are there any testing software

Discussion in 'Computer Information' started by Chet, Oct 15, 2009.

  1. Chet

    Chet Guest

    looking to upgrade from my buddys Dell Dimension 4550 computer, there
    are alot of nice used computers around for sale, is there somekind of
    software I can run my his present Dell and when he is shopping around we
    can run it on the computer he is looking and see what the performance
    gains are.

    Chet
    Chet, Oct 15, 2009
    #1
    1. Advertising

  2. Chet

    JD Guest

    Chet wrote:
    > looking to upgrade from my buddys Dell Dimension 4550 computer, there
    > are alot of nice used computers around for sale, is there somekind of
    > software I can run my his present Dell and when he is shopping around we
    > can run it on the computer he is looking and see what the performance
    > gains are.
    >
    > Chet


    Hi there

    you might want to try a program called "Sandra" (made by Sisoft, Google
    "sisoft sandra"), it allows you to run benchmarks on various things such
    as the whole pc or single things like memory, disk speed etc. they used
    to do a free/home version.

    JD
    JD, Oct 15, 2009
    #2
    1. Advertising

  3. Chet

    Paul Guest

    Chet wrote:
    > looking to upgrade from my buddys Dell Dimension 4550 computer, there
    > are alot of nice used computers around for sale, is there somekind of
    > software I can run my his present Dell and when he is shopping around we
    > can run it on the computer he is looking and see what the performance
    > gains are.
    >
    > Chet


    It depends on what kind of gains you're looking for. And also,
    what kind of operating system you expect to find on the machine
    that is for sale.

    Processors can have single cores or multiple cores. A lot of software,
    uses one core at a time, and in such a situation, the clock rate of the
    processor helps. The benchmark that corresponds to "software running
    on a single core" is SuperPI.

    http://www.xtremesystems.com/pi (original home, down right now)

    http://www.techpowerup.com/downloads/366/ (alternate download location)

    SuperPI can be run with various numbers of digits. Popular values might
    be 1M or 32M, standing for 1 million digits or 32 million digits. A
    1 million digit run, would have been suitable for the Dimension 4550,
    because processors with small caches, cannot hold the entire benchmark
    in cache. But on newer systems (say a Core2 Quad with 12 megabyte cache),
    there is a possibility the benchmark will fit entirely in cache.
    Enthusiasts use a larger benchmark run, to prevent that from happening.
    If you use 1 million digits for all your runs, then a Core2 Quad with
    12MB cache would get an unfair advantage, and would benchmark faster than
    the speedup you might see for other software. A 1 million digit run takes
    about 8 megabytes of system memory to hold the test. A 32 million digit
    run uses a lot more system memory, so it can't fit in the cache. A 32 million
    digit run takes too long, if you're standing around in someone's house.
    So it is a bit of a problem. It could standardize on some intermediate
    value, like 2M or 4M.

    To rate multicore performance, the simplest way would be to multiply the
    SuperPI result, by the number of cores. That is roughly correct. Because
    the Core2 Quad "chokes a bit on its FSB", a quad core Core2 is about 3.5x
    faster when its four cores are all being used. So the "scaleup" isn't perfect.
    But multiplying by 3.5x may be simpler than trying to find a multicore
    benchmark you can believe in.

    SuperPI mainly is interested in CPU and memory, and if it doesn't fit in
    the processor cache, then the memory gets tested and influences the results.
    There are many other benchmarks around, which test more of the system,
    including elements of the hard drive. But in my opinion, that tend to cloud
    the issue, and doesn't really tell you what part of the system sucks and
    what part of the system is a superstar. If a benchmark makes too much use
    of the hard drive, then all the computers will look like "slugs".

    This benchmark is for graphics. This particular version is rather old, but
    will make any video card look like a winner. 3Dmark 2001 SE (Build 330).
    My main reason for picking this, is it is a relatively small download at
    40MB. But for your purposes, you could pick one of the later versions that
    is hundreds of megabytes, and put it on a CD. Another difference is,
    this one has to be installed, which might not be the easiest thing to
    do when visiting people with used computers. SuperPI on the other hand,
    simply runs, without installation.

    http://majorgeeks.com/3Dmark_d99.html

    When looking at used machines, you also have the option of doing your
    homework in advance, making your own "cheat sheet" based on already
    available benchmarks for various processor families or types, various
    graphics cards and so on. That might be a viable alternative if there
    was some consistency in the benchmarking world, but that is harder than
    it seems.

    This site contains at least half a million benchmark runs, which is good.
    The bad part, is the interface on the web site. You can figure it out
    after a while, but even when you learn how it works, it doesn't always
    give the information when you want it. For example, I get SuperPI 1M and
    32M results from this site, but it doesn't always have results for
    a given processor at stock speed. And when it does have stock speed
    results, it won't give them to you without a fight. I have to manually
    edit the URL on some of the web pages, to get what I want.

    http://www.hwbot.org/

    To give examples of what kind of speedup to expect, my previous generation
    of computers were P4 ~3GHz and AthlonXP 3200+. They would complete
    "SuperPI 1M" in 45 to 50 seconds. The slower run was on the machine with
    antivirus software running in the background. This is my new machine
    with a Core2 Duo running at 2.6GHz. You can see that FSB (interface on
    processor) and memory configuration, make some difference, but the
    core clock speed is what really helps. The "worlds record" is somewhere
    below 10 seconds for this, but not at stock speed. That would be on
    an overclocked system. Lower times are better.

    -------------------------------------------------------SuperPI Memtest 1.65
    -------------------------------------------------------1M (sec) Bandwidth
    -------------------------------------------------------lower is (MB/sec)
    ------------------------------------------------------- better

    200 x 13 = 2.60GHz, FSB800, DDR2-533, Single channel 24.05 2203
    (stock) Dual channel 22.87 2668

    266 x 10 = 2.66GHz, FSB1066, DDR2-533, Single channel 23.47 ----
    Dual channel 22.52 ----

    266 x 13 = 3.46GHz, FSB1066, DDR2-533, Single channel 19.37 2419
    Dual channel 18.42 3305

    The memtest in that example, comes from here. It tests memory integrity.
    Test 5 is probably the most important one to test, as more failures
    are seen there. This happens to have a "bandwidth readout", which I use
    for quick comparisons when fooling with computer settings in the BIOS.
    But the "bandwidth readout" is not accurate when comparing different
    computer architectures, as some architectures cheat and have prefetch
    schemes. Sisoftware Sandra recent versions, do a better job of testing
    memory bandwidth. But wasting time testing memory isn't really of as much
    value, as you can see how much effect it has above.

    http://www.memtest.org/

    Before buying a machine, I might want to see this run for 10 minutes, to
    see if the computer is stable. Prime95 uses all cores. It will make the
    processor get warm. It does a computation with a known answer, which is
    how it detects errors. But to be really acceptable, I'd want a four hour
    run with no errors at all, and if you're visiting someone's house,
    you can't very well wait that long. This program has "stress testing" and
    also has the ability to work on a project, and in this case, hardware
    people use the "I'm just stress testing" feature. You don't have to
    "join GIMPs" to use it. Their objective, is to search for Mersenne Prime
    numbers, a branch of mathematics.

    http://www.mersenne.org/freesoft/

    A computer should be able to complete one benchmark run of 3DMark2001SE
    without crashing. It should be able to complete at least ten minutes of
    Prime95. So perhaps in about half an hour, you could squeeze in a 60 second
    SuperPI 1M, maybe 10 minutes for 3DMark2001, then 10 minutes for Prime95.
    You'd do the first two, to see if you were interested in buying the
    computer, and the third one before taking your purchase home. If a
    computer isn't stable, that might hint at why the seller is getting
    rid of it. Ten minutes with Prime95 is a better test of stability,
    than staring at an idle computer for 100 hours, to give you some
    idea how a seller could "cheat". Many unstable computers look fine,
    when they're idle.

    HTH,
    Paul
    Paul, Oct 16, 2009
    #3
    1. Advertising

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

It takes just 2 minutes to sign up (and it's free!). Just click the sign up button to choose a username and then you can ask your own questions on the forum.
Similar Threads
  1. Steelman
    Replies:
    12
    Views:
    3,215
    ┬░Mike┬░
    Dec 6, 2004
  2. Replies:
    3
    Views:
    391
    Pavel A.
    May 12, 2006
  3. Networking Student
    Replies:
    4
    Views:
    1,295
    vreyesii
    Nov 16, 2006
  4. tttrainer
    Replies:
    0
    Views:
    1,025
    tttrainer
    Nov 3, 2008
  5. priyanka.madhu97
    Replies:
    0
    Views:
    1,210
    priyanka.madhu97
    Dec 30, 2008
Loading...

Share This Page