Jump to content
HWBOT Community Forums

Lucky_n00b

Members
 View Profile  See their activity

  • Posts

    852

  • Joined

  • Last visited

  • Days Won

    11

 Content Type 

Posts posted by Lucky_n00b

    Updating my feeling

    in League Discussions

    Posted

    I usually don't bench purely for points, but looking from some opinions provided here I can understand the frustration of the 3D benchers (which I totally respect, I mainly does 2D as my current job is to test motherboards and RAM, but I'm aiming to be better 3D bencher as well).

     

    A 'difficulty multiplier' as suggested could work, but as Pieter said:

     

    Imho, from the looks of it I think it's a bad idea to measure the difficulty of a benchmark by the amount of people use LN2. We need a different method of defining difficulty.

     

    As an idea, maybe the difficulty multiplier need to take into account:

    - How the benchmark scales to non-hardware tweaks (OS choice/OS Tweaks, Driver Tweaks/adjustment, anything non-hardware related), in this case Pi 32M and 3DMark2001SE should be getting on the high side just because the sheer number of tweaks needed to be done to get it ran right.

    - How the benchmark scales to some hardware parameters. Is the benchmark Really CPU-bound and doesn't really scale much to other components?(e.g like GPUPI & Hwbot Prime shows little scaling to memory compared to Geekbench/XTU). Or how Fire Strike Ultra is really GPU-oriented(when you do 1GPU) and does not scale that much according to cpu, compared to 3d11 1GPU when CPU and RAM also mattered a lot.

    - How many hardwares need to be configured/monitored properly during an all-out bench session. In this case, anything 3-way/4-way is much more difficult compared to doing a single card session.

     

     

    The hard part is, some benchmark's "difficulty' will also depend on the hardware. E.g:

    Benching 3DMark Fire Strike on GT 730 or lower don't need that much attention to CPU(you can leave CPU on air), compared to you benching the same benchmark on GTX 980 Ti where the CPU start to give score scaling because the big GPU Score on GTX 980 Ti need much CPU Score to give an optimal score.

     

    That's what i thought(sorry for the bad English),

    Skylake 6700K/6600K Batch and Serial Numbers

    in Skylake/Kaby Lake/Coffee Lake (-X) (Z170/Z270/Z370/X299) OC

    Posted · Edited by Lucky_n00b

    Hi,

    Tested 10 pcs L525B503 Malaysia retail(and one ES for comparison) with three kind of tests:

    1) Maximum Boot to Windows 7 64-bit, 4C/8T, 1.35V VCore

    2) Maximum Cinebench R15 Frequency, 1.35V VCore (there's Vdroop to around 1.31-1.32 IIRC)

    3) Lowest VCore to run 4.6Ghz Cinebench R15 4C/8T

     

    *Aircooled, Noctua NH-U12S+Noctua NF-F12 iPPC-3000, 25C ambient*

     

    This is the result:

     

    JagatOC_6700K_binning.jpg

     

    All the details about the specs, methods can be found here if you guys interested(need google translate though :P)

     

    Cheers,

    The official MSI Godlike OC Tournament: Ambient 3D Battle thread.

    in HWBOT Competitions

    Posted

    I don't know if everyone knows about this but I came across a forum on Tomshardware the other day and someone gave a useful like for a protection plan from Intel for OCing CPUs

     

    heres the link here http://click.intel.com/tuningplan/

     

    it covers quite a few CPUs for a small fee and one time of course but hopefully it will let people feel a little more comfortable OCings

     

    I used that for some of my retail cpus, making CPU RMA in my place way easier and no questions asked :)

    The official MSI Godlike OC Tournament: Ambient 3D Battle thread.

    in HWBOT Competitions

    Posted

    I'm trying my best to learn on how to improve IGP preformance don't really have any knowledge other then what most electronic stores will tell you about IGP vs dedicated GPUs to get you to spend more if you don't know anything about computers

     

    I'm no IGP expert, but I might have some advice for you :)

    IGP tuning is actually pretty straightforward, maximise its IGP clock, and memory performance matters a LOT(especially in AMD's IGP).

     

    1) For Intel's IGP solution, for example HD 4600 found in i7 Haswell, the frequency of the IGP is found on the 'GT Ratio' on MSI's motherboard, this is a 'multiplier', and you get IGP Frequency from this formula:

     

    IGP Frequency = GT Ratio * (BCLK/2)

    So basically on 100Mhz BCLK, every 1 GT ratio increase will give you 50Mhz more IGP Frequency.

     

    The one voltage that's responsible for the IGP is the GT Voltage (IGP Volt or Vaxg in some mobo). You can really push the IGP volts, 1.35-1.4V is quite safe for this and will give around 1550-1600Mhz IGP frequency or more. Got a better cooling and you can crank this up ;)

     

    2)After finding your maximum GT ratio for 3D benchmarks, time to maximise performance from your memory system, if you can do 2800CL9 with typical samsung timings you'll do fine. IGP OC tend to favor more on frequency instead of timings on most cases :)

     

    3)Lastly, when your IGP GT Ratio and Memory Frequency is maxxed out, time to do some fine tuning with BCLK. Increase 1 - 2 Mhz here will give you that extra 15-20 Mhz you'll probably need for competition purposes.

     

    Oh, and if you're on ambient cooling, sometimes reducing CPU OC can gain just a little bit of IGP frequency. If your benchmark doesn't require high CPU scores, you might try lowering your cpu oc & cpu volts a bit to gain IGP frequency.

    (useful for GPU-bound scenarios,like 3DMark Fire Strike on HD4600 .But on 3DMark Cloud Gate CPU speed is still needed)

     

    Well, hope that helped. Good Luck ;)

    • Thanks 1
×
×
  • Create New...