Database tool of CPU performance/power consumption

Discussion in 'STH Suggestions and Updates' started by nry, Mar 20, 2018.

  1. nry

    nry Active Member

    Joined:
    Feb 22, 2013
    Messages:
    311
    Likes Received:
    60
    Often when reviewing server CPUs I turn to the articles posted on here for benchmarks and power consumption.

    Sometimes the CPUs I’m comparing don’t show up in the same graphs, it would be amazing if we could select specific models and do a side by side comparison of figures for specific models.

    Also idle power consumption is a big one, although I appreciate this is hard to report when you have additional hardware.

    I’d be happy to help out with a backend app for this!
     
    #1
    Son of Homer likes this.
  2. Patrick

    Patrick Administrator
    Staff Member

    Joined:
    Dec 21, 2010
    Messages:
    11,156
    Likes Received:
    4,114
    That was the original concept around Linux-Bench. Perhaps it is something to revisit next month as we prepare to update the suite.
     
    #2
    malloot likes this.
  3. Son of Homer

    Son of Homer Member

    Joined:
    May 9, 2016
    Messages:
    170
    Likes Received:
    15
    I second getting even an indication of idle power consumption.
     
    #3
  4. Patrick

    Patrick Administrator
    Staff Member

    Joined:
    Dec 21, 2010
    Messages:
    11,156
    Likes Received:
    4,114
    I actually want to make a 50-70% load number what we use going forward.

    Few people run at 100% and heavy AVX workloads. If you are, then you know you are going to use a lot of power.

    Idle is less exciting these days. Remember, the AST2500 BMC, RAM chip for the BMC, PSU idle and etc all account for a very significant portion of 1P system idle.
     
    #4
    i386 likes this.
  5. nry

    nry Active Member

    Joined:
    Feb 22, 2013
    Messages:
    311
    Likes Received:
    60
    My main reason for idle power consumption is due to the kit running at home, every watt makes a difference to the electricity bill [emoji3]
     
    #5
  6. Evan

    Evan Well-Known Member

    Joined:
    Jan 6, 2016
    Messages:
    2,296
    Likes Received:
    317
    @Patrick all for a 50% used (think 70% is too high) but I also do like idle, big difference a home server idling at 30w vs 50w and that’s what a lot of systems do at night say.

    Which is why I am looking to optimize a system or pair of systems for always on usage but still able to run ok when a bit loaded, example I guess is it looks like an X11SDV will end up idling certainly no less than an X10SRM with e5-2640v4

    Either your figures or mine seem to be off a bit.

    (I live in the land of 20+ cent kwhr power)
     
    #6
  7. BackupProphet

    BackupProphet Well-Known Member

    Joined:
    Jul 2, 2014
    Messages:
    716
    Likes Received:
    250
    Idle varies alot, it only makes sense for a complete system like Dell R230 og similar options. It is too hard to benchmark this.
    But power scaling over number active cores, may be interesting.
     
    #7
    Patrick likes this.
  8. malloot

    malloot New Member

    Joined:
    Oct 24, 2013
    Messages:
    11
    Likes Received:
    1
    Any news on the bench updates? It seems like there is no good tool to actually compare performance for compute based workloads out there, you guys could really make an impact with this. Possibly let people compare their own hardware to the benchmark article so they can compare better. One important aspect that is missing in the articles is the performance per dollar of parts, preferably in a way that you can change the price if you find a deal or get bulk pricing.

    Really hoping to hear updates about this!
     
    #8
  9. Patrick

    Patrick Administrator
    Staff Member

    Joined:
    Dec 21, 2010
    Messages:
    11,156
    Likes Received:
    4,114
    The model we have been using is more impactful: we have been running benchmarks for companies using their exact software/ workload.

    It is not a free service since it costs us a lot to do, but it has helped us scale while others are cutting back.

    Also, performance per dollar of a part is really only valid to compare in the same platform and on a TCO basis. We do put some of that together, but that is why it is not emphasized as much.
     
    #9
  10. malloot

    malloot New Member

    Joined:
    Oct 24, 2013
    Messages:
    11
    Likes Received:
    1
    Hey, so where does my company sign up for that service? We could really benefit from that as we are currently always having to deal with suppliers for samples and long lead times in order to bench hardware. Can we PM about this?
     
    #10

Share This Page