@Bikeman: Have you got a card yet or are we just confusing you even more? It would be interesting to find out how fast the 9800GT "green" cards are. Plus you can use it to practise your cuda programming skills :-)
Haven't got it yet, should be delivered next week. The fastest CUDA card I now have is the GeForce 8600M GS in my notebook :-), it's so slow that it won't finish a GPUgrid task in time. "On paper", the new card should be about 8 times faster. The "green" variant is also about 10% slower than the regular 9800 GT (again, "on paper").
it's so slow that it won't finish a GPUgrid task in time.
may be client just need more optimized settings to run at full power?
No, the card is just too slow :-). GPUgrid maintains a list of cards that have enough power to meet the deadline, and my notebook's GPU is clearly "out".
Go on ditch it. After all you seem to have an abundance of i7's now :-)
I pulled my 2 9800GT's out and replaced them with GTS250's, plus another pair for good measure. A definate speed improvement there with GPUgrid and Seti. Now just waiting for the Einstein cuda app to be released...
Abundance?
Three is now abundance?
Now if I had 10 ... that might be abundance ...
The good news is that I am completing good tasks. THe bad news is that they have not purged the system of the bad tasks and it looks like they are making more of the same. And it affects almost all task name types ... then they seem to have run out of work ... sigh ... some days I wonder why I get out of bed ...
@Bikeman: Have you got a card yet or are we just confusing you even more? It would be interesting to find out how fast the 9800GT "green" cards are. Plus you can use it to practise your cuda programming skills :-)
Haven't got it yet, should be delivered next week. The fastest CUDA card I now have is the GeForce 8600M GS in my notebook :-), it's so slow that it won't finish a GPUgrid task in time. "On paper", the new card should be about 8 times faster. The "green" variant is also about 10% slower than the regular 9800 GT (again, "on paper").
CU
Bikeman
Well at least you can practise your cuda programming on the laptop while you wait for the 9800GT to arrive.
My 9800GT's (not green by any means) were taking 13 hours on the old one and 12 hours on the newer one to complete GPUgrid work. This is before their current rash of work units that have issues with the 185.85 driver.
Well at least you can practise your cuda programming on the laptop while you wait for the 9800GT to arrive.
Exactly, doing that at the moment. Oh, and I was wrong, I used an even slower card on GPUgrid (GPUgrid would not even run on my notebook because it's a 32 bit Linux and GPUgrid only has 64 bit Linux apps. The card that happened to ship with my quad-core 64 bit system has only one multiprocessor, and took about a week per GPUgrid task :-).
Issue # 11 of the German c't magazine mentions on p 148 that ATLAS now has a small rack of NVIDIA Geforce and Tesla equipped hosts. That doesn't mean that ATI will be neglected, of course, but at the moment NVIDIA seems to offer better support for universities and other research institutions.
Meanwhile...my GeForce 9800 GT "Eco" has arrived, is now installed in a 64 bit Linux system with a 4850e Athlon X2 CPU, and undergoing 24/7 stability tests. The card itself is mounted horizontally in the case, meaning that the massive heat-sink is facing downwards, which is opposite to what would be optimal for convection to carry the hot air away from the heat sink. There's a rather small fan installed on the card as well...
but ambient temp in the room can get rather hot and I'm wondering whether this will work out ok without installing additional ventilation.
According to
nvidia-settings -q gpucoretemp
the GPU temp now is 76 deg C under full load (GPUgrid), which seems ok. We'll see.
RE: @Bikeman: Have you got
)
Haven't got it yet, should be delivered next week. The fastest CUDA card I now have is the GeForce 8600M GS in my notebook :-), it's so slow that it won't finish a GPUgrid task in time. "On paper", the new card should be about 8 times faster. The "green" variant is also about 10% slower than the regular 9800 GT (again, "on paper").
CU
Bikeman
RE: it's so slow that it
)
may be client just need more optimized settings to run at full power?
RE: RE: it's so slow that
)
No, the card is just too slow :-). GPUgrid maintains a list of cards that have enough power to meet the deadline, and my notebook's GPU is clearly "out".
CU
Bikeman
RE: Go on ditch it. After
)
Abundance?
Three is now abundance?
Now if I had 10 ... that might be abundance ...
The good news is that I am completing good tasks. THe bad news is that they have not purged the system of the bad tasks and it looks like they are making more of the same. And it affects almost all task name types ... then they seem to have run out of work ... sigh ... some days I wonder why I get out of bed ...
RE: RE: @Bikeman: Have
)
Well at least you can practise your cuda programming on the laptop while you wait for the 9800GT to arrive.
My 9800GT's (not green by any means) were taking 13 hours on the old one and 12 hours on the newer one to complete GPUgrid work. This is before their current rash of work units that have issues with the 185.85 driver.
BOINC blog
RE: Well at least you can
)
Exactly, doing that at the moment. Oh, and I was wrong, I used an even slower card on GPUgrid (GPUgrid would not even run on my notebook because it's a 32 bit Linux and GPUgrid only has 64 bit Linux apps. The card that happened to ship with my quad-core 64 bit system has only one multiprocessor, and took about a week per GPUgrid task :-).
CU
Bikeman
RE: RE: Well at least
)
Howdy,
If GPU apps are used at Einstein. I hope it is not Cuda only.
My ATI HD4850 GPU already has Very Powerful double precision capabilities and,
I would like to put it to use on Einstein.
As a test, I hooked it up to MilkyWay at Home this morning with their ATI
GPU app and, crunched 12 of their work units in 2.94 minutes ! ( cpu time )
Bill
IIRC an OpenCL (Cross
)
IIRC an OpenCL (Cross platform) app is in the works, but the CUDA version is much farther along because nVidia supplied engineering support for it.
Issue # 11 of the German c't
)
Issue # 11 of the German c't magazine mentions on p 148 that ATLAS now has a small rack of NVIDIA Geforce and Tesla equipped hosts. That doesn't mean that ATI will be neglected, of course, but at the moment NVIDIA seems to offer better support for universities and other research institutions.
Meanwhile...my GeForce 9800 GT "Eco" has arrived, is now installed in a 64 bit Linux system with a 4850e Athlon X2 CPU, and undergoing 24/7 stability tests. The card itself is mounted horizontally in the case, meaning that the massive heat-sink is facing downwards, which is opposite to what would be optimal for convection to carry the hot air away from the heat sink. There's a rather small fan installed on the card as well...
but ambient temp in the room can get rather hot and I'm wondering whether this will work out ok without installing additional ventilation.
According to
nvidia-settings -q gpucoretemp
the GPU temp now is 76 deg C under full load (GPUgrid), which seems ok. We'll see.
CU
Bikeman
RE: .... 76 deg C under
)
Ah, a heater with calculating functions. Neat!
Put some liquid cooling mods in and you'll have a kettle for hot chocolate and stuff. :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal