SMP only using one core

GargGarg Purveyor of Lincoln Nightmares Icrontian
edited September 2012 in Folding@Home
My googling only brought up old threads, mostly full of people who aren't setting the SMP flag.

My work computer, an Athlon II X4 635 @ 3400, usually runs SMP just fine. Lately, though, it's been stuck using just one core. A few minutes ago, it was using all four, then I closed it, restarted the computer, starting folding again, and it's only using one. Restarting the client doesn't help. There are no warning messages in the console that I've noticed, and I've got -forceasm and -smp 4 in the startup options.

Anyone have any ideas? It's too hot to fold at home, so this is my main producer at the moment.

Comments

  • TushonTushon I'm scared, Coach Alexandria, VA Icrontian
    There is the line "Mapping NT 1 to 1" so it is only detecting one core, whereas looking at the previous log it was NT 4 to 4. I would backup preferences/WUs and move the folder out, re-download and run it with -configonly the first time to make sure it's all setup correctly and try it again.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    I got that a few times, tunred out to be the WUs, not the setup. BUT, try new revision of FAHControl, as with it I no longer get WUs that force to single core folding. I set it up in its setup to be SMP, let it choose its own SMP variable, and it works with 8 cores folding very nicely. Intel CPU, though. so YMMV.
  • GargGarg Purveyor of Lincoln Nightmares Icrontian
    Tushon said:

    There is the line "Mapping NT 1 to 1" so it is only detecting one core

    Ah, I didn't realize that corresponded to cores. I'll give that a shot. Thanks!
    Straight_Man said:

    I got that a few times, tunred out to be the WUs, not the setup. BUT, try new revision of FAHControl, as with it I no longer get WUs that force to single core folding.

    I had to Google FAHControl, turns out there's a whole new FAH7 out. I'll have to check it out. I've had my clients running without any interference for so long I hadn't noticed. Gracias!

  • TushonTushon I'm scared, Coach Alexandria, VA Icrontian
    I'm still running 6.34 for my bigadv Linux client, but yeah, v7 has a whole new interface and is officially supported, though 6.34 should still work fine.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    I will explain a bit more about single-core versus multi-core and work units. What Folding calls uniprocessor is older single-core processors. And, there are developed projects that still use uniprocessor WUs that are still in the wild in a sense of being still active. The WUs from this kind of uniprpocessor WU project, as has been somewhat complexly noted and obfuscated in the Folding Forum(not here, Folding@Home's main Support forum), CAN be assigned to SMP machines for work. The New v7 setup for Folding is still beta, though public Beta now. some folks are finding (me included) that with v7 set of Folding software, the client tells Folding@Home's main servers for assignment that it is running on an SMP machine if it is detecting such or has been forced to such, and also how many cores are folding. The assignment servers are more and more cooperating with the client's info and giving assignments to servers with SMP-intended WUs on them if such are available and have WUs.

    v7 then appears to again request SMP WUs from the WU server assigned to by assignment server each time an assignment is done. If, and this DOES happen, folding has an outage or problem with networking or load issue on servers with SMP WUs, it will override and assign uniprocessor intended WUs to SMP clients, but more and more Stanford is trying to minimize this. So, it is not purely software at client end, but also availability of WUs for clients that rules sometimes.
  • GargGarg Purveyor of Lincoln Nightmares Icrontian
    Straight_Man said:

    So, it is not purely software at client end, but also availability of WUs for clients that rules sometimes.

    Yeah, that was one reason I was asking, in case there was some known issue in provisioning SMP WUs right now. Hopefully the new client improves things, but at least installing that should mean a fresh start without any corrupted files.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    Benn on v7 beta since Tushon ran his article about howto Linux fold. It does improve, noticeably, but not perfectly due to other factors also influencing what WUs are assigned. No client, by itself, rejects an assigned WU.

    True as to corrupted files, but if setup files were corrupt would suddenly be all the time single core over lots of WUs worth of work time. If it switches between single core and multicore folding over lots of WUs, is WUs that are the variable cause and not setup or machine factor most probably. If a change causes result change, demonstrably and consistently change resulting in change, then change contributed to result change-- Occam's Razor derivative applied to folding logic.
  • GargGarg Purveyor of Lincoln Nightmares Icrontian
    After wiping my work and cores, it downloaded another single core unit. Installed v7, and worked on all 4 straightaway. I'm digging the control interface so far. :thumbsup:
  • GargGarg Purveyor of Lincoln Nightmares Icrontian
    Being able to manage GPU & SMP from the same console is totally rad.

    I guess I'll believe it when it shows up on my stats, but right now it's showing an estimated 10k ppd out of my lowly Athlon II & Radeon 6670 combo. I just turned on the GPU for kicks, figuring its utilization of one CPU core wouldn't be worth it, but I guess we'll see. I don't even have SMP scaled back to 3 threads, yet.
  • GargGarg Purveyor of Lincoln Nightmares Icrontian
    Re: PPD - estimate was wrong as it hadn't had time to adjust to the second client. Seems better off with just SMP, since GPU takes a whole core.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    Unless the GPU is quite powerful, that will be the case. I have an i7 CPU here, built-in graphics capability. I had to tell the client to just SMP and it is doing better using SMP only, by about a 10-15% PPD improvement on average. That is a week or so with both versus months on SMP only now. AND, things are running cooler temp-wise.
  • TushonTushon I'm scared, Coach Alexandria, VA Icrontian
    @Gargoyle is referencing his AMD GPU, not an onboard IGP like the i7 sports. The OpenCL implementation is still really bad in terms of optimization, as he noticed, and not worth sacrificing a whole core to feed. CUDA still wins there and almost any Nvidia graphics card will be worth turning on for the 1-3% of CPU usage it might hit.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    I know, I was using the IGP as an example of a low-end GPU folding and how ineffective that is.

    Folding developer folks had to choose what they would tune for when they strategized GPU folding. The IGP and the lower end and mid-range AMD stuff uses things for graphics that favor CUDA processing. Thanks, I understand that and was not demeaning the equipment choices of another. I would consider an Nvidia card and an i7, if I could afford the card and the cooling it would need in order to be a great producer for folding. I am kinda stuck, as are most of us, with a limited budget. So, I SMP fold.
  • GargGarg Purveyor of Lincoln Nightmares Icrontian
    It'd probably take a pretty beefy AMD GPU to make it worthwhile. I used to have a 5770 in that rig, and it was just worth it. I've got two 5770s at home on my i5 rig, now, but since the 2500K's cores are more potent than the Athlons, it's better to just keep it on SMP, too. (when it's not too hot to fold at all, anyway)
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    As for the 2500K, I have a 2600K i7, and folding thinks it has 8 processors and folds SMP detecting same itself accordingly. As to beefy GPU, the GPUs that are beefy fold well, but folding favors Nvidia cards over AMD cards because Nvidia does raw graphical oriented preprocessing in GPU to a greater degree than AMD does. Thus, I could take a high-end Nvidia card that supported CUDA well, and the i7 would feed it semiraw data with one thread and not hobble the card to a great degree.
  • TushonTushon I'm scared, Coach Alexandria, VA Icrontian
    Straight_Man said:

    As for the 2500K, I have a 2600K i7, and folding thinks it has 8 processors and folds SMP detecting same itself accordingly. As to beefy GPU, the GPUs that are beefy fold well, but folding favors Nvidia cards over AMD cards because Nvidia does raw graphical oriented preprocessing in GPU to a greater degree than AMD does. Thus, I could take a high-end Nvidia card that supported CUDA well, and the i7 would feed it semiraw data with one thread and not hobble the card to a great degree.

    To clarify: it is solely CUDA optimization in the folding program that causes the huge advantage. Most Nvidia and AMD cards at the various price points are identical in terms of speed or other performance measures. They are fed identical data, whether it is translated through OpenCL or CUDA is the difference maker. The difference on SMP vs GPU with an AMD card has everything to do with the individual performance of the chips and cards in that configuration and has to be tested by the folder, as some CPUs will outproduce the SMP+GPU folding together if left alone, as stated.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    Tushon said:


    To clarify: it is solely CUDA optimization in the folding program that causes the huge advantage. Most Nvidia and AMD cards at the various price points are identical in terms of speed or other performance measures. They are fed identical data, whether it is translated through OpenCL or CUDA is the difference maker. The difference on SMP vs GPU with an AMD card has everything to do with the individual performance of the chips and cards in that configuration and has to be tested by the folder, as some CPUs will outproduce the SMP+GPU folding together if left alone, as stated.

    Um, there appears to also be two things to consider that can be figured into designing a folding machine:

    1. OpenCL loads the CPU more, and AMD cards are desinged to rely on that support some. So, to get SMP+GPU out of an AMD system, you need a faster AMD (or overclocked more) CPU And a higher-end or overclockable midrange GPU. Thus, in a high-heat ambient environment relatively (warmer to hot climate) you should plan on water or liquid or peltier cooling for the AMD stuff.

    2. Intel i7s are stable at high temps, my 2600k has run for 7 months straight at 140%+ load almost at 65C average CPU temp max. I also have it forced to run in Turbo 2.0 mode-- that is the extra 40% load. It would feed a high-quality CUDA tuned GPU with half a core at that speed, and the GPU would be running all-out. CUDA and OpenCL supporting calc systems are literally strategies for graphics calc implemented into a processing tuning software and hardware set ideally.

    The base difference is how much floating point calc and vector calc is done within the GPU, which determines how much has to precalced by the CPU before handing over to the GPU. The more the GPU does of this internally, IF it is fast enough and fast enough Graphics RAM is supplied on card, the freer the CPU is to also SMP fold more to full capacity of CPU. That is basic, lots of other things like OS load, graphics driver load, etc come into play here in truely tuning a system for Folding.

Sign In or Register to comment.