catalyst 6.11

Might not be the thread for it but i'll try :)

trying the gpu folding malarky on a x19xt-x and getting 2.0 secs a frame :eek: well its a surprise to me seeing as others mentioned in here are 8mins-ish so wtf going on? using the 6.10 cats + sample_X1950_8-301-060926a-036790 foldy bit as per their site, mind you does seem to be doing some awful big wu's taking a good few hours to complete just 1.
 
Steves_pc said:
Might not be the thread for it but i'll try :)

trying the gpu folding malarky on a x19xt-x and getting 2.0 secs a frame :eek: well its a surprise to me seeing as others mentioned in here are 8mins-ish so wtf going on? using the 6.10 cats + sample_X1950_8-301-060926a-036790 foldy bit as per their site, mind you does seem to be doing some awful big wu's taking a good few hours to complete just 1.
Click to expand...

unless im wrong theres only one type of GPU WU at the moment, its 330 points, has the equivelant of 100 frames, and even with a highly OC'd GPU will take you hours to finish

are you by any chance using the graphical version and looking at each individual step? these will go that quick but there's literally like a million of them

if you just use the command line client it tells you each frame (100 frames per GPU WU) or monitor the graphical client via fahmon, it wont give you any project info or a completion time, but it will tell you how many frames youv'e finished and you can easily see in the log how long each one is taking

from personal experience at 8mins per frame it takes ~14 hours non-stop to complete a WU
 
also the Graphical GPU client should give the same output to the logfile as the console (command line version) displays on screen anyway so you can see a comparable figure there

as mikeymike said what are normally being referred to as frames here (eg. 6-8mins a frame) are 1% of a full WU and not what the graphical client calls a frame
 
Hi Guys,

Yep sounds about right then, just fire up the graphicy bit now n then to see whats happening - i dont leave that bit of it running.

Going from memory (risky!) the last time i did a bit of folding it was taking about the same time but using just the cpu, thought the gfx based jobby was supposed to be a bit better but it doesnt seem to be making much difference :confused: maybe i need to go read up a bit more.

Systems an X2 3800 @2.5 with 2gb ram if it makes any difference.
 
assuming your using normal 2D settings, i am pretty sure the GPU client will still be folding faster than one of your cores will, by this i mean you will average a higher PPD, also dont forget you can fold on the second CPU core still, you just need to leave one core available for the GPU client to use (with the current beta GPU clients anyway) non-stop 24hr folding id say you can easily average ~650PPD on that system if you fold on the GPU@2D settings and also fold on the second CPU core :)

oh also i'd use the console version of the client, or always keep the GPU window minimised as i think you are doing anyway, having it open does actually slow down your folding :/
 
At slightly over 3D speeds (using ATI Tray Tool as in Concorde's guide) of 640/730MHz my X1900XT is giving me well over 700PPD compared to the 140PPD that I'm getting from a Standard Gromacs WU on the other core of my [email protected]

The GPU client is actually doing 20-40 times the work of the CPU client but since Stanford still need plenty of normal CPU clients running they've decided to limit the GPU points to 6x the CPU benchmark of 110PPD.
The points are a delicate balance to try and keep everyone happy, the GPU folk think they should be higher wheras those still relying on CPUs for most of their points would rather they were lower :o
 
ponts are less about how much sci is done, more about how much resorces it uses.
I feel that the 600PPD you get from the current GPU client is low. I can get that with some WUs on my E6300 at 3.2GHz, but I have to give up a whole core to run the GPU client. therfore I should get my CPU clients ppd + the GPU extra ontop. unless they can get the GPU not to whore the CPU so I can run 2CPU and 1GPU client.
 
You must log in or register to reply here.
Back
Top Bottom