This other user has a supercomputer compared to my quite old gaming rig.
AMD GPU's are quite fast crunchers indeed.
At some projects yes they are, but not at ALL projects.
Interesting. the 491 machine in the comparison running an identical AMD card to mine but on a windows box is turing in a run time of around 2400. My machine on Linux with the same AMD is turning in run times of around ~4500. I do have a mix of WUs so I am not sure how this could effect run time performance
... the 491 machine in the comparison running an identical AMD card to mine but on a windows box is turing in a run time of around 2400. My machine on Linux with the same AMD is turning in run times of around ~4500.
The '491' host belongs to ZomBiE80, who started this recent discussion. As a new participant, I'm guessing he/she may be running tasks 1x. How many concurrent tasks are you running? Even if only 2x, you're probably already beating the '491' host :-).
This other user has a supercomputer compared to my quite old gaming rig.
AMD GPU's are quite fast crunchers indeed.
At some projects yes they are, but not at ALL projects.
Quite correct. Depends on the project and the app itself. Sometimes an OpenCL app is applicable to a given project and can be faster on Kepler cards than its CUDA counterpart.
It's been a while since I ran Einstein Arecibo tasks, but the following are the results running one Arecibo GPU task at a time, along with 3 intensive AVX CPU tasks from another project.
The GTX TITAN is faster than the Tesla K20 in distributed computing apps and really shines when double-precision is called. It seems that both AMD and NVIDIA have seen fit to cripple double-precision performance in their latest offerings (other than the TITAN) and are concentrating on the gaming market.
Now, if you are bitcoin mining, then AMD is the way to go as far as consumer/enthusiast graphics cards.
Version 337.88 - WHQL
Release Date Mon May 26, 2014
.....ah yes the fun of d/l'ing and installing the drivers on 6 of my 7 hosts
this will take a while
Have you tested it to see if it is better then your current drivers? I never install the newest drivers until after other test it out first and start reporting back that it IS helpful for crunching. I am not a gamer so no need for me to jump in too early, what I have works.
Have you tested it to see if it is better then your current drivers?
I upgraded from 335.23 to 337.88 on my GTX 660 Ti (Win7 64-bit), and don't see any difference with S6CasA, being 28 1/2 minutes in either case. Also, I run GPUGrid on a pair of GTX 750 Ti's, and don't see any difference with the CUDA work there (WinXP).
Note however that for some curious reason I don't understand, after upgrading the 660 Ti, all of the remaining S6 CasA work units in the buffer errored out. I had to download a new group before they ran properly (BOINC 7.3.19 x64).
I'm running the notebook version of 337.88 for 2,5 days now (755M on Wintendo 7 pro 64bit) and it works well for FGRP3. No erroring out, invalids or increase in runtime. Just as if I didn't upgrade at all.
(I haven't tried other applications yet, because I like FGRP3-opencl being suboptimal enough to not overheat this stupid laptop.)
Quite interesting result of
)
Quite interesting result of Workunit 187859516 http://einsteinathome.org/workunit/187859516
This other user has a supercomputer compared to my quite old gaming rig.
AMD GPU's are quite fast crunchers indeed.
RE: Quite interesting
)
At some projects yes they are, but not at ALL projects.
RE: RE: Quite interesting
)
Interesting. the 491 machine in the comparison running an identical AMD card to mine but on a windows box is turing in a run time of around 2400. My machine on Linux with the same AMD is turning in run times of around ~4500. I do have a mix of WUs so I am not sure how this could effect run time performance
RE: ... the 491 machine in
)
The '491' host belongs to ZomBiE80, who started this recent discussion. As a new participant, I'm guessing he/she may be running tasks 1x. How many concurrent tasks are you running? Even if only 2x, you're probably already beating the '491' host :-).
Cheers,
Gary.
RE: RE: Quite interesting
)
Quite correct. Depends on the project and the app itself. Sometimes an OpenCL app is applicable to a given project and can be faster on Kepler cards than its CUDA counterpart.
It's been a while since I ran Einstein Arecibo tasks, but the following are the results running one Arecibo GPU task at a time, along with 3 intensive AVX CPU tasks from another project.
http://einsteinathome.org/account/7867/computers
The GTX TITAN is faster than the Tesla K20 in distributed computing apps and really shines when double-precision is called. It seems that both AMD and NVIDIA have seen fit to cripple double-precision performance in their latest offerings (other than the TITAN) and are concentrating on the gaming market.
Now, if you are bitcoin mining, then AMD is the way to go as far as consumer/enthusiast graphics cards.
nVidia GeForce driver time
)
nVidia GeForce driver time again
http://www.geforce.com/drivers
Version 337.88 - WHQL
Release Date Mon May 26, 2014
.....ah yes the fun of d/l'ing and installing the drivers on 6 of my 7 hosts
this will take a while
RE: nVidia GeForce driver
)
Have you tested it to see if it is better then your current drivers? I never install the newest drivers until after other test it out first and start reporting back that it IS helpful for crunching. I am not a gamer so no need for me to jump in too early, what I have works.
RE: Have you tested it to
)
I upgraded from 335.23 to 337.88 on my GTX 660 Ti (Win7 64-bit), and don't see any difference with S6CasA, being 28 1/2 minutes in either case. Also, I run GPUGrid on a pair of GTX 750 Ti's, and don't see any difference with the CUDA work there (WinXP).
Note however that for some curious reason I don't understand, after upgrading the 660 Ti, all of the remaining S6 CasA work units in the buffer errored out. I had to download a new group before they ran properly (BOINC 7.3.19 x64).
RE: I upgraded from 335.23
)
I forgot. It is a GTX 660. I moved the Ti to another board; there is relatively little difference on the S6CasA.
I'm running the notebook
)
I'm running the notebook version of 337.88 for 2,5 days now (755M on Wintendo 7 pro 64bit) and it works well for FGRP3. No erroring out, invalids or increase in runtime. Just as if I didn't upgrade at all.
(I haven't tried other applications yet, because I like FGRP3-opencl being suboptimal enough to not overheat this stupid laptop.)