OK, here are some numbers for a GTX470 and Linux 64Bit:
Load Time
55% 0h55m for 1 WU
79% 1h15m for 2 WU
84% 1h55m for 3 WU
The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.
Michael
@mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!
OK, here are some numbers for a GTX470 and Linux 64Bit:
Load Time
55% 0h55m for 1 WU
79% 1h15m for 2 WU
84% 1h55m for 3 WU
The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.
Michael
@mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!
E@H doesn't yet have an app for ATI GPUs. Only CUDA GPUs are supported, and CUDA is an nVidia-only technology. I'm sure there are a lot of OpenCL-capable ATI/AMD cards out there that would love to get in on the action.
OK, here are some numbers for a GTX470 and Linux 64Bit:
Load Time
55% 0h55m for 1 WU
79% 1h15m for 2 WU
84% 1h55m for 3 WU
The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.
Michael
@mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!
There are not many projects that have ATI applications for LINUX. The only one that works without too much trouble is DNETC. That's what it is doing, though I would love to do something else with that card - like Einstein or Milkyway :)
OK, here are some numbers for a GTX470 and Linux 64Bit:
Load Time
55% 0h55m for 1 WU
79% 1h15m for 2 WU
84% 1h55m for 3 WU
The CPU is a Phenom IIx4 945 at stock speed. Running RNA on the free cores.
Michael
@mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!
There are not many projects that have ATI applications for LINUX. The only one that works without too much trouble is DNETC. That's what it is doing, though I would love to do something else with that card - like Einstein or Milkyway :)
Michael
But your rac over there is only 20k, tou should be waaaaay above that with a 58xx card! I have a 5870 on Dnetc and am up above 260k and still climbing, I use Windows but that can't be the difference!
Here are the times and credits for one of my units:
904.91 11.95 3,091.20 DNETC@Home v1.31 (ati14)
and for one of your units:
10,777.46 99.83 3,292.45 DNETC@Home v1.02 (ati14)
For some reason yours is using a much different version of the Dnetc software, probably the Linux version. I can't believe it is that much slower!! My pc is doing units in about 15 minutes each, why is yours taking soooo much longer?!
Sorry for being ignorant, but I have a few questions.
I'd like to try and run more than one WU on my GTX 480 as well. Running just one seems awfully inefficient. From what I've read I think I'll need an XML file for that corresponding to the cuda WU I want to run. I guess it goes into einsteins project folder in my boinc data directory.
But do I just create a new XML document with something like the XML code posted earlier in this thread? I realize it'd need to have updated file versions and I'm also on win 7 64 bit instead of xp 32 bit.
Are there are pre-made XML files I could just use floating around or any how to guides? I tried to google but it's late and I'm sleepy and thus not finding anything useful.
Sorry for being ignorant, but I have a few questions.
I'd like to try and run more than one WU on my GTX 480 as well. Running just one seems awfully inefficient. From what I've read I think I'll need an XML file for that corresponding to the cuda WU I want to run. I guess it goes into einsteins project folder in my boinc data directory.
But do I just create a new XML document with something like the XML code posted earlier in this thread? I realize it'd need to have updated file versions and I'm also on win 7 64 bit instead of xp 32 bit.
Are there are pre-made XML files I could just use floating around or any how to guides? I tried to google but it's late and I'm sleepy and thus not finding anything useful.
I will throw something in the mix here, you need to see how much memory each current unit is using, then you can figure out if playing with an app_info.xml file will do any good. If for instance your current gpu has 512 meg of memory and an Einstein unit takes 350 meg then there is not enough space left over for another unit to run at the same time. If however you have a 2gb gpu and the same 350mb unit, then yes you can run multiple units on the same card.
To use multiple cards, which I think is what you are asking, you also need to check the current status of Boinc Manager and see what it says, if it says you are already running more than one gpu unit at a time, 4 in your case, than you may be maxed out depending on the memory requirements of each unit. If however you only see one unit running but have 4 gpu's in one machine then it may NOT an app_info.xml file you need but a cc_config.xml file which is entirely different. A line in the cc_config.xml file will ensure all gpu's on the machine are working, if all gpu's on the machine ARE working and crunching one unit, then you may not be able to separate them.
You probably looked at my old pc, it has an old 8800GT with 512 MB, my current pc has a GTX480 with 1500 MB. I'm not asking to use multiple cards btw, I only got 1 GPU per system, but I'd like that one GPU, at least the 480 to crunch more than one WU. If I can't it's much more productive for it to just run GPUGRID all the time but I like einstein@home.
Anyways, thanks for the tips, I found the stuff about the einstein apps in my client_state.xml. I'll need that for the app_info and then change a cuda setting from 1 to 0.25 right?
Any more advice is more than welcome though, a bit busy right now with other things so I'm waiting a bit before I start fiddling around with making an app_info.
Any more advice is more than welcome though, a bit busy right now with other things so I'm waiting a bit before I start fiddling around with making an app_info.
spare at least 2 CPU-cores if you want to run 4 cuda-jobs at maximum speed.
don't start 4 of them at a time - give them an interleave of a minute.
sorry to say, but currently einstein has the worst performing cuda app around and it's a real pita to squeeze out a little bit on our side..
of course they could do what you want server side and give us an option to select the number of jobs we want to run in parallel. at least as long as the app is as it is now.
RE: OK, here are some
)
@mickydl* Where is your "CAL ATI Radeon HD5x00 series (Redwood) (1024MB) driver: 1.4.815" crunching? It seems that pc, AuthenticAMD AMD Phenom(tm) II X4 940 Processor [Family 16 Model 4 Stepping 2] (4 processors), is ONLY doing cpu units!
RE: RE: OK, here are some
)
E@H doesn't yet have an app for ATI GPUs. Only CUDA GPUs are supported, and CUDA is an nVidia-only technology. I'm sure there are a lot of OpenCL-capable ATI/AMD cards out there that would love to get in on the action.
-- Tony D.
RE: RE: OK, here are some
)
There are not many projects that have ATI applications for LINUX. The only one that works without too much trouble is DNETC. That's what it is doing, though I would love to do something else with that card - like Einstein or Milkyway :)
Michael
RE: RE: RE: OK, here
)
But your rac over there is only 20k, tou should be waaaaay above that with a 58xx card! I have a 5870 on Dnetc and am up above 260k and still climbing, I use Windows but that can't be the difference!
Here are the times and credits for one of my units:
904.91 11.95 3,091.20 DNETC@Home v1.31 (ati14)
and for one of your units:
10,777.46 99.83 3,292.45 DNETC@Home v1.02 (ati14)
For some reason yours is using a much different version of the Dnetc software, probably the Linux version. I can't believe it is that much slower!! My pc is doing units in about 15 minutes each, why is yours taking soooo much longer?!
I don't want to hijack this
)
I don't want to hijack this thread with an off-topic discussion so I'm moving it to ATI performance
Michael
Sorry for being ignorant, but
)
Sorry for being ignorant, but I have a few questions.
I'd like to try and run more than one WU on my GTX 480 as well. Running just one seems awfully inefficient. From what I've read I think I'll need an XML file for that corresponding to the cuda WU I want to run. I guess it goes into einsteins project folder in my boinc data directory.
But do I just create a new XML document with something like the XML code posted earlier in this thread? I realize it'd need to have updated file versions and I'm also on win 7 64 bit instead of xp 32 bit.
Are there are pre-made XML files I could just use floating around or any how to guides? I tried to google but it's late and I'm sleepy and thus not finding anything useful.
RE: I tried to google but
)
You could google for "
But you should try to get some sleep beforehand. ;-)
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: Sorry for being
)
I will throw something in the mix here, you need to see how much memory each current unit is using, then you can figure out if playing with an app_info.xml file will do any good. If for instance your current gpu has 512 meg of memory and an Einstein unit takes 350 meg then there is not enough space left over for another unit to run at the same time. If however you have a 2gb gpu and the same 350mb unit, then yes you can run multiple units on the same card.
To use multiple cards, which I think is what you are asking, you also need to check the current status of Boinc Manager and see what it says, if it says you are already running more than one gpu unit at a time, 4 in your case, than you may be maxed out depending on the memory requirements of each unit. If however you only see one unit running but have 4 gpu's in one machine then it may NOT an app_info.xml file you need but a cc_config.xml file which is entirely different. A line in the cc_config.xml file will ensure all gpu's on the machine are working, if all gpu's on the machine ARE working and crunching one unit, then you may not be able to separate them.
You probably looked at my old
)
You probably looked at my old pc, it has an old 8800GT with 512 MB, my current pc has a GTX480 with 1500 MB. I'm not asking to use multiple cards btw, I only got 1 GPU per system, but I'd like that one GPU, at least the 480 to crunch more than one WU. If I can't it's much more productive for it to just run GPUGRID all the time but I like einstein@home.
Anyways, thanks for the tips, I found the stuff about the einstein apps in my client_state.xml. I'll need that for the app_info and then change a cuda setting from 1 to 0.25 right?
Any more advice is more than welcome though, a bit busy right now with other things so I'm waiting a bit before I start fiddling around with making an app_info.
RE: Any more advice is more
)
spare at least 2 CPU-cores if you want to run 4 cuda-jobs at maximum speed.
don't start 4 of them at a time - give them an interleave of a minute.
sorry to say, but currently einstein has the worst performing cuda app around and it's a real pita to squeeze out a little bit on our side..
of course they could do what you want server side and give us an option to select the number of jobs we want to run in parallel. at least as long as the app is as it is now.