i may just stick w/ 301.42 on Einstein@Home then...if it ain't broke, don't fix it.
for what its worth, my DistrRTgen run times went from ~00:29:45 to ~00:28:30 (a 4.2% improvement) by going from 301.42 to 306.23. this is with an MSI N560GTX-Ti Hawk GPU, which is factory OCed to 950MHz/2100MHz.
i may just stick w/ 301.42 on Einstein@Home then...if it ain't broke, don't fix it.
for what its worth, my DistrRTgen run times went from ~00:29:45 to ~00:28:30 (a 4.2% improvement) by going from 301.42 to 306.23. this is with an MSI N560GTX-Ti Hawk GPU, which is factory OCed to 950MHz/2100MHz.
I wanted to upgrade another one of my hosts but the one quad-core I still have is one in a tiny box so no way would a new PS fit inside (no big deal) but it doesn't look like a new vid card will fit in the box without cutting the box and I don't want to take all the time to get one of my full sized box's out of my stash and moving everything over to one (just saying that makes me want to do it now)
But I just ordered a EVGA GeForce GTX 550 Ti Fermi since the price is low and added 8GB more Ram and a 750watt PS so it will either go in the 3-core that would just take a few minutes or......maybe the quad if I take a bit longer to do it. (then decide on the drivers)
i may just stick w/ 301.42 on Einstein@Home then...if it ain't broke, don't fix it.
for what its worth, my DistrRTgen run times went from ~00:29:45 to ~00:28:30 (a 4.2% improvement) by going from 301.42 to 306.23. this is with an MSI N560GTX-Ti Hawk GPU, which is factory OCed to 950MHz/2100MHz.
I wanted to upgrade another one of my hosts but the one quad-core I still have is one in a tiny box so no way would a new PS fit inside (no big deal) but it doesn't look like a new vid card will fit in the box without cutting the box and I don't want to take all the time to get one of my full sized box's out of my stash and moving everything over to one (just saying that makes me want to do it now)
But I just ordered a EVGA GeForce GTX 550 Ti Fermi since the price is low and added 8GB more Ram and a 750watt PS so it will either go in the 3-core that would just take a few minutes or......maybe the quad if I take a bit longer to do it. (then decide on the drivers)
That's probably the reason I still stick with 275.33 and 280.xx drivers, since
no real problems for (older) drivers have to be replaced using NVidias GTX 470 & 480 GPUs.(Older FERMI GPUs like GTX470 & 480).
(Also the use of over-sized PSUs, to ensure the Mobo/CPU/DRAM/GPU(s) power-
hungry methods as a result of too many WUs/GPU is possible!)
Also (I) stick, with 1 per GPU (or 2) or even more simultanious) WUs (BRP4CUDA/ATI) per GPU; if it appears to be more effective running more
then 1 or 2 at a time per GPU.
The availebillity of work is equal inportant, see SETI@home where sometimes
1 MB WU (6.08/6.09/6.10) is available! Or AstroPulse work.
Takes some time to see or know, it's really worthwhile doing more then 1,
or more per GPU, as there are a lot of (hard-ware) reasons, to do only
1 at a time per GPU!
Depending on GPU alone is hard to determine. As you're whole system depends
on doing this whithout making errors/overheating, not only Mobo/CPU/DRAM & in 'older systems' that's first QUAD-Core (Q6600/6700/--Q9550/Q{X}9650) CPUs
and first ' FERMI' GPUs, GTX460/470 etc.,
to the Sandy-Bridge/Ivy-bridgeCPUs combined with AMD/ATI >5000
or NVidias 600 (KEPPLER) and ofcoarse a suitable Mobo! (And 'case')!
Stepped back, for 24 to 48 hours to see if running more then 1 WU/GPU, is
worthwhile, that is if effeciency gains and GPU load increases,
whithout overloading the GPU(s) or the whole system to the point it starts
making errors! (As I've seen my I7-2600, shuts down, cause it reaches
95C, which is setted as safeguard-temp!).
Each project has it's own specific CPU/GPU, "mix" and best to find out
it's 'best value' per projects, as this may or can be different on each project!
Ok I have my next version of the GTX 550Ti 2GB fermi running for 24 hours with no problems and it is slightly faster running cuda X2 than the other 550Ti 1GB that is overclocked
At first I was having problems and the times were slow and it was acting like it was a heat problem but after checking with GPU-Z that wasn't the problem and I even tried using EVGA Precision to overclock to keep bring the clock speeds up to normal.......no luck......just continued dropping the clock down to as low as 100ghz
I continued messing around and at the same time was talking to a EVGA tech who made the basic suggestions and at the same time I was watching it run it's 2nd task without failing so I decided to just change one more setting to performance instead of quality.
Well soon after I watched it finish a few more tasks (until 5am) and the next day when I checked it finally decided to run cuda X2 and has been running non-stop with no problem @ X2 per 53mins and the other overclocked 550Ti 1GB runs X2 average 56mins
I didn't think this new one would be faster since it is my older 3-core and it barely has 4GB Ram and the other one is a quad-core with over 8GB Ram
HD 7950 ------> 1x 1,145
GTX 560 Ti ----> 1x~1,100
7950 outperforms 560 ti by 2 times counting in tflops, but has the same WU's crunch time.
Wherefrom do you get these crunching-times?
My HD6950 crunches 2wu's in 62 min win7 64 i7 two cores left free 5632 GFLOPS peak
My GTX550ti crunches 2 wu's in 53min win7 64 i3 two cores left free 891 GFLOPS peak
In general I agree, nVidia crunches faster in many projects, but the CNG-Cards (HD7xxx series) should perform much better than your crunching-times show.
Maybe you have a setup-problem?
i may just stick w/ 301.42 on
)
i may just stick w/ 301.42 on Einstein@Home then...if it ain't broke, don't fix it.
for what its worth, my DistrRTgen run times went from ~00:29:45 to ~00:28:30 (a 4.2% improvement) by going from 301.42 to 306.23. this is with an MSI N560GTX-Ti Hawk GPU, which is factory OCed to 950MHz/2100MHz.
RE: i may just stick w/
)
I wanted to upgrade another one of my hosts but the one quad-core I still have is one in a tiny box so no way would a new PS fit inside (no big deal) but it doesn't look like a new vid card will fit in the box without cutting the box and I don't want to take all the time to get one of my full sized box's out of my stash and moving everything over to one (just saying that makes me want to do it now)
But I just ordered a EVGA GeForce GTX 550 Ti Fermi since the price is low and added 8GB more Ram and a 750watt PS so it will either go in the 3-core that would just take a few minutes or......maybe the quad if I take a bit longer to do it. (then decide on the drivers)
RE: RE: i may just stick
)
That's probably the reason I still stick with 275.33 and 280.xx drivers, since
no real problems for (older) drivers have to be replaced using NVidias GTX 470 & 480 GPUs.(Older FERMI GPUs like GTX470 & 480).
(Also the use of over-sized PSUs, to ensure the Mobo/CPU/DRAM/GPU(s) power-
hungry methods as a result of too many WUs/GPU is possible!)
Also (I) stick, with 1 per GPU (or 2) or even more simultanious) WUs (BRP4CUDA/ATI) per GPU; if it appears to be more effective running more
then 1 or 2 at a time per GPU.
The availebillity of work is equal inportant, see SETI@home where sometimes
1 MB WU (6.08/6.09/6.10) is available! Or AstroPulse work.
Takes some time to see or know, it's really worthwhile doing more then 1,
or more per GPU, as there are a lot of (hard-ware) reasons, to do only
1 at a time per GPU!
Depending on GPU alone is hard to determine. As you're whole system depends
on doing this whithout making errors/overheating, not only Mobo/CPU/DRAM & in 'older systems' that's first QUAD-Core (Q6600/6700/--Q9550/Q{X}9650) CPUs
and first ' FERMI' GPUs, GTX460/470 etc.,
to the Sandy-Bridge/Ivy-bridge CPUs combined with AMD/ATI >5000
or NVidias 600 (KEPPLER) and ofcoarse a suitable Mobo! (And 'case')!
Stepped back, for 24 to 48 hours to see if running more then 1 WU/GPU, is
worthwhile, that is if effeciency gains and GPU load increases,
whithout overloading the GPU(s) or the whole system to the point it starts
making errors! (As I've seen my I7-2600, shuts down, cause it reaches
95C, which is setted as safeguard-temp!).
Each project has it's own specific CPU/GPU, "mix" and best to find out
it's 'best value' per projects, as this may or can be different on each project!
Ok I have my next version of
)
Ok I have my next version of the GTX 550Ti 2GB fermi running for 24 hours with no problems and it is slightly faster running cuda X2 than the other 550Ti 1GB that is overclocked
At first I was having problems and the times were slow and it was acting like it was a heat problem but after checking with GPU-Z that wasn't the problem and I even tried using EVGA Precision to overclock to keep bring the clock speeds up to normal.......no luck......just continued dropping the clock down to as low as 100ghz
I continued messing around and at the same time was talking to a EVGA tech who made the basic suggestions and at the same time I was watching it run it's 2nd task without failing so I decided to just change one more setting to performance instead of quality.
Well soon after I watched it finish a few more tasks (until 5am) and the next day when I checked it finally decided to run cuda X2 and has been running non-stop with no problem @ X2 per 53mins and the other overclocked 550Ti 1GB runs X2 average 56mins
I didn't think this new one would be faster since it is my older 3-core and it barely has 4GB Ram and the other one is a quad-core with over 8GB Ram
7970 3 WU = 41:5333
)
7970 3 WU = 41:5333
Surprised no gtx460
)
Surprised no gtx460 figures...this is now
I´m running two gtx460s, i3, nVidia 295.71 and Linux. brp1.28
GTX 460 -> 1x3000, 2x4800 PCIe x16, 768MB
GTX 460 -> 1x4700, 2x8400 PCIe x4, 768MB
They run at a cool 41-46C, and so rather quiet - and cheap second hand!
In hindsight a mobo with more PCIex16s would have been wiser.
I´m hoping to get a regular 50+ BRP4 WUs done per day once I ween myself off some other projects Primegrid etc.
AMD GPUs not in favor now? as
)
AMD GPUs not in favor now?
as far as i see nvidia cards give better results for the similar performance.
for example:
HD 7950 ------> 1x 1,145
GTX 560 Ti ----> 1x~1,100
7950 outperforms 560 ti by 2 times counting in tflops, but has the same WU's crunch time.
RE: HD 7950 ------> 1x
)
Wherefrom do you get these crunching-times?
My HD6950 crunches 2wu's in 62 min win7 64 i7 two cores left free 5632 GFLOPS peak
My GTX550ti crunches 2 wu's in 53min win7 64 i3 two cores left free 891 GFLOPS peak
In general I agree, nVidia crunches faster in many projects, but the CNG-Cards (HD7xxx series) should perform much better than your crunching-times show.
Maybe you have a setup-problem?
here
)
here http://einsteinathome.org/node/196380&nowrap=true#119110
right now i have no any setup problem with amd card because i have no amd card at all. but i'm looking to buy something on GCN.
i have several results
)
i have several results too:
GTX 560 Ti ----> 1x~1,100, 2x 2,000, 4x 4,100, 5x 5,200
GTS 450 ----> 1x~2,200, 2x 4,200
GT 240 ----> 1x~5,400, 2x 10,500