Opteron question

Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

I am contemplating the purchase of a dual processor opteron WINXP
machine. The motivation is that currently several of my programs bring
my processor (single procsessor machine) to its knees. So, if I get a
dual processor machine my idea is that the program that currently brings
my one processor machine to its knees will only bring ONE of the TWO
processors to its knees and leave one processor available to do other
tasks. I am assuming here that none of the software is specifically
targeted (written) to use multi processors.

Also, I have checked-into why my one processor is on its knees (100%
utilized) and it is not because I am running out of RAM, got plenty left.

Anyhow, with WINXP how will this dual opteron machine partition all the
processor power between programs?

Thanks

Paul
9 answers Last reply
More about opteron question
  1. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    > I am contemplating the purchase of a dual processor opteron WINXP
    > machine. The motivation is that currently several of my programs bring
    > my processor (single procsessor machine) to its knees. So, if I get a
    > dual processor machine my idea is that the program that currently brings
    > my one processor machine to its knees will only bring ONE of the TWO
    > processors to its knees and leave one processor available to do other
    > tasks. I am assuming here that none of the software is specifically
    > targeted (written) to use multi processors.
    >
    > Also, I have checked-into why my one processor is on its knees (100%
    > utilized) and it is not because I am running out of RAM, got plenty left.
    >
    > Anyhow, with WINXP how will this dual opteron machine partition all the
    > processor power between programs?

    One of the largest benefits of dual-procs in a desktop is the increased
    responsiveness. On one hand, with a decent OS, you're only doing half of
    the context-changes, you can handle twice as many interrupts, and the APIC
    interrupt controller does things more intelligently than the older PIC
    controllers.

    On my machine and other machines around the office, we have a few apps
    that will make a single-proc machine of any speed pretty unusable while
    they're running. As long as the apps aren't doing something like using
    twice the memory that the machine actually has, dual-CPU machines are always
    MUCH more usable when running those apps. The actual throughput of one
    individual app probably won't go up, it may even go DOWN a little - but
    you're still able to use the machine while they're running.

    In fact, right now I'm on a dual Athlon MP 2600, and at home I have an
    AthlonXP 3200. For any single app, it will run faster on the single-proc
    machine: But I have a number of apps which, when run on the single-proc
    machine, bog it down enough that I can't do anything else while they're
    running. On the dually, there are VERY few apps that will have the same
    effect.

    In fact, some time ago I was able to overload the power supply enough to
    cause instability. Both CPUs were pegged at 100% by various processes and a
    SETI to help out, I was ripping a DVD to a hard drive, copying files from
    one hard drive to another, and burning another DVD from files over the
    network - all at the same time. During that, the machine was nicely
    responsive and usable, at least until the power supply gave out. = )

    I regularly use some FAST single-, dual-, and quad-CPU machines, and no
    matter how fast the single-CPU machine is, it never feels as responsive
    under load as the multi-proc machines: But for a single app, a fast
    single-CPU machine will get the job done as fast or faster, and for a lot
    less money. For the cost of a reasonable dual-CPU setup, you can easily set
    up two single-proc machines!

    In fact, some time ago a person asked about buying a dual Xeon system for
    running DVD shrink on a good number of DVDs at the same time. With some
    actual prices, I pointed out that for the same cost, he could build five(!)
    single-proc stations. = )

    steve
  2. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    Do your other programs specifically support dual cpu's? If not then you
    won't see any benefit. What cpu,mobo,ram do you have now?
    "Paul S." <pauls@nospam.com> wrote in message
    news:Nv-dnQgq4Y2RZ7jfRVn-3Q@seanet.com...
    > I am contemplating the purchase of a dual processor opteron WINXP
    > machine. The motivation is that currently several of my programs bring
    > my processor (single procsessor machine) to its knees. So, if I get a
    > dual processor machine my idea is that the program that currently brings
    > my one processor machine to its knees will only bring ONE of the TWO
    > processors to its knees and leave one processor available to do other
    > tasks. I am assuming here that none of the software is specifically
    > targeted (written) to use multi processors.
    >
    > Also, I have checked-into why my one processor is on its knees (100%
    > utilized) and it is not because I am running out of RAM, got plenty left.
    >
    > Anyhow, with WINXP how will this dual opteron machine partition all the
    > processor power between programs?
    >
    > Thanks
    >
    > Paul
    >
  3. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    Hi Dawg,
    Well, Let's see, I have the following hardware currently:

    Pentium 4 2.53GHz
    Motherboard: Intel D850EMV2
    RAM: 1Gig PC800, FSB = 132.6MHz * 4 = 530.3MHz
    *note: my system uses RAMBUS RAM


    None of my programs can specifically support 2 CPU's. My motivation for
    getting a two processor machine is that when my simulation program
    gobbles-up all the resources on one CPU I'll still have one other CPU
    left to get other work done with. At least, that is what I am hoping in
    terms of how it would work

    I have noticed that I am never even close to running out of RAM (in
    terms of what happens when my simultion software bogs down the machine),
    but, if I use the cntrl-alt-delete "performance monitor" that the
    processor is flat-out, but, the ram is only maybe 30% used. So, I think
    the processor is the bottle-neck.

    So, my question is whether having a two CPU machine, with software NOT
    specifally targeted to a multi-processor machine will allow me to
    essentially load down one processor while still having another to get
    other tasks done with.

    What do you think?

    Thanks!

    Paul


    dawg wrote:
    > Do your other programs specifically support dual cpu's? If not then you
    > won't see any benefit. What cpu,mobo,ram do you have now?
    > "Paul S." <pauls@nospam.com> wrote in message
    > news:Nv-dnQgq4Y2RZ7jfRVn-3Q@seanet.com...
    >
    >>I am contemplating the purchase of a dual processor opteron WINXP
    >>machine. The motivation is that currently several of my programs bring
    >>my processor (single procsessor machine) to its knees. So, if I get a
    >>dual processor machine my idea is that the program that currently brings
    >>my one processor machine to its knees will only bring ONE of the TWO
    >>processors to its knees and leave one processor available to do other
    >>tasks. I am assuming here that none of the software is specifically
    >>targeted (written) to use multi processors.
    >>
    >>Also, I have checked-into why my one processor is on its knees (100%
    >>utilized) and it is not because I am running out of RAM, got plenty left.
    >>
    >>Anyhow, with WINXP how will this dual opteron machine partition all the
    >>processor power between programs?
    >>
    >>Thanks
    >>
    >>Paul
    >>
    >
    >
    >
  4. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    Paul S. wrote:
    > I am contemplating the purchase of a dual processor opteron WINXP
    > machine. The motivation is that currently several of my programs bring
    > my processor (single procsessor machine) to its knees.
    [...]
    > Anyhow, with WINXP how will this dual opteron machine partition all
    > the processor power between programs?

    Each process will spend about 50% of the time on one CPU and 50% of the time
    on the other, unless you tell it otherwise. I'm not 100% sure what happens
    on a NUMA system (I think SP2 added NUMA support?), but it should leave a
    process on one CPU the whole time.

    While a dual-opteron setup would certainly solve you problem, you might also
    want to try lowering the priority of the CPU-intensive process, which should
    increase the responsivenes of the single-processor setup.

    --
    Michael Brown
    www.emboss.co.nz : OOS/RSI software and more :)
    Add michael@ to emboss.co.nz - My inbox is always open
  5. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    Michael Brown wrote:
    > Paul S. wrote:
    >
    >>I am contemplating the purchase of a dual processor opteron WINXP
    >>machine. The motivation is that currently several of my programs bring
    >>my processor (single procsessor machine) to its knees.
    >
    > [...]
    >
    >>Anyhow, with WINXP how will this dual opteron machine partition all
    >>the processor power between programs?
    >
    >
    > Each process will spend about 50% of the time on one CPU and 50% of the time
    > on the other, unless you tell it otherwise. I'm not 100% sure what happens
    > on a NUMA system (I think SP2 added NUMA support?), but it should leave a
    > process on one CPU the whole time.
    >
    > While a dual-opteron setup would certainly solve you problem, you might also
    > want to try lowering the priority of the CPU-intensive process, which should
    > increase the responsivenes of the single-processor setup.
    >

    Hi Michael,
    Thanks for the info. What is a NUMA system?
    And, is there a "way to tell it otherwise" in XP pro SP 2.0?


    Thanks

    Paul
  6. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    Paul S. wrote:
    > Michael Brown wrote:
    [...]
    >> Each process will spend about 50% of the time on one CPU and 50% of
    >> the time on the other, unless you tell it otherwise. I'm not 100%
    >> sure what happens on a NUMA system (I think SP2 added NUMA
    >> support?), but it should leave a process on one CPU the whole time.
    [...]
    > Thanks for the info. What is a NUMA system?

    Non Uniform Memory Access. It's what you've got got a dual-Opteron system.
    In a good dual-Opteron system, each CPU has it's own memory. To access the
    memory hanging off the CPU1, CPU0 has to ask CPU1 through the HT to give it
    the data (and vice versa for CPU1 accessing for CPU0's memory). Doing this
    decreases performance: memory off the "other" CPU is slower than memory off
    "this" CPU. Hence why it's called non-uniform.

    If the OS is non-NUMA aware, then it randomly allocates an application's
    memory from both CPUs. So on average, the program will be accessing local
    memory half the time and non-local memory the other half of the time. A
    NUMA-aware OS will allocate all the application's memory off one CPU, and
    keep that CPU locked onto that processor (otherwise it will again be
    spending time accessing non-local memory). This increases the performance of
    the system compared to a non-NUMA-aware OS.

    > And, is there a "way to tell it otherwise" in XP pro SP 2.0?

    Yup. Open up task manager, right-click on the process and click "set
    affinity". Make sure only one CPU is checked. Again, I'm not sure what
    happens on a NUMA-aware OS. If you set the affinity to be different to where
    all the memory is allocated then you've just shot yourself in the foot :)
    This option may not exist on uniprocessor machines ... the only Windows
    machine I've got is a duallie :)

    Alternatively, there's programs like prioaff that will let you start up an
    application with a particular affinity.

    --
    Michael Brown
    www.emboss.co.nz : OOS/RSI software and more :)
    Add michael@ to emboss.co.nz - My inbox is always open


    ------------ And now a word from our sponsor ----------------------
    For a quality mail server, try SurgeMail, easy to install,
    fast, efficient and reliable. Run a million users on a standard
    PC running NT or Unix without running out of power, use the best!
    ---- See http://netwinsite.com/sponsor/sponsor_surgemail.htm ----
  7. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    Steve Wolfe wrote:
    >>I am contemplating the purchase of a dual processor opteron WINXP
    >>machine. The motivation is that currently several of my programs bring
    >>my processor (single procsessor machine) to its knees. So, if I get a
    >>dual processor machine my idea is that the program that currently brings
    >>my one processor machine to its knees will only bring ONE of the TWO
    >>processors to its knees and leave one processor available to do other
    >>tasks. I am assuming here that none of the software is specifically
    >>targeted (written) to use multi processors.
    >>
    >>Also, I have checked-into why my one processor is on its knees (100%
    >>utilized) and it is not because I am running out of RAM, got plenty left.
    >>
    >>Anyhow, with WINXP how will this dual opteron machine partition all the
    >>processor power between programs?
    >
    >
    > One of the largest benefits of dual-procs in a desktop is the increased
    > responsiveness. On one hand, with a decent OS, you're only doing half of
    > the context-changes, you can handle twice as many interrupts, and the APIC
    > interrupt controller does things more intelligently than the older PIC
    > controllers.
    >
    > On my machine and other machines around the office, we have a few apps
    > that will make a single-proc machine of any speed pretty unusable while
    > they're running. As long as the apps aren't doing something like using
    > twice the memory that the machine actually has, dual-CPU machines are always
    > MUCH more usable when running those apps. The actual throughput of one
    > individual app probably won't go up, it may even go DOWN a little - but
    > you're still able to use the machine while they're running.
    >
    > In fact, right now I'm on a dual Athlon MP 2600, and at home I have an
    > AthlonXP 3200. For any single app, it will run faster on the single-proc
    > machine: But I have a number of apps which, when run on the single-proc
    > machine, bog it down enough that I can't do anything else while they're
    > running. On the dually, there are VERY few apps that will have the same
    > effect.
    >
    > In fact, some time ago I was able to overload the power supply enough to
    > cause instability. Both CPUs were pegged at 100% by various processes and a
    > SETI to help out, I was ripping a DVD to a hard drive, copying files from
    > one hard drive to another, and burning another DVD from files over the
    > network - all at the same time. During that, the machine was nicely
    > responsive and usable, at least until the power supply gave out. = )
    >
    > I regularly use some FAST single-, dual-, and quad-CPU machines, and no
    > matter how fast the single-CPU machine is, it never feels as responsive
    > under load as the multi-proc machines: But for a single app, a fast
    > single-CPU machine will get the job done as fast or faster, and for a lot
    > less money. For the cost of a reasonable dual-CPU setup, you can easily set
    > up two single-proc machines!
    >
    > In fact, some time ago a person asked about buying a dual Xeon system for
    > running DVD shrink on a good number of DVDs at the same time. With some
    > actual prices, I pointed out that for the same cost, he could build five(!)
    > single-proc stations. = )
    >
    > steve
    >
    >
    Hi Steve,
    Thanks for your informative answer to my questions. Quite the conundrum
    really since it does sound like a dual proc machine would be much more
    responsive under load, but, may actually slow down individual programs
    as compared to a single proc machine. I , of course, have the need to
    run a given program (a simulationprogram) as fast as possible, while
    still being able to get other work done like emails, documents, and the
    like. Do you have any idea what the effect of NUMA capability (Perhaps
    WinXP pro sp2 has this) would add to the mix? That is, with NUMA, could
    you regain any speed loses that the dual proc machine has?

    Thanks a bunch!

    Paul
  8. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    > Hi Steve,
    > Thanks for your informative answer to my questions. Quite the conundrum
    > really since it does sound like a dual proc machine would be much more
    > responsive under load, but, may actually slow down individual programs
    > as compared to a single proc machine. I , of course, have the need to
    > run a given program (a simulationprogram) as fast as possible, while
    > still being able to get other work done like emails, documents, and the
    > like. Do you have any idea what the effect of NUMA capability (Perhaps
    > WinXP pro sp2 has this) would add to the mix? That is, with NUMA, could
    > you regain any speed loses that the dual proc machine has?

    First, the slowdown: Assuming identical processers, the performance hit
    of running a single app on a dual-proc machine is pretty minor - you'll
    rarely see more than a couple of percent, so it's not something to worry
    about.

    As for NUMA, if the operating system's scheduler does a good job of
    scheduling programs close to their allocated memory, then you end up with
    lower-latency memory requests, if not, then the memory latency is a bit
    higher. Just how much of a benefit or detriment that is to your program
    depends on your program. Note that in a dual-Opteron setup, even requesting
    "far" memory (memory attached to the other processer) still has latency
    equal to or lower than a memory request on a Xeon, going through the
    northbridge.

    In any case, unless your program is incredibly sensitive to memory
    latency, then the NUMA implementation could help or hurt by a small bit
    (probably just a few %) depending on how well it was implemented - I haven't
    tried XP SP2.

    In the end, it will be far cheaper to just buy a second machine on which
    to run the process - the money you save will let you buy a faster CPU, disk,
    and all of the other goodies. If you have no choice but to run it on the
    same machine that you'd like to get other work done on, then using a
    dual-CPU machine will at least let you get something done while your process
    is chugging away.

    steve
  9. Archived from groups: alt.comp.hardware.overclocking.amd (More info?)

    Steve Wolfe wrote:
    >>Hi Steve,
    >>Thanks for your informative answer to my questions. Quite the conundrum
    >>really since it does sound like a dual proc machine would be much more
    >>responsive under load, but, may actually slow down individual programs
    >>as compared to a single proc machine. I , of course, have the need to
    >>run a given program (a simulationprogram) as fast as possible, while
    >>still being able to get other work done like emails, documents, and the
    >>like. Do you have any idea what the effect of NUMA capability (Perhaps
    >>WinXP pro sp2 has this) would add to the mix? That is, with NUMA, could
    >>you regain any speed loses that the dual proc machine has?
    >
    >
    > First, the slowdown: Assuming identical processers, the performance hit
    > of running a single app on a dual-proc machine is pretty minor - you'll
    > rarely see more than a couple of percent, so it's not something to worry
    > about.
    >
    > As for NUMA, if the operating system's scheduler does a good job of
    > scheduling programs close to their allocated memory, then you end up with
    > lower-latency memory requests, if not, then the memory latency is a bit
    > higher. Just how much of a benefit or detriment that is to your program
    > depends on your program. Note that in a dual-Opteron setup, even requesting
    > "far" memory (memory attached to the other processer) still has latency
    > equal to or lower than a memory request on a Xeon, going through the
    > northbridge.
    >
    > In any case, unless your program is incredibly sensitive to memory
    > latency, then the NUMA implementation could help or hurt by a small bit
    > (probably just a few %) depending on how well it was implemented - I haven't
    > tried XP SP2.
    >
    > In the end, it will be far cheaper to just buy a second machine on which
    > to run the process - the money you save will let you buy a faster CPU, disk,
    > and all of the other goodies. If you have no choice but to run it on the
    > same machine that you'd like to get other work done on, then using a
    > dual-CPU machine will at least let you get something done while your process
    > is chugging away.
    >
    > steve
    >
    >
    Hi Steve,
    Hey, I finally got back to read your reply. Thanks a lot for your email.
    I ended up ordering a duul proc machine. I'll post something at this
    site when I get the new machine delivered and make some benchmark tests
    and let you know what I find out!

    Thanks again,

    Paul
Ask a new question

Read More

AMD Dual Processor Processors Opteron Overclocking