How much does memory speed matter

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

bmacklin

Member
Dec 10, 2013
96
5
8
Two questions:

  • How much performance do you gain from using 1600 MHz ram instad of 1333 in a server or workstation workload with two processors? A lot of discussion using gaming benchmarks, but I can't find much on stuff like linux bench scores by changing the memory
  • How much power savings from using 1.35V DDR 3 vs 1.5 V ram? I found some 1600MHz 1.35V (8x8GB) for $200, and 8x8GB 1333 MHz 1.5 V for $140. Wondering if the $60 difference is worth it in terms of performance and power savings?
 

andrewbedia

Active Member
Jan 11, 2013
698
247
43
If you're using Redis, it matters. doesn't matter for anything else, really. very negligible, if any performance increase. RAM speed is almost never the bottleneck in a system.
 

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
7,625
2,043
113
If you're using Redis, it matters. doesn't matter for anything else, really. very negligible, if any performance increase. RAM speed is almost never the bottleneck in a system.
Any "in memory" usage would technically benefit from faster memory the problem is USUALLY things go in and out so fast you don't notice but it becomes more noticeable the harder you're hitting the memory... more is still subjective because maybe for your workload 1s faster per-whatever is what you need where others may want 5s faster or maybe someone wants 1ms faster, etc... also remember that memory frequency isn't the only thing, you should also be sure to check latency. I made a thread a couple months ago that linked to numerous resources that went into very good details about DDR3, DDR4 frequency & latency, etc...

Would I spend 10% more for fastest/faster memory sure, would I spend more than that not likely.... my home server runs on 1033 I believe not so fast, but affordable and high density.

Remember too that if you populate more channels the board may down clock the frequency anyway.
 

Keljian

Active Member
Sep 9, 2015
428
71
28
Melbourne Australia
The difference between 1.5 vs 1.35 v is about 2-3w for four sticks. Most 1.5v ram will run quite happily at 1.40-1.42 which basically negates the difference in power.

Regarding speed, that's a big "it depends". If you are running anything that demands high memory bandwidth, then you will notice a difference. I notice it when doing video work (I run 32 gig of ddr3-2400 in my desktop). For general office apps you won't notice it.
 

Deslok

Well-Known Member
Jul 15, 2015
1,122
125
63
34
deslok.dyndns.org
Knowing what application you're planning to use primarily would help us evaluate the impact of memory speed, Scale matters here a lot as well, dual channel DDR3(not sure on speed off hand) is fine for a VM host i have at home but can be limiting in an environment I'd use at work(far higher vm count so more potential demand for bandwith) My understanding is that one of the favorite filesystems here ZFS also could show an impact on memory speed but the amount would be based on use case specifically.
 

EffrafaxOfWug

Radioactive Member
Feb 12, 2015
1,394
511
113
I'd concur with Keljian that the power savings amount to about 0.5-1W per stick as far as DDR3 goes; I don't think we've done any testing regarding different voltages on DDR4 yet (probably due to availability, think all of our DDR4 stuff is 1.2V). By our lifecycle costs the marginal extra cost of the lower-power stuff pays for itself over the course of a year or so.

Very much a £0.02/YMMV statement, but I don't think I've seen any real-life workloads since the start of the opteron era where memory speed made a significant difference to the overall result, and I think especially as of nehalem it's become even less of a bottleneck. There's almost certainly a bunch of workloads (and e-peen benchmarks) that are highly dependent on memory throughput, just saying that I've not run into any in years, I'd be interested to see if any readers have run into any.

TechReport's memory latency review from 2005 still makes for an interesting read. The Anandtech Haswell one is more up to date but not as well presented.
 
  • Like
Reactions: bmacklin