Personal computing discussed

Moderators: renee, Flying Fox, Thresher

 
meerkt
Gerbil Jedi
Topic Author
Posts: 1754
Joined: Sun Aug 25, 2013 2:55 am

Number of DRAM ranks vs performance

Wed Dec 07, 2016 9:23 am

Any idea if there are cases where there's a performance difference between single- and dual-rank DRAM modules (1Rx8/2Rx8), either in single- or dual-channel uses?

Assuming you're unlikely to hit the memory controller rank limit, are there any other potential advantages to one over the other?
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 9:43 am

More ranks allows more banks to be kept open, which could reduce average memory latency.

Edit: On the downside, more ranks may require looser timings, because there are more chips connected to the same lines on the DRAM bus.
Nostalgia isn't what it used to be.
 
TheRazorsEdge
Gerbil Team Leader
Posts: 219
Joined: Tue Apr 03, 2007 1:10 pm

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 10:07 am

What JBI said. Also, it puts more load on the memory controller and can affect your overclock of that component.

Back in the day, you could usually get higher clocks with 2 DIMMs, but I don't know how much of a factor it is today. I don't really need to push overclocks like I used to.
 
maxxcool
Gerbil Elite
Posts: 855
Joined: Thu Sep 12, 2002 8:40 am
Location: %^&*%$$
Contact:

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 10:33 am

just brew it! wrote:
More ranks allows more banks to be kept open, which could reduce average memory latency.

Edit: On the downside, more ranks may require looser timings, because there are more chips connected to the same lines on the DRAM bus.

This ^

There is a peak of diminishing returns.  Dual/tripple and quad channel ram DOES increase performance by increasing bandwidth and allowing more data to be stored in ram versus pagefile.

*but* the more banks you populate increases the timing sensitivity of the dram modules, and by result can cause latency to go up.  Overall that latency is not truly that impacting.. but it is areal thing.

With some older chipsets and or dram silicon adding additional banks will require reducing the dram speed to allow for stability. 
Cybert said: Capitlization and periods are hard for you, aren't they? I've given over $100 to techforums. I should have you banned for my money.
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 10:37 am

As an aside, some of the server motherboards I deal with at the day job require *quad* rank registered DDR3 DIMMs if you want to max out the memory capacity. Those things are like moon rocks; we had to order pulls from some random Amazon seller to max out the RAM in some test servers we were putting together.
Nostalgia isn't what it used to be.
 
meerkt
Gerbil Jedi
Topic Author
Posts: 1754
Joined: Sun Aug 25, 2013 2:55 am

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 10:51 am

Has anyone ever done an in-depth study/review on number of ranks vs timing needed for stability?

BTW, why isn't >1 rank treated as dual channel?

TheRazorsEdge wrote:
Back in the day, you could usually get higher clocks with 2 DIMMs
Versus 1?!

maxxcool: I'm wondering about in-module ranks, not modules/channels.

An aside: any non-ECC mobos/chipsets that can use ECC RAM (unregistered I suppose) as normal RAM?
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 11:17 am

meerkt wrote:
BTW, why isn't >1 rank treated as dual channel?

Because it's not the same thing. Each DIMM socket is connected to a specific memory channel, so all of the ranks on a given DIMM are connected to the same channel of the memory controller.

meerkt wrote:
An aside: any non-ECC mobos/chipsets that can use ECC RAM (unregistered I suppose) as normal RAM?

AFAIK this is actually fairly common. Since the extra parity lines aren't connected, they are simply ignored.
Nostalgia isn't what it used to be.
 
Bauxite
Gerbil Elite
Posts: 788
Joined: Sat Jan 28, 2006 12:10 pm
Location: electrolytic redox smelting plant

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 11:48 am

Did a lot of research on this for my latest xeon build, its actually very complicated depending on the CPU IMC(s) and motherboard dimm population which varies a lot between generations and even core type - some have multiple. There is also a difference between aiming at latency or bandwidth.

For desktop/115x socket at "normal" memory amounts (2 or 4 unbuffered dimms and you can actually overclock ram) I think dual rank is better at the same clock speeds.
TR RIP 7/7/2019
 
TheRazorsEdge
Gerbil Team Leader
Posts: 219
Joined: Tue Apr 03, 2007 1:10 pm

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 4:36 pm

meerkt wrote:
Has anyone ever done an in-depth study/review on number of ranks vs timing needed for stability?

BTW, why isn't >1 rank treated as dual channel?

TheRazorsEdge wrote:
Back in the day, you could usually get higher clocks with 2 DIMMs

Versus 1?!

Being able to use one module is relatively rare. Most consumer CPUs are optimized for two.

With very old systems (286/386/Pentium), you installed SIMMs in pairs. With dual-channel setups starting in the late-90s, you installed DIMMs in pairs. I.e., Athlon/P4 and later.

IIRC, only the 486, Pentium II, and Pentium III used single modules. And their derivative Celeron parts, of course.

Even then, you usually couldn't get enough RAM on a single stick, so the choice was effectively 2 vs 3-4 modules.
 
DPete27
Grand Gerbil Poohbah
Posts: 3776
Joined: Wed Jan 26, 2011 12:50 pm
Location: Wisconsin, USA

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 4:50 pm

meerkt wrote:
Any idea if there are cases where there's a performance difference between single- and dual-rank DRAM modules (1Rx8/2Rx8), either in single- or dual-channel uses?

Do you participate in paid overclocking competitions?
If yes - See other's responses and/or seek out advice from a professional overclocker.
If no - No, it doesn't make a damn bit of difference.
Main: i5-3570K, ASRock Z77 Pro4-M, MSI RX480 8G, 500GB Crucial BX100, 2 TB Samsung EcoGreen F4, 16GB 1600MHz G.Skill @1.25V, EVGA 550-G2, Silverstone PS07B
HTPC: A8-5600K, MSI FM2-A75IA-E53, 4TB Seagate SSHD, 8GB 1866MHz G.Skill, Crosley D-25 Case Mod
 
meerkt
Gerbil Jedi
Topic Author
Posts: 1754
Joined: Sun Aug 25, 2013 2:55 am

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 5:04 pm

TheRazorsEdge wrote:
Being able to use one module is relatively rare. Most consumer CPUs are optimized for two.

The main difference I've encountered is more bandwidth in dual-channel. A wider bus so that's what you get. But special optimization?

DPete27 wrote:
Do you participate in paid overclocking competitions?

Just curious in general. Though, if given a choice I'd take an extra 1% for free. :)
 
just brew it!
Administrator
Posts: 54500
Joined: Tue Aug 20, 2002 10:51 pm
Location: Somewhere, having a beer

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 5:09 pm

TheRazorsEdge wrote:
Being able to use one module is relatively rare. Most consumer CPUs are optimized for two.

There's a big difference between performing better with two, and requiring two. Modern dual-channel systems will generally run quite happily with a single stick (but with half the RAM bandwidth).

TheRazorsEdge wrote:
With very old systems (286/386/Pentium), you installed SIMMs in pairs. With dual-channel setups starting in the late-90s, you installed DIMMs in pairs. I.e., Athlon/P4 and later.

IIRC, only the 486, Pentium II, and Pentium III used single modules. And their derivative Celeron parts, of course.

32-bit Athlon systems prior to the nForce2 chipset (which was introduced in 2002) were single-channel, as were all Socket 754 Athlon64 and Sempron systems.
Nostalgia isn't what it used to be.
 
TwistedKestrel
Gerbil Elite
Posts: 686
Joined: Mon Jan 06, 2003 4:29 pm

Re: Number of DRAM ranks vs performance

Wed Dec 07, 2016 5:14 pm

I think we have some semantic confusion in this thread between "rank" and "bank", they are not the same thing
 
TheRazorsEdge
Gerbil Team Leader
Posts: 219
Joined: Tue Apr 03, 2007 1:10 pm

Re: Number of DRAM ranks vs performance

Fri Dec 09, 2016 5:34 pm

just brew it! wrote:
There's a big difference between performing better with two, and requiring two. Modern dual-channel systems will generally run quite happily with a single stick (but with half the RAM bandwidth).

32-bit Athlon systems prior to the nForce2 chipset (which was introduced in 2002) were single-channel, as were all Socket 754 Athlon64 and Sempron systems.

While this is true, I don't think it matters.

In cases where DRAM tuning could offer measurable performance gains, the first order of business would be to populate both memory channels. Very few workloads would trade minor latency gains for a huge bandwidth deficit.

Virtually all enthusiast systems would be built this way. World-record overclock attempts would be the notable exception.

At this point, two modules has been the expected configuration for the majority of the home computing history.
 
Vhalidictes
Gerbil Jedi
Posts: 1835
Joined: Fri Jan 07, 2005 2:32 pm
Location: Paragon City, RI

Re: Number of DRAM ranks vs performance

Fri Dec 09, 2016 5:44 pm

It turns out that latency is almost as important as speed. Considering 1) budget, and 2) what speeds your chipset/motherboard/CPU can handle, latency is probably more important.

For more details, feel free to refer to: http://www.anandtech.com/show/7364/memory-scaling-on-haswell/7

Who is online

Users browsing this forum: No registered users and 1 guest
GZIP: On