Jump to content
HWBOT Community Forums

S_A_V

Members
 View Profile  See their activity

  • Posts

    291

  • Joined

 Content Type 

Posts posted by S_A_V

    PCMark05 , keep or remove from hwbot.

    in Benchmark software

    Posted

    SAV: That is good proof of scaling. Do you have any proof of scaling beyond 4 cores? I haven't seen any scaling beyond 4 cores in any subtest.

     

    I don't have direct comparison with more than 4 cores with exact the same settings.

    Just found PCMark04 results with FX-4100 @ 6918 MHz and FX-8120 @ 6782 MHz:

    4-core: 49.3 web page rendering, 317 fps WMV - http://hwbot.org/submission/2231053_

    8-core: 55.2 web page rendering, 362 fps WMV - http://hwbot.org/submission/2240067_

    Don't look at Overall score, FX-4100 result was done with wrong Grammar Check.

     

    Windows was updated with AMD Bulldozer Performance Updates:

    Windows6.1-KB2592546-x64.msu

    Windows6.1-KB2645594-x64.msu

    Windows6.1-KB2646060-v3-x64.msu

     

    I haven't seen scaling past 4 threads testing on FX-8120 in this subtest. I haven't tested results much with other subtests and more than 4 cores.

    Do you use WMV settings suggested in PCMark05 Tweaking Thread?

    I mean this: "Force NumThreads"=dword:00000002

    Try playing with other values ;)

    PCMark05 , keep or remove from hwbot.

    in Benchmark software

    Posted

    There should be a maximum number of cores for certain benchmarks, yes (like for PCMark04 we only need single core and multicore, as it only benefits from two threads). PCMark05 up to 4. Vantage and 7 I don't really know. I bet the number of threads is specified in some white papers.

    Both PCMark04 and PCMark05 uses some OS components that scales from number of cores - web browser, audio-video codecs, etc.

     

    Don't look at "whitepapers", just open process manager, run benchmark and see real processor usage per cores.

     

    The same with 3DMarks03/05: they don't need multicore CPU, but video drivers support multicore CPU, so we have better scores with more cores.

     

    PCMark04 - Web Page Rendering - Phenom II 560:

    2-core - 60.6 - 6346mhz

    3-core - 64.8 - 6300mhz

    4-core - 70.9 - 6346mhz

     

    PCMark04 WMV Video Compression subtest also scales from number of cores on Windows 8 with right settings.

    PCMark05 Tweaking Thread

    in Benchmark software

    Posted

    Can't get working Transparent Windows Bob's Tweak on two different setups :confused:

    Tried many times both ways - by Stelaras and by bob80.

    11k is the best run on air so far:

     

    light36048pcmark05pheno.png

     

    On the video bob says do log off and then log in.

    After "login" I see PCMark05 already (still) running.

    But while logging off windows close all running tasks, including PCMark05.

    Ok, I tried lock / unlock windows and change user / back to logged used, it's not works for me too.

    PCMark05 Tweaking Thread

    in Benchmark software

    Posted

    Damn 34K transparent windows...

    I only know how to run transparent windows without mouse moving and default mouse settings (all my pcmark05 runs done without mouse tweaks), but it gives me only the same scores that others can do with mouse tweaks. Unfortunately my way to tweak transparent windows, when combined with mouse tweaks, gives not better scores than when both method used separately.

    S_A_V - Phenom II X4 960T BE @ 2006MHz - 14sec 749ms wPrime 32m

    in Result Discussions

    Posted

    Nothing different to the 3 cores you talk about, and this is not especially because of your submission or the global points but something in general- no one would have thought about making ranking for 6800LE for example with 9, 10 or 11 pipes unlocked to- this will bring only a flood of new results going on with number of cores in the future. With this we will see 7,9,11,13 core and so on in the future untile bot drowns in this :D - I would have limited boints to count of cores cpu was sold at - again, nothing against you personally ;)

    Now I understand your point:

    CPU configured to 3 cores by manufacturer - exist.

    CPU configured to 5 (7,9,11,13) cores by manufacturer - not exist.

    Yes, that is the difference.

    But if some CPU's not exist by now it not mean they will not released later.

    Anyway, ability to posting 5x-core scores already exist in this revision of hwbot engine, I just use it :)

    S_A_V - Phenom II X4 960T BE @ 2006MHz - 14sec 749ms wPrime 32m

    in Result Discussions

    Posted

    there have been enough 960 which have been unlocked to 6 cores- it doesn´t count as result if you downgrade a gulftown to five cores, so why does this count?^^ - nice idea, but hwbot tricking and searching for weaknesses in rules and testing these wasn´t what was supposed as the bot was made. Just my 5 cents ;)

     

    We already have 3-core ranking filled with results with Phemon II / Athlon II 2-cores by default (4-cores physically) unlocked to 3-core, so why do you think Zosma (4-core by default) unlocked to 5-core is something different?

     

    I understand it is easiest global points ever made and not competitive for now, but anyone can buy 960T for 100$ on ebay and join me in 5-core ranking (both wprime, pcmark, unrar bench, etc).

     

    Sure, I will make some backups with LN2, so it will be not so easy anymore ;)

    PCMark05 Tweaking Thread

    in Benchmark software

    Posted

    I also think the high SSD-based virus scan scores are interesting. Sav just posted a 1200+ virus scan yesterday using 2x SSDs from onboard RAID0. That is higher than the actual disks are capable of, and is done through ramcaching of onboard raid software. Ramcaching software is disallowed in the rules, but these submissions are being allowed - not sure where the line is drawn on ramcaching software. These scores are getting the same virus scan performance of hardware raid cards with onboard RAM, but its done through software.

     

    Write-back cache is standard functionality of many software, comes with both on-board and discrete RAID-controllers. Intel RST, AMD RAIDExpert, Promise WebPAM, etc. It is already used for ages for PCMark scores, way before someone discovered third-party caching software like Fancy Cache or Super Cache.

    There is a big difference between that two types of software with caching ability.

    Third-party software can use so much RAM for cache that you want, and it gives you performance very close to RAM Disk driver (if have enough RAM installed).

    "Bundled" software that comes with hardware not allow you to specify amount of ram for cache and don't have "mirroring partition" feature. You can only enable or disable that option. And if enabled it gives the same boost no matter how much system RAM you use - 4 Gb or 64 Gb.

    PCMark05 Tweaking Thread

    in Benchmark software

    Posted

    I still have to understand many things there..

    for example S_A_V 's general usage or mtech page rendering... and I believe there are still things to discover :)

     

    I still need to improve my page rendering and transparent windows.

    As for general usage on SSD there is no special tweaks.

    Use latest firmware for SSD.

    Use latest Intel RST - not from official Intel site, look at station-drivers for example. See driver build date in .inf-file, not version number.

    Use small stripe size (try 4, 8 or may be 16) and enable write-back cache.

    Run TRIM manually every time before you make RAID with SSD.

    First run after you make RAID always better, because there is still no TRIM support for RAID.

    Every next run without rebuilding and trimming RAID will be worse than previous.

    Don't use RAID partition for XP Startup, if possible, use some other storage (single SSD, RevoDrive, etc).

    Don't write anything on RAID partition.

    Disable creating all temp/system files/folder for RAID partition ($RECYCLE.BIN, System Volume Information, etc.).

    This is only for SSD in RAID, you don't need it for iRAM/Acard.

     

    And the most important part - don't use slow SSD: no SATA2, no Intel, no SandForce, and no less than 120 Gb.

    For now, I can only recommend Crucial m4, Corsair Performance Pro and Plextor M3.

    S_A_V - Core i5 2500K @ 5989MHz - 50226 marks PCMark 2005

    in Result Discussions

    Posted

    Wow great score! Amazing - only 2 SSDs using onboard controller again?

    Yes, 2x SSD onboard SATA3-ports.

    I tried 3x and 4x SSD in SATA2 mode - it gives worst scores than only two in SATA3 mode.

    Also compared 2x Crucial m4 and 2x SandForce2-based (Vertex3 and HyperX) - scores with Crucial m4 are much better, especially with new firmware that gives nice boost in general usage.

    edit - carefull with that SB - she is a gem for sure! I have killed my share with too much cpuv :(

    You used gigabyte board when CPU died?

    S_A_V - Core i5 2500K @ 5996.8MHz - 5min 36sec 672ms SuperPi 32m

    in Result Discussions

    Posted

    try the m4e p67?

    M4E sold, "upgrading" to M4EZ was my big mistake :o

     

    Very nice efficiency!

    You kidding, right? :)

    Your 5:35 at 5950 is very efficient, like it should icon14.gif

    This 5996*336.672=2018k is not, because of high tRCD.

    5:33 should be possible with 6 GHz CPU and 8 Gb hypers.

     

    no waza or just not tightened subs?

    No problems with copywaza on this ram.

    Not all subs tightened - I just set tRFC 72 and some others at random.

×
×
  • Create New...