1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

Overclocking Overclocking on X58

Discussion in 'Hardware' started by dec, 20 May 2016.

  1. dec

    dec [blank space]

    Joined:
    10 Jan 2009
    Posts:
    322
    Likes Received:
    10
    I recently inherited a dusty office computer that had an i7 960 with 24GB of ram in it. Once I cleaned it and got it back in working order I noticed some things that I thought were very strange.

    1. The cpu is meant to run at 3.2ghz (3.46ghz with turbo) but the bios had it set to 2.93ghz max after turbo boost and 2.5ghz without turbo.

    2. The memory was running at 1066mhz at 1.5V even though the kit is rated to 1600mhz and 1.65V

    I set out to get both the cpu and ram to run at the speeds on the tin. The motherboard I'm using is an intel DX58SO2 and there is a 650W antec earthwatts 80+ bronze PSU as well. The memory in use is 6x4GB of ddr3 1600 CL9 made by kingston. I had very few options to control multiplers so this is where I wound up.

    Base clock: 150mhz (stock was 133mhz)
    CPU multipler: 22 to give 3.3ghz
    CPU voltage: (left at stock) 1.2V
    Uncore multipler: 20 to give 3.0ghz
    Uncore voltage: 1.25V
    memory multipler: 10 to make 1500mhz
    memory voltage 1.55V. at 1066mhz the motherboard put it to 1.5V
    memory timings: 9-9-9-27. command rate left to auto.
    QPI strap: 5.8GT/s to make QPI speed 6.602GT/s
    QPI voltage: (left at stock) 1.1V

    I have a windows 7 and manjaro linux dual boot running. Windows was perfectly happy with those settings and when I turned back on turbo boost and speed step even bumped the CPU up to 3.7ghz (reported using cpuz, coretemp, and hwmonitor). On the other hand manjaro (using conky) reported the CPU never going above 2.93ghz. I resolved this issue by disabling turbo boost and speed step to keep the cpu fixed at 3.3ghz which both OS's report as being the same.

    Trying to run some CFD work in manjaro using openfoam has been giving me a hell of a time. Very early in my simulation the process would crash either with a floatint point exception or a segmentation fault. After ensuring the cause was not related to the simulation itself, I decided to change some things with the hardware as follows.

    Raised memory voltage to 1.65V
    Raised uncore voltage to 1.3V
    Raised QPI voltage to 1.15V

    After putting in these settings I have had no issues. I've had undervolted memory be the culprit of CFD errors before and this was a big undervolt to begin with so I wasn't surprised by having the change that. The part that is confusing to me was how much of a pain in the neck this was to test and retest just to get the memory to anywhere near its rated speed. This particular memory kit did not have XMP either.

    I have haswell for my home desktop and changing anything on that is extremely easy. This was the first time I got to play with parts from the X58 days. Was it typical to have to change this many settings for something so simple? Is this motherboard just very restrictive?

    Its also worth noting the system couldn't post with the base clock anywhere between 133mhz and 160mhz that was not 133mhz and 150mhz.
     
  2. LennyRhys

    LennyRhys Oink!

    Joined:
    16 May 2011
    Posts:
    5,797
    Likes Received:
    308
    Yes - X58 was very intensive to overclock, which is what made it so much fun to tinker with. Then along came Sandy Bridge and all you had to do was set 100x50 in the BIOS and an easy 5GHz fell into your lap.

    Memory overclocking was definitely the sticking point on X58 because it was a new thing to have the memory controller on the CPU die, and coming from P45/X48 it was a steep learning curve. It took me a long time (and a lot of X58 hardware) to settle into a comfort zone with X58, and I've never left it. :)

    I'm also currently running 24GB of RAM (6x4GB) and I knew I was going to need a lot of uncore volts to get it stable running at 1568MHz on a base clock of 196MHz and uncore of 3136MHz. I was running at 1600/3200 for a while, which required 1.38v in BIOS, but I've since taken it down because it's just too hot. The higher end X58 CPUs allow you to run more relaxed dividers between memory and uncore (1:1.5 instead of 1:2) which makes overclocking a lot easier, but I eschew this because of the performance hit to memory bandwidth.

    X58 was never a plug-and-play system, even with XMP - which, in my experience, was a complete waste of time. You're far better off setting the voltages manually. And your POST problems may be motherboard or BIOS related; the motherboard you have is not an enthusiast level product by any stretch and you've done well to get as far as you have!
     
  3. Vault-Tec

    Vault-Tec Green Plastic Watering Can

    Joined:
    30 Aug 2015
    Posts:
    9,136
    Likes Received:
    931
    Intel boards have always fallen short of other manus. That's why they gave up (but by crikey, they did make some lovely looking boards !).

    Whenever I overclocked X58 I would start with the base clock. Lower the multi to the lowest then start cranking on the BCLK to see how high I could run it. After that I just added mutli and volts.

    But Intel boards (and especially their X58 stuff, X79 was better but still 10% below average for OC) can be a real pain.

    Have you looked up your ram timings and set it all manually? without XMP that is the first place I would start :)
     
  4. rollo

    rollo Well-Known Member

    Joined:
    16 May 2008
    Posts:
    7,700
    Likes Received:
    99
    Intel boards are super stable though is one of there big + points.
     
  5. Otis1337

    Otis1337 aka - Ripp3r

    Joined:
    28 Nov 2007
    Posts:
    4,480
    Likes Received:
    117
    yeah intel boards where by far the most rock solid and put though immense testing. they might not be the fastest, but they will out live any other board hands down.

    loving my x58 system still 8 years later! :p
     
  6. dec

    dec [blank space]

    Joined:
    10 Jan 2009
    Posts:
    322
    Likes Received:
    10
    After doing some initial playing around I realized it wasn't going to be as easy as everything post-sandy bridge so I got a notebook and filled a page with combinations of settings that worked and failed, using manufacter info as a reference. Yes those memory timings are manually set.

    It was actually very fun. I spent a lot of time reading up on why the uncore and memory voltages should be near each other as well as keeping the uncore, memory and QPI frequencies all in the correct order. The whole process was like solving a riddle in almost painful contrast to my haswell desktop that was far too easy.

    I'd like to push this system further but I don't want to have to explain to the boss how any part of it died should something go terribly wrong. Hopefully all the office computers get replaced before too long and I get to take this piece home.
     

Share This Page