just theories on the interwebs... that's why I came to the experts to see if anyone had actually done the scientific research to back it up.I've never heard of such degradation/performance differences...please share.
Last edited:
- Joined
- Mar 7, 2008
The number of physical modules doesn't matter, what does matter is the number of ranks on a channel. The modules mentioned I believe are B-die so single rank at 8GB modules. I have no experience with 16GB sticks, would they be dual rank? If so, running 4x8 would be no different from running 2x16. What difference does rank have? In synthetic tests, having 2R per channel give slightly more bandwidth at slightly more latency compared to 1R per channel at otherwise same settings. I've seen bigger throughput differences running 2R per channel with Prime95 like workloads which are heavily ram bandwidth dependent. If I had a choice, I'd rather run 2R per channel than 1R, but that suits my workloads. If you're purely overclocking for speed, 1R per channel is the way to go.
- Thread Starter
- #8
Hmmmm... so I can get 2 more 8GB sticks and be in for a total of $409.98. Or I can return the 8GB sticks and buy the 16GB sticks (F4-3200C14D-32GTZSK or F4-3200C14D-32GTZ) and be in for $411.49. (Due to price changes since I bought the first sticks and restocking fees).
So which one is better for general performance / gaming? I'm not interested in trying to reach crazy clock speeds.
If all memory modules run at the same settings (frequency and all timings, main and sub) then results should be exactly the same. However, 16GB modules have more relaxed timings so at auto, any configuration at higher capacity memory will be slower (not much but will be). The same, 4 memory modules can be a bit slower than 2 ( counting that you compare your 2x8GB to 4x8GB ) but this is not a rule and depends how motherboard will set timings which are in BIOS under auto.
In new IC, 16GB = dual rank, 8GB = single rank. Dual rank in some applications is faster but in similar way act 2 dual rank as 4 single rank modules.
Single rank modules are overclocking better.
Nothing from above really matters while playing games so just pick whatever is easier or cheaper for you.
In new IC, 16GB = dual rank, 8GB = single rank. Dual rank in some applications is faster but in similar way act 2 dual rank as 4 single rank modules.
Single rank modules are overclocking better.
Nothing from above really matters while playing games so just pick whatever is easier or cheaper for you.
- Joined
- Mar 7, 2008
Agree with others. It doesn't really make a noticeable difference for gaming outside of benchmarks. Just get enough for your needs and don't over-think this.
- Joined
- Jan 12, 2012
Windows needs a pagefile. No matter how small. Some things will not run correctly without one.
Designate a static size of like 1024 and you'll be fine.
- Thread Starter
- #14
I can try that when I get more RAM. Just booting up, loading the stupid apps to manage AIO, GPU Fans, etc, and a game causes the system to use 17GB. It's generally around 20GB though. When I turned all Paging files off the apps would crash, I presume due to OOM beacause task manager would show me getting right up to the 16GB limit before crash.
- Thread Starter
- #16
I can grab the screen shot later tonight. There's DropBox, Backblaze and whatever other base stuff starts. Intel RST, Steam, Origin, etc. I haven't even installed VirtualBox or any of my development stuff yet, so there's nothing be sucked up by VMs. Infact I still have virtualization disabled in the BIOS. For some reason Windows just likes to load GBs of cache. I'll do some experiments tonight. I still haven't bought the additional RAM yet.
- Joined
- Mar 7, 2008
Don't count the cached amount in your usage. That it is used is normal, and is a good thing. Basically it will use any otherwise unused ram to cache disk content. So when you reload a file it is reading from ram, faster than a SSD. And the best thing is, because it is a temporary copy, Windows can free it in an instant if a real app really wants to use it.
- Thread Starter
- #18
The only problem I have with it is Windows is throwing it back onto my SSD via the PageFile.sys. If it's really going to cache stuff, fine, but keep it in RAM.
- Joined
- Mar 7, 2008
Run a pagefile. It isn't going to impact performance and might clear up the problems you're having from not having one. If you're really concerned about flash SSD wear buy a small Optane module and put the pagefile on that, but it really isn't anything to be worried about.
- Thread Starter
- #20
I thought Optane was only for SATA interfaced drives? Not NVMe. The drive I has is a full NVMe, not just M.2 interfaced SATA III.
https://www.intel.com/content/www/u...4/memory-and-storage/intel-optane-memory.html
https://www.intel.com/content/www/u...4/memory-and-storage/intel-optane-memory.html
