UNSOLVED

Marcel1971

updated

3 months ago

0

123681

January 3rd, 2014 11:00

CPU performance R610 2x X5650 vs. R620 2x E5-2630

Hi Dell Team,

I have 1 server R610 running Ubuntu with 2 Xeon CPU's X5650 (bought in 2011). and i have 1 R620 running Ubuntu with 2 Xeon E5-2630 CPU's (bought 2013).

I would expect the E5-2630 CPU's to be faster, but i noticed that with the same job it was significant slower. So i started to do a test.

I tested with: sysbench --test=cpu --cpu-max-prime=20000 run
This starts a test with 1 thread for a maximum prime number check.

With R610 it lasted 23.47 seconds

With R620 it lasted 32.67 seconds.

I would think the R620 would be a little bit faster then the R610, however it is a lot slower!

I have set the BIOS values as described in this paper:http://i.dell.com/sites/content/shared-content/data-sheets/en/Documents/configuring-low-latency-environments-on-dell-poweredge-12g-servers.pdf

And i upgraded the Bios of the R620 to 2.1.3 (latest version), but no still no improvement.

Difference between the R610 and R620:

R610: Ubuntu 11 - 12 Gb Ram

R620: Ubuntu 12 - 8 Gb Ram

The processes i use (imagemagick) do almost not use memory. The CPU power is all it needs....

What else can i do to improve the CPU speed for the R620. As my program heavily depends on the CPU.

  • Daniel My

    12 Elder

    •

    6204 Posts

    3192

    0

    Posted January 3rd, 2014 13:00

    Hello again Marcel

    There have been many reports of performance benchmark tools rating older servers as better than our 12th generation servers. This is typically due to new power saving features. Our 12th generation servers are the most power efficient we have released. These power saving features can cause slower performance in some scenarios. For instance, if everything is idled down and you start a benchmark it may come back with bad results because there is a startup delay on increasing the voltage on memory and CPU processing.

    To make sure this is the issue I would recommend that you boot into the system BIOS and change the system profile and disable all power saving features. Run your tests again and if they produce the desired results you should adjust the power saving features to meet your needs. I would not recommend leaving all power saving features disabled as it could lead to a dramatic increase in your electric bill. In a single server it may not matter much, but in a datacenter the change could be very significant.

    If the issue persists after disabling all power saving features then let me know and we could look into other possibilities.

    Thanks

  • 3192

    0

    Posted January 3rd, 2014 14:00

    Hi Daniel,

    Thanks for the answer.

    I guess i can not change the BIOS settings from a remote location through an iDrac Express?

    And you mention to disable "all power saving features". Which are those exaclty? (before i start making my trip to the datacenter)

    Are that the following settings?

    - "CPU Power Management" -> "Max Performance"

    - "Turbo boost" -> enabled

    - C States -> disabled

    - C1E -> disabled

    - Monitor / MWait -> disabled?

    Are that the ones, or did i skipped some settings?

    Regards,
    Marcel

  • Daniel My

    12 Elder

    •

    6204 Posts

    3192

    0

    Posted January 3rd, 2014 14:00

    I just edited my above post to correct a very big typo. "QPI speed, cache, and all prefetching features on the processor should be enabled." Changed disabled to enabled on that line.

  • Daniel My

    12 Elder

    •

    6204 Posts

    3192

    0

    Posted January 3rd, 2014 14:00

    And you mention to disable "all power saving features". Which are those exaclty? (before i start making my trip to the datacenter)

    If you followed the article that you listed and set to all of the recommended settings then all power saving features would be disabled. For testing purposes turbo boost, C states, advanced memory error correction, and scrubbing/monitoring should be disabled. QPI speed, cache, and all prefetching features on the processor should be enabled.

    Yes, if you don't have console access via a remote KVM or iDRAC enterprise then you will need physical access to change BIOS settings.

    Thanks

  • Solli

    1 Rookie

    •

    11 Posts

    3199

    0

    Posted January 6th, 2014 00:00

    The X5650 is a 6C 2.66Ghz (max turbo 3Ghz)

    The E5-2630 is a 6C 2.30Ghz (max turbo 2.8Ghz)

     

    I would expect the X5650 to be faster for pure cpu intensive tasks.

     

     

  • pcmeiners

    6 Operator

    •

    1771 Posts

    3199

    0

    Posted January 8th, 2014 11:00

    Have not used this but it might help find the bottleneck...

    http://bluestripe.com/express/monitoring-tool-features.html

  • 3199

    0

    Posted January 23rd, 2014 13:00

    Hi Daniel,

    I have done what you asked, but it did not help.

    When turning off the Turbo, the perfomance measure took 36 seconds. (Whit Turbo on, it was 32 seconds).

    So i turned Turbo on.

    See attached screen shots of the Bios Settings.

    Integrated Devices: http://www.online-image-editor.com/dell/integrated_devices.jpg

    Memory Settings: http://www.online-image-editor.com/dell/memory_settings.jpg


    Processor Settings: http://www.online-image-editor.com/dell/processor_settings.jpg

    System Profile Settings: http://www.online-image-editor.com/dell/system_profile_settings.jpg

    Regards,

    Marcel

  • Daniel My

    12 Elder

    •

    6204 Posts

    3199

    0

    Posted January 23rd, 2014 15:00

    Marcel

    I see one option disabled which will have a significant impact on performance. The very first option under processor settings for Logical Processor is disabled. This disables hyperthreading on the CPU.

    Most newer operating systems are capable of switching between hyperthreading on and off with little or no issues, but please check on your specific version of Ubuntu to make sure there will not be OS corruption when enabling this feature.

    Thanks

  • 3199

    0

    Posted January 24th, 2014 01:00

    I enabled that option. But still no improvement of performance. 32.6 seconds for the cpu test.

  • 707

    0

    Posted January 25th, 2014 12:00

    Enabling Monitor/MWait speeded up to 30.4 seconds. However, still far away from the X5650.