1. This forum section is a read-only archive which contains old newsgroup posts. If you wish to post a query, please do so in one of our main forum sections (here). This way you will get a faster, better response from the members on Motherboard Point.

Disappointing upgrade: Sempron 2200+ --> Phenom II 4X 840

Discussion in 'AMD Overclocking' started by DK, Jan 5, 2012.

  1. DK

    DK Guest

    Wanting more raw computing power, I went from Sempron 2200+
    in ASUS Asus A7V400-MX (socket A) to Phenom II 4X 840 in Asus
    M4N68T-M V2 (socket AM3).

    I must say I am quite disappointed. I was hoping for about 10X speed
    boost even with applications that can only use a single core. After all,
    more than six years separate the two (expected boost suggested
    by Moore's law would be 64X). The real life tests show nothing
    of this sort! Not even close to 10X.

    DIVX encoding using VirtualDub and DivX 5 codec became
    faster by only about 3X. OK, this may also have something to do
    with the video card change (does it? - I don't really know; went
    from dedicated nVIDIA Quadro FX 3000, 256 MB to integrated
    nVIDIA GeForce 7025, 512 MB - not sure what is supposed to be
    better; I need occasional hardware stereo so Quadro will be
    going back into the slot).

    To make sure it's not GPU, I also tested using a program that does
    no graphics (multiparametric fitting anf FFT; essentially raw numbers
    crunching). Alas, a fairly common task that took 23 min on the
    6.5 years old cheap system now took 9 min - only ~2.5X faster.

    Granted, programs that can use SMP run correspondingly faster
    (about 3.3X over just using a single core) but still... Are my results
    observations fairly typical? If not, what gives? This $200 upgrade
    pales in comparison with the $160 upgrade 6.5 years ago when I
    went from dual Celerons 366 in ABit BP6 to the Sempron 2200+.
    There, the speed boost was of near cosmic proportions.

    Any comments? Thanks!

    - Dima
    DK, Jan 5, 2012
    1. Advertisements

  2. DK

    Paul Guest

    While DivX encoding is running, arrange a copy of CPU-Z running
    at the same time, and verify the CPU clock value is correct.

    (A "no-install" version is fine...)

    Using Cool N' Quiet, the processor will switch to the low multiplier
    when the system is quiet. When busy, the processor should be
    running at the high multiplier. And this should be reflected
    in the core clock speed value listed in the main panel of CPUZ.

    You can try disabling Cool N' Quiet, which should leave the
    processor at the high multiplier.

    Depending on the OS used, a "cpu driver" may be required.
    I'm not sure where to get the best information on that now.

    This CPU driver is for WinXP. I would expect Vista or Win7 to
    already have this driver as a hidden built-in. It took a
    number of steps on the AMD driver download page to find this.
    It wasn't that easy.



    3.2GHz quad core L1=128Kx4 L2=512Kx4 L3=missing 95W


    Performance ratio (as via this benchmark)


    AMD Phenom II X4 840 3,735 (Passmark)


    AMD Sempron 2200+ 336 (Passmark)

    So according to that, the ratio is supposed to be 10x.
    And for that to be true, the benchmark would have to
    be able to run on multiple cores.

    Paul, Jan 5, 2012
    1. Advertisements

  3. DK

    Wes Newell Guest

    Do the math. 366MHz celery to what, 1800MHz? Sempron 2200+ to what~
    2500MHz X4 Phenom. And core speed is only one factor of many in system
    speed. That said, check your cool n quite settings. It's possible your OS
    is not set to run the cpu at max speed.
    Wes Newell, Jan 5, 2012
  4. DK

    DK Guest

    Thanks, Paul!

    I did install something called "CPU driver" off Asus DVD. Will go back
    and see what happens if I don't install it or install an updated version.
    Yeah, that's the numbers I was looking at when planning an update.
    Sounds like this is exactly what I would see if I get my 2.5-3X boost
    for a single core and multiply it by 3-4X. But then, at work
    I have Core i7 950 and its Passmark is value 6,365. I just tested
    DivX encoding on it using the same test as done for my Phenom
    home system. At home I was getting ~ 40 FPS (versus around 15 with
    the Sempron), while at work it is around 125 FPS - considerably
    more than what the 1.7X difference in Passmark scores would

    As the other reply points out, I guess I simply did not appreciate the
    fact that CPU clock frequency in modern processors seems to have hit
    a wall and that the main inroads into increasing computing speed is
    now in SMP scaling.

    I also haven't realized that Cool'n'Quiet can get in the way of things -
    will investigate this further. Does it work on a per core basis or does
    it look at the whole chip?

    - Dima
    DK, Jan 5, 2012
  5. DK

    Paul Guest

    As far as I know, it applies to the whole chip. You never know
    though, they could change that. It just makes it more of a pain
    for the CPU cores to use shared resources, if done that way.

    In this article, it implies application to the whole chip.


    Paul, Jan 5, 2012
  6. DK

    Yousuf Khan Guest

    The CPU driver usually is there to take advantage of newer power-saving
    modes, and stuff. Even with a generic CPU driver (afterall, these are
    all just x86 processors, no matter what), I doubt that performance will
    be affected all that much. Power consumption on the other hand might
    become significantly better though.

    I wouldn't base too much importance on synthetic benchmarks. Synthetic
    benchmarks typically measure individual component performances to the
    exclusion of other components, such as RAM speed vs. CPU speed vs. I/O
    speed, etc. Real-world benchmarks involve real applications, and real
    applications are a combination of CPU/RAM/IO. Even if the CPU part is
    10x faster, doesn't mean that the RAM is, and certainly not the disk.
    The DivX codec uses an Intel compiler that's purpose built to exclude
    non-Intel hardware. It does this by turning off most modern instruction
    sets such as SSE on non-Intel processors, even if the non-Intel
    processors have those instruction sets available. It was the subject of
    an anti-trust suit between Intel and AMD, which Intel lost and was
    forced to pay AMD $1.2B. Intel used the compilers to gain unfair
    advantages in benchmarking suites.

    Divx using Intel compiler? - AnandTech Forums
    I think CnQ and other clock-throttling technologies are there to save
    power, but they dynamically scale speed of the processors to the
    workload needed. I can't see CnQ would interfere with ultimate
    performance by more than a few percent (I'm guessing less than 5%,
    probably less than 3%).

    Yousuf Khan
    Yousuf Khan, Jan 5, 2012
  7. DK

    Paul Guest

    I've run into a couple cases of "stuck multiplier" problems in
    the newsgroups. The sad thing in one case, was the problem
    resolved itself, and the OP couldn't say exactly what
    step fixed it (changing too many things at the same time,
    to be sure). So it does happen, that a user finds
    their CPU stays at the low multiplier, and they don't
    know why.

    You can achieve that result artificially, by using a power
    schema. For example, the RMClock utility, will install it's
    own power schema (that's in addition to the ones already
    there like "Always On" option). If a user cranked the
    multiplier values down in their RMClock schema and left
    the RMClock schema selected in the Power Options control panel,
    that could give a "stuck multiplier" effect. I occasionally
    run into people, who get a machine from a relative, which
    is hacked to bits, and full of surprises. Like the
    guy with a funny multiplier problem, which was resolved
    when he pulled the Athlon processor, and found a jumper
    wire stuffed into the socket holes (wire mod). I didn't
    even suggest he look there, but he managed to find it.

    So there are ways to mess it up, given enough effort. It
    just depends on how clever the previous owner was.

    Paul, Jan 5, 2012
  8. DK

    SC Tom Guest

    I was following this since I upgraded from an Athlon 64 x2 4800+ (2.5GHz) on an Asus M2NPV-VM MB (4GB RAM) to a Phenom
    II x2 555 Black Edition (3.2 GHz) on a M4N68T-M V2 (4GB RAM). I have noticed quite a difference in speed, and when I
    used the Asus unlocker to open two more cores, I noticed a real difference in most applications I use (some, no change
    at all, but that's mostly older small-footprint apps that weren't too slow anyhow). Windows now has it listed as a
    Phenom II x4 B55. In Device Manger, open each core and update the driver. Let it go to the MS site to get it. The link
    that Paul provided points to a driver for XP SP2, and is dated 2009. I'm afraid that driver would be counterproductive
    on a XP SP3 system with a processor manufactured after 2009. I may be wrong, but I got better results with the MS driver
    (which I normally don't use for driver updates). The "CPU driver" on the MB DVD is nothing more than Cool 'n Quiet. If
    you run the setup again, it will ask if you're sure you want to uninstall it? I left it on mine. I don't find that it
    interferes with what I do.

    My Passmarks are: B55 3,857 (left as a stock x2 555 = 1,993)
    Athlon 64 4800+ 1,312
    I have no way to scientifically prove it, but the system runs a lot faster than the 3-times difference in Passmark

    I haven't done any DIVX editing, but I've done a few home movies and movie editing with Microsoft Movie Maker, and found
    that processing and saving the movies is at least 5 or 6x faster than with my old MB/CPU combo. Movies that used to take
    3/4 to an hour to process are now done in 5-10 minutes.
    Part of your problem may be the onboard video. I found it to be clunky and slow, and I took my Nvidia GT 240 out of my
    old system and use it with this one. Tremendous improvement in hard-core video-intensive games like Modern Warfare
    (which was unplayable on the onboard 7025) , and in overall PC speed.

    Some other things to look at are your RAM timing speeds, whether they're set up as ganged or unganged, etc. You can find
    your current settings using CPU-Z and go from there.

    Overall, I am very satisfied with this upgrade. Plus the fact that I got the MB, CPU and 4GB of G-Skill DDR3-1333 Ripjaw
    RAM for $170 including shipping (Black Friday sale from Newegg).
    SC Tom, Jan 5, 2012
  9. DK

    Paul Guest


    There was an extremely long thread on this topic a while back,
    but I haven't been tracking it. The Version 4 update is mentioned here.


    You have to go to the end of the thread, to get the latest experiences
    with that stuff. I don't know if the original poster in that thread,
    continues to update the first post with the "answer" or not (last update
    stamp says 2007).

    It could be the current AMD driver, checks the update status of the
    OS before installing anything, so just about any behavior in 2011 is
    possible. It really depends on whether AMD felt like deferring to
    Microsoft or not. AMD is always playing cleanup it seems (like
    getting Bulldozer support - patch listed by Microsoft, then

    Paul, Jan 5, 2012
  10. DK

    Flasherly Guest

    It's a difficult job, programming for multiple cores, not to mention
    BIOS and MB implementation between features and microprocessor model
    enhancements constantly being added to multiple cores. When a chip
    design is capable of automatically routing a linearly constructed
    program, written and compiled on a single core, through a networked
    die of multiple cores at x64 a factor of speed, either we'll be ready
    and can sign up for a brain-implanted chip for vastly improved AI-
    augmented intelligence, or begin to cover our asses when the first
    Terminators are released.

    While watching a PBS show on DNA, one of the researchers noted for her
    work in statistical reduction sets over intersecting gnome patterns in
    coupling the animal world generally to humankind, mentioned she is as
    well a computer programmer. Her claim then being, is she wrote a
    program, on camera for such as might a you academic be prone to say -
    "I got geeky" [to accomplish the above task] - while presumably on and
    using university facilities, at a factor of one a single afternoon
    connotes, to do what she claims an average desktop would accomplish in
    35 years!

    "There are liars, there are damn liars, and then there are the
    statisticians." -anon.
    Flasherly, Jan 5, 2012
  11. DK

    SC Tom Guest

    As usual, you go above and beyond in your research, and I for one appreciate it.

    I read a few of the 64 pages of the article, including the version 4 patch and the other articles toward the end of the
    article. They all mention SP2, including the version 4 patch. Also intimated in the article is that it was all included
    in SP3, and that everything was taken care of with it also. Don't know about that since some people still have
    multi-core problems.

    Also, since the article seems to deal more with dual core than with 3 or more, I don't know how much of it would be
    appropriate to 3+ cores. Some of the optimizing tips might be useful; if I get brave, I may try some of them. Kinda
    scary playing with the boot.ini. If I do decide to mess around with it, I'll be sure to have a new image of my drive
    before starting on it :)

    I did find this troubleshooting guide:


    And even though the AMD driver file from your earlier link says "Updated 2009", the setup file inside the zip is dated
    9/7/2007. I think I'll leave mine alone for now; it's working fine for me :)
    SC Tom, Jan 5, 2012
  12. DK

    Bob F Guest

    Is your hard drive 10x as fast?
    Bob F, Jan 6, 2012
  13. DK

    Paul Guest

    Benchmarking, after a hardware upgrade, is all part of the fun.

    You want to verify your system is tuned up properly, and also
    that you got your money's worth.

    All it takes is a benchmark which isn't disk limited, and you
    can get some idea how much the new gear is helping.

    An example is Cinebench, a synthetic benchmark that scales
    perfectly (so doesn't really stress much except the processor
    cores). That's better than nothing.


    Another test you can use, is 7ZIP in ultra mode compression. My
    puny machine can only do about 5MB/sec compression, which means
    the hard drive isn't stressed at all. 7ZIP or WinRAR testing
    is good for including memory subsystem performance as a component
    of the test. The Phenom without an L3 may be marginally slower
    than a Phenom with 6MB L3 (at the same clock speed). In many other
    situations, the missing L3 isn't a factor. While those benchmarks
    are running, you want to pop open Task Manager and make sure the
    graphs are well filled (in case your test case is only using one core
    by accident).

    Paul, Jan 6, 2012
  14. DK

    DK Guest

    Just tried 7ZIP ultra compression. The version I have seems to only use
    a single core for the ~1.3 MB/sec on average (Task Manager shows no
    more than 30% CPU utilization and that is with email, browser and
    newsreader). Is your 5 Mb/sec for a single core as well? If yes, what
    CPU do you have?

    DK, Jan 6, 2012
  15. DK

    DK Guest

    No, absolutely not, I kept my PATA drives. But the tests I ran are
    not in any way limited by disk access.

    DK, Jan 6, 2012
  16. DK

    Paul Guest

    To do a fair comparative test, we'd have to start with
    the same data file. My processor for the test was
    the E8400 dual core.

    7ZIP is a bit picky about using multiple cores. In some cases,
    the interface claims it will use multiple cores, when at runtime
    it is obviously not doing so. That's why it is important to
    use Task Manager, to make sure it's working.

    The slower compression options, like Ultra, are the ones that
    may trigger multi-core operation.

    If you wanted another benchmark, there is "pigz" for Linux,
    which is a multithreaded GZIP. If I need to compress entire
    disk drive images, that's my tool of choice. It's a good
    tradeoff between compression speed, and amount of compression
    achieved. I don't know if there is a Windows port. In Linux,
    you get this from your package manager and install it there.


    OK, there's a Windows port here, just found it. I won't
    be testing this until later.




    To have a fair test, we can create files with similar entropy.

    Download a copy of "dd".



    Unzip the file. Open an MSDOS (command prompt) window and "cd"
    (change directories) to the directory holding the dd.exe file.
    Then try

    dd if=/dev/random of=test.bin bs=1048576 count=1024

    That will create a 1GB file of pseudo-random numbers. Since the seed used
    isn't likely to be the same, we won't have exactly the same file.
    On my machine, it takes 24 seconds to make the test file.

    To test the seed, I created two files with the tool, and the
    checksums don't match. So it uses a new seed for each file.

    Now, for the test. (If I wanted to do this right, I'd compute
    the md5sum of a 7GB movie.iso file, to flush the file cache. But
    that's not necessary as this test is CPU bound.)

    I did an "Add to archive" in 7ZIP version 4.60 beta (from 2008).

    Archive format 7z
    Compression level Ultra
    Compression method LZMA
    Dictionary size 64MB
    Word size 64
    Solid Block size 4GB
    Number of CPU threads 2 (E8400 3GHz Dual Core 6MB L2 cache)

    The test uses about 700MB of memory (of 4GB of DDR2-800 CAS5 memory).

    Basically, the file is incompressible (high entropy), and the
    output file "test.7z" is bigger than the original. The
    1048576 KB input file "test.bin", becomes 1062758 KB on output as "test.7z".

    The test completes on my machine in 7 minutes 13 seconds.
    1,073,741,824 bytes in 433 seconds is 2,479,773 bytes/sec
    compression rate. During the test, the CPU runs at only
    75%, meaning there is 25% of the dual core not used. (And that
    means this test does not scale well, apparently.)

    You should be able to beat me, with a quad core, by a bit.
    Assuming 7ZIP scales to four threads. I don't know if it
    does or not. If you see "Number of CPU threads 4",
    the contest should be interesting.


    I'm going to run it again now with one core instead of two.
    I'm seeing the Task Manager at 50%, which means running with
    one thread is much more efficient for this version of software.
    Two threads should have run at 100%, but only manage 75%, which
    means the threads aren't independent of one another.

    That wouldn't happen with the Cinebench CPU test, which scales
    perfectly, and doesn't really measure bottlenecks in the processor
    quite as well. 7ZIP or WinRAR put more pressure on the cache levels.

    Archive format 7z
    Compression level Ultra
    Compression method LZMA
    Dictionary size 64MB
    Word size 64
    Solid Block size 4GB
    Number of CPU threads 1 (E8400 3GHz Dual Core 6MB L2 cache)

    In this case, I feel I should be able to beat a single core on
    your processor. My time on this one is 11 minutes 17 seconds.
    1,073,741,824 bytes in 677 seconds is 1,586,029 bytes/sec

    Paul, Jan 6, 2012
  17. DK

    SC Tom Guest

    Mine has the 6MB L3, and the 7Zip score is 8.6MB/sec. using 4/4 CPU threads. I watch all 4 cores in Task Manager and
    they are like synchronized swimmers at the Olympics.
    I love the decompression rate of 135MB/sec.
    SC Tom, Jan 6, 2012
  18. DK

    DK Guest

    I had version 4.65 installed but just to be sure downloaded and installed
    version 4.60 beta. And for some strange reason I do not see quite the
    same GUI options as you list:
    Only option is "Create solid archive" - I checked it.
    Only option I see is "multi-threading" - I checked it.
    Using 670 Mb here.
    Alas, I am not even close, confirming my first impression that
    this system is considerably slower than it "should" be:

    The test took 18 min 36 sec with the overall CPU usage between
    30 and 33%.

    Without multi-threading, CPU usage is 25% and it takes
    correspoindingly slightly longer (canceled before it finished).

    CPUZ show stock bus speed (200) and multiplier (16), so
    nothing fishy in the CPU setup. Same for memory - no OC,
    expected frequency.

    Hmm, what going on?

    - Dima
    DK, Jan 6, 2012
  19. DK

    Paul Guest

    That motherboard seems a bit weird. Check out this suggestion.



    If that doesn't help, I'd try setting a few BIOS things to manual. I
    have a couple Asus motherboards, where "manual" exposes the
    most settings, and some of those settings can be interesting.

    For example, I'd set "CPU Overclock" to [Manual] rather than [Auto].
    It may allow you to observe clock frequencies, while in the BIOS.
    And check if something isn't right.

    Your memory "DCT Unganged Mode" is set to [Always] by default, which
    is best for a processor like a quad core. I don't even see an option
    to disable it, which might be beneficial if you were using
    a single core processor. Maybe there aren't any single core
    processors that work in the motherboard ?

    Do you have two sticks of RAM, or only one ? Using one stick,
    might drag down the multithreaded efficiency.

    I was toying with the idea of suggesting a "cache and memory" type test,
    which gives bandwidth curves. But the problem is finding a program
    that does a nice job. My copy of SiSoftware Sandra didn't do a good
    job. It took me forever to figure out how to get the displayed
    results to appear (copy of Sandra from 2009).

    Paul, Jan 6, 2012
  20. DK

    Rodney Pont Guest

    Memtest86+ shows the speed of the memory and the three levels of cache
    (if present).
    Rodney Pont, Jan 6, 2012
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.