mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2018-11-15, 19:29   #1
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

487 Posts
Default Fascinating Lenovo memory configuration paper

Presented here, for your education and edification, is a rather interesting paper from lenovo I ran across that describes and illustrates, via lots of diagrams, what kind of memory access percentages you can get from the latest 6 channel Xeon processors with different configurations of dimm slots populated.
For instance, it shows 97% of maximum possible bandwidth if you have 6 dimms configured correctly. If you add a seventh dimm, bandwidth drops to 30% !!!

http://lenovopress.com/lp0742.pdf

But, you may say, "I have no Xeons at all, let alone any with 6 channel memory access".
Remember, these are Intel processors and lessons learned here are probably applicable to most of their other processors, even if they are 2 channel ( most are in this category ) or 4 channel.

As George has said many times, with the newer processors, Prime95 is mostly memory access constrained.

Last fiddled with by tServo on 2018-11-15 at 19:32
tServo is offline   Reply With Quote
Old 2018-11-16, 09:33   #2
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

33·5·73 Posts
Default

There is some valuable information here. There is an explanation of increased memory performance with multiple rank DIMMs.
Quote:
Within a memory channel, a second level of interleaving called memory rank interleaving can occur. A memory rank is a block of data created from the memory chips on a memory DIMM.A memory rank is typically 64 bits wide. If ECC is supported, an additional 8 bits are added for a total of 72 bits. A DIMM may contain multiple memory ranks with one, two and four rank DIMMs being the most common.

Memory rank interleaving generally improves memory performance as the total number of ranks on a memory channel increases, but only up to a point. The Intel architecture is
optimized for two to four memory ranks per memory channel. Beyond four ranks per memory channel, performance can slightly degrade due to electrical turnaround time on the memory channel when the memory controller switches between memory ranks.

Last fiddled with by kladner on 2018-11-16 at 09:35 Reason: Weird quote box behavior
kladner is offline   Reply With Quote
Old 2018-11-16, 10:12   #3
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

7·19·41 Posts
Default

I just tell the minions to fill all the slots on the mobo with all modules matched. Saves a lot of explaining, fiddling, and time wasting.
retina is offline   Reply With Quote
Old 2018-11-16, 11:16   #4
mackerel
 
mackerel's Avatar
 
Feb 2016
UK

37510 Posts
Default

It is interesting looking at the unbalanced configurations. I like to keep it simple myself. Generally I run one module per channel, same module types on all channels. On the rare occasion I have surplus ram, I might put two modules per channel, which if they were single rank, sees some gain by having two rank per channel in that case.

I did find it interesting on my laptop, which came with single channel 8GB, so I put in another module from my old laptop bringing it to dual channel 16GB. Where it got interesting is that my old module was 2R, the new module was 1R. In tests I still got the bandwidth upgrade and Prime95 benches scaled within tolerances.
mackerel is offline   Reply With Quote
Old 2018-11-16, 14:10   #5
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

487 Posts
Default

Quote:
Originally Posted by retina View Post
I just tell the minions to fill all the slots on the mobo with all modules matched. Saves a lot of explaining, fiddling, and time wasting.
That could be quite expensive.
Some Dell workstations ( T7910, T7920 ) have 16 slots.
I'm sure lenovo and HP are the same.
tServo is offline   Reply With Quote
Old 2018-11-16, 14:16   #6
tServo
 
tServo's Avatar
 
"Marv"
May 2009
near the Tannhäuser Gate

487 Posts
Default

Quote:
Originally Posted by mackerel View Post
It is interesting looking at the unbalanced configurations. I like to keep it simple myself. Generally I run one module per channel, same module types on all channels. On the rare occasion I have surplus ram, I might put two modules per channel, which if they were single rank, sees some gain by having two rank per channel in that case.

I did find it interesting on my laptop, which came with single channel 8GB, so I put in another module from my old laptop bringing it to dual channel 16GB. Where it got interesting is that my old module was 2R, the new module was 1R. In tests I still got the bandwidth upgrade and Prime95 benches scaled within tolerances.
What that says, to me, is that having the extra channel overrides the rank.
This paper has made me examine my machines that run memory & cpu intensive workloads ( like Prime95 ) and plan some upgrades.
tServo is offline   Reply With Quote
Old 2018-11-16, 18:32   #7
retina
Undefined
 
retina's Avatar
 
"The unspeakable one"
Jun 2006
My evil lair

545310 Posts
Default

Quote:
Originally Posted by tServo View Post
That could be quite expensive.
Some Dell workstations ( T7910, T7920 ) have 16 slots.
I'm sure lenovo and HP are the same.
Compared the the amount of time wasting and lost work, it is much cheaper to buy extra memory sticks than have them playing around trying fewer modules in various arrangements.
retina is offline   Reply With Quote
Old 2018-11-17, 16:28   #8
chris2be8
 
chris2be8's Avatar
 
Sep 2009

71116 Posts
Default

There's an earlier paper http://lenovopress.com/lp0501.pdf which studies Xeons with 4 or 8 memory channels per core. That may be more relevant to consumer kit.

Chris
chris2be8 is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Fascinating periodic sequence pairs doctornash Other Mathematical Topics 7 2018-07-14 00:06
Optimal LL configuration aurashift Hardware 11 2015-09-22 14:09
configuration for max memory bandwidth smartypants Hardware 11 2015-07-26 09:16
Best configuration for linux + dual P4 Xeon + hyperthreading luma Software 3 2003-03-28 10:26
Multiple systems/multiple CPUs. Best configuration? BillW Software 1 2003-01-21 20:11

All times are UTC. The time now is 15:19.

Sat Jun 6 15:19:08 UTC 2020 up 73 days, 12:52, 1 user, load averages: 1.65, 1.78, 1.74

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.