Intel S2600GL - Low memory performence

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

Ponury Typ

Member
Oct 28, 2015
57
12
8
32
AS threat says, its only about 7964 mb/s according to memtest and aida. All settings in bios are setup to performance. Memory clock is 1600. Bios updated to latest version.

Config:
S2600GL
16x4GB DDR3 1600mhz ecc
2x e5-2650l

Any ideas what is causing this low performance?
 
Last edited:

Ponury Typ

Member
Oct 28, 2015
57
12
8
32
So, i contacted my dope dealer, he is gonna send me some 8GB sticks. Then i am gonna test shit out of this server.
 

gabor

New Member
Mar 28, 2017
26
1
3
27
Mine is doing the same thing, I suppose the memtest86+ is reporting the wrong speeds. Tested different modules in a DELL R710, which behaves the same way.. You should check the memory speeds by the OS or via IPMI
 

Ponury Typ

Member
Oct 28, 2015
57
12
8
32
I checked them under debian (host os) and windows (kvm). And everyone is reporting low speeds.
So additional memory came today and i did some additional tests.

1x e2603 + 4x 8GB ~10234 MB/s
1x e2603 + 4x4GB ~ 10141 MB/s
1x 2650l + 4x8GB ~10342 MB/s
2x 2650l + 4x8GB (2 dimms per cpu) ~ 8029 MB/s
2x 2650l + 16x4GB 7342 MB/s

Avrage memory latency 52.547 ns!

Like wtf, i cant get my head around it. Anyone knows what the heck is going on ?
 

Terry Kennedy

Well-Known Member
Jun 25, 2015
1,140
594
113
New York City
www.glaver.org
Mine is doing the same thing, I suppose the memtest86+ is reporting the wrong speeds. Tested different modules in a DELL R710, which behaves the same way.
memtest86+ bandwidth / SPD / etc. support is sadly lacking for modern (where "modern" means at least as far back as Xeon 5400 series) processors / chipsets. Sometimes it completely misses some DIMMs when reporting SPD data (it doesn't seem to understand 3-bank memory, for example). memtest86 (non-plus) might be better, but recent versions require UEFI and then complain about something when they finally boot. They're probably fine for memory testing, not so good for memory info.

If you have the time for it, try bandwidth for real memory subsystem benchmarking (as opposed to testing). You'll get a nice output graph with all sort of interesting stuff, showing you the results of the various cache levels, different types of accesses, etc.

This is a sample (some sort of Xeon 56xx CPU, I forget which). The dips are because the process wasn't nice'd and there were other things going on:
 

Ponury Typ

Member
Oct 28, 2015
57
12
8
32
So i did today tests with some additional programs like mbt or aida and i recived same results as i got on memtest ...
 

MBastian

Active Member
Jul 17, 2016
205
59
28
Düsseldorf, Germany
Check your NUMA(Edit: You checked with one CPU, so that should be ok) and PowerSafe state settings. If in doubt reset your BIOS. You might disable C6 and higher states and set your profile to max performance. Don't change anything else, mostly this does more harm than good on server boards.
In Linux: Check your CPU/power settings. E.g. your cpufreq governor with 'cpupower frequency-info'. As Terry Kennedy already said, use a 'serious' tool like bandwith.
In Windows: Install Passmark or some other benchmarkt tool (that also gives you a false sense of superiority and/or the urge to buy new stuff you don't really need. :) ) with a database and compare.
Otherwise: Maybe everything is as it should be ... as far as I can gleam in a few minutes, an e5-2650l has 2/3-3/4 the memory bandwidth of an E5-2670.
 
Last edited:

MBastian

Active Member
Jul 17, 2016
205
59
28
Düsseldorf, Germany
Abbreviated bandwidth results for my dual E5-2670 Workstation(Asus Z9PA-d8, 8x16GB Samsung M393B2G70BH0-YK, Arch Linux)

Code:
Sequential read (64-bit), size = 128 B, loops = 2059403264, 50271.0 MB/s
Sequential read (64-bit), size = 64 kB, loops = 2753536, 34415.2 MB/s
Sequential read (64-bit), size = 512 kB, loops = 231680, 23160.7 MB/s
Sequential read (64-bit), size = 512 MB, loops = 117, 11903.9 MB/s

Random read (64-bit), size = 256 B, loops = 716701696, 34982.6 MB/s
Random read (64-bit), size = 512 MB, loops = 27, 2753.9 MB/s

Sequential write (64-bit), size = 128 B, loops = 984088576, 24017.2 MB/s
Sequential write (64-bit), size = 64 kB, loops = 1784832, 22300.8 MB/s
Sequential write (64-bit), size = 512 kB, loops = 191104, 19098.3 MB/s
Sequential write (64-bit), size = 512 MB, loops = 100, 10193.9 MB/s

Random write (64-bit), size = 256 B, loops = 515112960, 25144.4 MB/s
Random write (64-bit), size = 512 MB, loops = 7, 662.1 MB/s

Sequential read (128-bit), size = 128 B, loops = 4118806528, 100555.9 MB/s
Sequential read (128-bit), size = 512 MB, loops = 126, 12867.8 MB/s
And with CPU frequency clamped at 2.1GHz ( frequency-set -d 2.1GHz; cpupower frequency-set -u 2.1GHz)
Code:
Sequential read (64-bit), size = 128 B, loops = 1311244288, 32003.0 MB/s
Sequential read (64-bit), size = 64 kB, loops = 1774592, 22180.0 MB/s
Sequential read (64-bit), size = 512 kB, loops = 149248, 14912.8 MB/s
Sequential read (64-bit), size = 512 MB, loops = 92, 9325.2 MB/s

Random read (64-bit), size = 256 B, loops = 455868416, 22258.6 MB/s
Random read (64-bit), size = 512 MB, loops = 22, 2209.3 MB/s

Sequential write (64-bit), size = 128 B, loops = 626524160, 15286.9 MB/s
Sequential write (64-bit), size = 64 kB, loops = 1136640, 14204.3 MB/s
Sequential write (64-bit), size = 512 kB, loops = 120192, 12014.9 MB/s
Sequential write (64-bit), size = 512 MB, loops = 70, 7073.0 MB/s

Random write (64-bit), size = 256 B, loops = 327942144, 16003.0 MB/s
Random write (64-bit), size = 512 MB, loops = 6, 529.1 MB/s

Sequential read (128-bit), size = 128 B, loops = 2622488576, 64024.0 MB/s
Sequential read (128-bit), size = 512 MB, loops = 97, 9924.8 MB/s
Everything seems to check out, so you can stop worrying. :)

Edit: Also forgot to nice up and should have set to 2GHz.
 
Last edited:

Ponury Typ

Member
Oct 28, 2015
57
12
8
32
Hmm ... i guess working with systems on v4 with ddr4 redefine how system should be "snapy" for me. Ok thx for the tests.