Hey guys back on the fold after the summer break, work shop needed heating again But the WU ppd seems eratic 6000-8500 ppd up and down in the same WU, running Server 08 R2 and the latest Nvid drivers with a GTX260. Any ideas?
Hey tom, good to hear you're back with us again Could you be more specific about which unit you're getting different point values for? It may just be a different variation of the same project, which will have similar PRCG ID's in fahmon.
on any work unit Im getting up to 2500 variation over the course of that 1 wu. Its a posh nas/workshop rig so hasnt been doing anything but fold. Also noticed if I remote desktop in it eue's the next wu. finishes the current one but eue's the next. All very strange goings on atm
I don't think i've ever heard of variation in WU point value. It may be a fahmon reporting error, but the fact that it's running on S2008 may have something to do with it as well. I don't know how easy or otherwise it is to fold on windows servers, but I can imagine it being somewhat different to the likes of XP or Vista. What are you using to access the machine via remote desktop? Built in windows software, VNC, logmein etc? It's probably putting too much load on the GPU causing it to crash (EUE) although I'd have assumed it would crash on the current unit and not the next. Hopefully someone else will be able to shed some light on these issues, your rig is definately displaying some strange errors like I've never heard of before.
if your using windows 2003 server 32 or 64 (its XP really) its the same, apart from 64bit more then 1.5GB been usable when you have 6 video cards installed (3x9800gx2) but i have not found 64bit 2003 or XP 32 more stable then the other both perfome the same for 2008 server that runs off vista code so you have issues getting more then the first video card folding if dummy plugs are not used (unless they have fixed that) in any case you must move the slider back back to Full hardware acceration and tick the box as well (2003 its set 1 from the bottom, 2008 as no driver maker has made support for lower modes i think its full speed default but dummy plugs may be needed) if your doing SMP folding and GPU at the same time its recommended that 64bit windows is used due to lack of usable ram (only an issue if your wanting to use the system as well) also set the SMP cores to 90% usage to prevent the SMP from draging the GPU2 cores down
Ok mate I've had another think about this and tried to come up with some solutions for you. Sorry if none of this is new information for you - I'm not sure how extensive your knowledge of the GPU2 client is. Crashing after being accessed via remote desktop - if you are, as I suspect, using the inbuilt windows remote access application, it will request more graphics resources than, say, VNC or Logmein do. It has always been a major resource hog in my experience, which is precisely why I use logmein (or VNC when on site). The other thing to ensure is that your client has the "core priority" flag set to idle rather than low. This means that when you send the GPU a request for resources, like playing a video in a web browser etc, the client will throttle back the core usage to accomodate the second GPU intensive task. I think remote desktop would use the card so this may be what's causing your problem. As for why it's EUEing on the next unit and not the current, i'm going to have to pass on that and hope that someone else can provide an explanation. Maybe temperature is an issue? As for your problems with unit point values - I think your fahmon monitoring settings may be wrong. If you're seeing the PPD fluctuate from one unit to the next and the units are the same, or even if you see it fluctuate during the same WU run, it's probably because in "preferences > monitoring" you have "Calculate PPD based on" set to the wrong value. You want it set on "last 3 frames" to give you a fairly accurate reading. It may be currently set on effective rate, which does fluctuate as the GPU is used for other tasks at some points during certain WU runs. Setting it to monitor the last 3 frames is more of a stable value for PPD from a client. If that's not it, check your stats carefully on EOC and see if your actual stats on there reflect your reported stats (PPD) from fahmon. I'm making the assumption of course that your problem is with the value that you see for the GPUs PPD in fahmon.
Tell Ive had a summer off from all this, the PPD variation was a simple fix.... disable the ruddy screensaver! Still not solved the remote desktop issue. I am indeedy using the windows 7/ server 2008R2 (win7 kernal) remote sign in. Still no matter what happens if I sign in, it EUE's the next WU, but actualy completes the current one. Ive set up FAHmon over the network and will have to walk out to the shed to play with the clocks
walk out to the shed!!!!!! hey i never thought of that one,rigs in sheds,now wheres that sheds "r" us catalogue gone cheers spunky
Hve you not been following my rack farm thread spunky? My rigs live in an old, coalshed behind my garage that I'm going to convert to a server room
Tom, you're best off using VNC or logmein to access remote PC's. Remote Desktop gets the folding clients confused.
