Can more memory can make a computer slower?

[IlllI]

I was reading a topic and someone typed out this tidbit of info:

"8gb is plenty for most, 16gb will actually be slower than 8gb most time unless you keep a ton of stuff open."

When questioned, it was elaborated:

"common tech. knowledge, that higher density memory is actually slower.
plus, boot times are slower on more memory as the P.o.s.t. memory check has to go thru it all; memory clocks should be enough to flesh that out for you."

This is the first I've heard of such a thing. Is it accurate? I can think maybe with old operating systems that were not designed to take advantage of the higher amount of ram today... but for windows 10 I can't fathom a computer running slower with 16Gb of ram vs 8. (I tried finding some benchmarks to see if the person was right and couldn't)

 

[nosirrahx]

That might have been true long ago, I can't see any modern situation where it would be.

In theory a memory speed that is stable at 1 stick per channel could become unstable at 2 sticks per channel but even that is very unlikely to ever happen.

 

[UsandThem]

Windows 10 loves RAM. If you have extra, it will gladly take extra and put it into standby. I don't think having "too much RAM" for the OS has been an issue for a very long time, maybe back to the early XP pre-service pack and earlier days. But I held on and stuck with Windows 2000 for about 18 months after XP was released, and I don't recall having "too much" memory issues even back then.

I briefly had 8GB of RAM installed with Windows 10 because of the high prices, but I was easily hitting the ceiling and it was causing Windows to use the page file on my disk more often. I finally sucked it up and got another 8GB. I guess it really comes down to how much you got going on at once.

 

[paperfist]

How can it be slower if it's not even being used?

 

[BonzaiDuck]

That might have been true long ago, I can't see any modern situation where it would be.

In theory a memory speed that is stable at 1 stick per channel could become unstable at 2 sticks per channel but even that is very unlikely to ever happen.

Stability is one thing, performance is another, but stability may affect configuration choices making memory slower.

I've always found G.SKILL tech-support to be helpful, insightful, and timely in their e-mail responses. I started with a set of 2x8GB=16GB TridentZ 3200 14-14-14. I decided that I had been short-sighted in rejecting a 32GB 2x16GB option for the TridentZ kit with the same specs and timings. Then RAM prices increased. I talked to tech-support about running two separately purchased kits with the same spec. They didn't want to guarantee stable operation for "unmatched" kits that weren't part of a single 4x8GB kit. I asked them about running them at 3000 14-14-14, and they gave it a thumbs up.

Then I bought the second kit and decided to start testing at the original spec. They passed an HCI Memtest 64 with 1,000% coverage. I don't think it's in the cards for them to run with CMD = 1, but I may eventually try it. I'm pretty sure I'd have to tweak the voltages higher for VDIMM and VCCIO / IMC.

Otherwise, I cannot imagine that more RAM automatically means a drop in performance. Perhaps I should've run the AIDA64 Memory Benchtest before I added the second kit, and then run it again to compare.

But it doesn't seem like a significant possibility, or a possibility that is significant. Anyway, I think this was being discussed in preceding posts as a matter of density -- say, an 8GB stick of my G.SKILL as opposed to a 16GB stick. But -- you see -- if I had the 2x16GB kit, I would bet anyone here $50 that I could run them as CMD=1 with nothing more than a tweak to the IMC voltage, and I could run them faster just as compensation. I still don't think there would be a difference at the same command rate, though.

 

[whm1974]

That might have been true long ago, I can't see any modern situation where it would be.

In theory a memory speed that is stable at 1 stick per channel could become unstable at 2 sticks per channel but even that is very unlikely to ever happen.

The last time I was aware that more memory would slow down your computer was back when L2 cache was on the motherboard and not on the CPU. Which hasn't been the case since the late 90's.

 

[Insert_Nickname]

I was reading a topic and someone typed out this tidbit of info:

"8gb is plenty for most, 16gb will actually be slower than 8gb most time unless you keep a ton of stuff open."

When questioned, it was elaborated:

"common tech. knowledge, that higher density memory is actually slower.
plus, boot times are slower on more memory as the P.o.s.t. memory check has to go thru it all; memory clocks should be enough to flesh that out for you."

This is the first I've heard of such a thing. Is it accurate? I can think maybe with old operating systems that were not designed to take advantage of the higher amount of ram today... but for windows 10 I can't fathom a computer running slower with 16Gb of ram vs 8. (I tried finding some benchmarks to see if the person was right and couldn't)

That might have been true long ago, I can't see any modern situation where it would be.

In theory a memory speed that is stable at 1 stick per channel could become unstable at 2 sticks per channel but even that is very unlikely to ever happen.

The only thing I can think of that'd make some sense is dual rank vs single rank memory. SR might be able to clock higher, but looses performance from there being less physical chips to spread R/Ws over. So essentially its usually a wash in actual performance, with the exception of latency. There high frequency SR has the advantage.

An example would be 2 high frequency SR DIMMs vs 4 DR DIMMs. The SRs might well be "faster" in theoretical bandwidth and actual latency, but would get demolished by the parallelism of the 4 DR DIMMs.

And that's why there is no "right" memory configuration. Always use the right tools for the job.

 

[Billb2]

Common tech. knowledge, that higher density memory is actually slower.plus, boot times are slower on more memory as the P.o.s.t. memory check has to go thru it all; memory clocks should be enough to flesh that out for you."

Low density/high density doesn't matter. The memory runs at the speed at which it is sold to run at.

Why wold any sane person look at memory checks over and over? Normal people turn function off in the BIOS. If there's a memory problem you'll know about it soon enough.

 

[dave_the_nerd]

Low density/high density doesn't matter. The memory runs at the speed at which it is sold to run at.

Why wold any sane person look at memory checks over and over? Normal people turn function off in the BIOS. If there's a memory problem you'll know about it soon enough.

The bios checks dont catch much anyway. Better to run memtest86 proper every now and again.

 

[Carfax83]

I don't think it has anything to do with memory density. Rather, it has something to do with more DIMM slots being populated. The more DIMM slots you have that are populated, memory latency is increased because the memory controller has to do more work. It's kind of a similar principle like when a hard drive slows down when more data is on it.

Basically, if you're going to add more memory, make sure you prioritize increased memory density rather than increased memory modules.

 

[HutchinsonJC]

What Carfax said has merit, but in addition to that, more RAM slots filled can also have an effect on stability when trying to OC a CPU because of the extra work the memory controller on the CPU is doing. That applies to a small % of users who would notice or care about that, though.

Adding more sticks of RAM later down the road, you'll find that the RAM will operate at the lowest common denominator... so in that sense, more RAM can slow you down if you're not taking care to buy the same RAM as what you've already got.

 

[LTC8K6]

Ram running at the lowest common speed, or having a bit higher latency, is not anything anyone would notice unless they scrutinize benchmarks closely.

 

[Billb2]

If, for example, you want 8 gigs total that can be:
1 x 8 gig
2 x 4 gig
4 x 2 gig
Which is available at the highest speed? The 2 gig modules of course.
So, would 4 x 2 gigs of 3000 Mz memory be faster that 2 x 4 gigs of 2400 MHz memory?
If you want performance, speed is king! If performance isn't your goal what are you doing at Anandtech anyway?

The discussion about overloading the memory controller is accurate but not germane.

 

[nerp]

Very overblown concern today. The more ram, the merrier. Any slight latency hit from having more ram is more than compensated for by . . . having more RAM.

 

[mv2devnull]

I don't think it has anything to do with memory density. Rather, it has something to do with more DIMM slots being populated. The more DIMM slots you have that are populated, memory latency is increased because the memory controller has to do more work.

Yes.

You can find documentation from Intel, or ...
https://www.kingston.com/us/business/server_solutions/best_practices/maximizing_performance

• Xeon E5 v3 series processors:
  • Up to one DDR4-2133 RDIMM per channel, or 1DPC (four modules per processor in quad-channel configuration).
    
    
  • In two DIMMs per channel, or 2DPC (eight modules per processor in quad-channel configuration), DDR4-2133 RDIMMs will operate at 1866MT/s.
    
    
  • In three DIMMs per channel, or 3DPC (twelve modules per processor in quad-channel configuration), DDR4-2133 RDIMMs will operate at 1600MT/s.

However, this is mostly for Xeons, where multiple modules per channel is possible, and furthermore memory not being fastest possible does not necessarily mean that your computer is observably slower.

 

[HutchinsonJC]

Unless I'm missing the point of what you're trying to exemplify, that documentation says 1 dimm per channel... 2 dimm per channel... 3 dimm per channel.

So a Xeon CPU's memory controller is taxed with more dimms per channel such that Intel felt it prudent that they lower the RAM's performance characteristics with more slots filled. This isn't a density issue... it's a quantity of sticks the memory controller has to deal with issue.

 

[mv2devnull]

Now that you did remind about density, this might be one of OP's "common tech knowledge" sources:
https://www.dasher.com/server-memory-rdimm-vs-lrdimm-and-when-to-use-them/

Again, should not apply to OP's "8 vs 16 on desktop" case.

 

[mikeymikec]

I've seen one or two boards that if all four memory slots were filled then it couldn't run them at as high a speed, but it's been a long time since I saw that happen on a new board (IIRC 2005). Even in that specific scenario, if your use of the system requires the extra memory, you'll see a net benefit despite the lower RAM speed.

 

[NathanBrazil]

The last time this really mattered for me was about 36 years ago. The Atari 800 had some overhead for RAM refresh, so a BASIC program running on a 16K system could be notably faster than on a 48K system. This could perhaps be found to affect some things on an Atari ST or Amiga but no mention comes to mind. Dedicated memory controllers have been common for so long it hardly seem worth worrying about outside of some highly specialized situations.

 

[Ankur Saxena]

More memory mean your ongoing application can more space to run so your system runs smoothly

 

[C1]

When messing around with RAM for performance then you'll need to read the user guide for your motherboard (memory section). There are all kinds of caveats concerning memory performance depending on the motherboard. These relate to types & brand of memory used, mix and match of brand and types issues, required slots to be installed in, BIOS settings, etc. Many times I see warnings about certain RAM on-board chip configurations that simply are not allowed or if used, will result in the lowest bus speed specification for the subject motherboard.