No, you don't understand me. I am not telling that I am some kind of superman or my PC is some kind or supercomputer. I was so confused with my E@H problem. In the next few days I will try to enable 2 WU at once.
... My GPU is 85% load, CPU is not load (just overclocked, but I think this could not be problem). My PC is stable. Sometimes I am running stable 4 VM with LHC@Home - Test4Theory.
An overclocked CPU could be a problem. In any case, if you have to run the GPU severely underclocked, there has got to be an unresolved issue somewhere. Overclocking will draw more power and underclocking the GPU will probably save some power so perhaps your PSU isn't quite up to running an overclocked CPU and the GPU at standard settings. What is the brand and rating of the PSU and how old is it? I don't suppose you have access to a decent alternative PSU that you could try?
PSU is Corsair TX 750. It deliveres good and stable voltage and amps. I am running Battlefield 4 @ Ultra settings without any problems. PC was tested with stock settings too but E@H only works with underclocking GPU and memory. It is stable right now and I dont wont to make any changes (It works - do not touch). I noticed delay of changed E@H preferences too. I have to wait couple of hours after changing of preferences to apply them (update and resetting project does not help).
... but E@H only works with underclocking GPU and memory.
There has got to be a problem with your setup somewhere. I don't know anyone else who is underclocking the GPU like you are just to get it to work. It would be really interesting to try a different GPU (same model) to see if there is some problem with the GPU itself. You don't have an accommodating parts supplier by any chance - just in case the GPU is actually faulty? I would imagine that E@H tasks probably put more stress on the card (or certain parts of it) than games. It's a significant investment and you want to make sure it's 100% functional.
Quote:
I noticed delay of changed E@H preferences too. I have to wait couple of hours after changing of preferences to apply them (update and resetting project does not help).
If you make changes of the GPU utilization factor, you have to download new work in order for the changes to be sent to your host. The easiest way is to make a small change to your work cache settings locally so that a work fetch is triggered. After that you can disable local prefs in order to restore the website prefs. For other website pref changes, an 'update' should be all that is needed to have the new information sent to the client. You should NEVER 'reset' the project just to get a change in prefs. Resetting throws away all current tasks, program files and data and then downloads all new copies of everything. In other words, a complete waste for a preference change.
Yes, from the beggining it was very strange with E@H because all other BOINC projects are running smooth. Almost all errors were caused by my kids (shutting down loaded PC).
It is incompatible to cut in half raw power of Matrix HD 7970 to get the same results like other E@H members. The GPU itself has LED indicator for indication of GPU load. It is green colour and it indicates idle but MSI Afterburner indicates 85% load.
I created completly new E@H account without any improvement. I will try to find nVidia GPU to test it with my setup.
Yes, from the beggining it was very strange with E@H because all other BOINC projects are running smooth. Almost all errors were caused by my kids (shutting down loaded PC).
It is incompatible to cut in half raw power of Matrix HD 7970 to get the same results like other E@H members. The GPU itself has LED indicator for indication of GPU load. It is green colour and it indicates idle but MSI Afterburner indicates 85% load.
I created completly new E@H account without any improvement. I will try to find nVidia GPU to test it with my setup.
Your gpu is doing better then mine is! One of my tasks is here:
2,625.18 504.96 3.32 1,000.00 Binary Radio Pulsar Search (Arecibo, GPU) v1.39 (BRP4G-opencl-ati)
While this is one of yours:
1,797.64 390.56 4.05 1,000.00 Binary Radio Pulsar Search (Arecibo, GPU) v1.39 (BRP4G-opencl-ati)
You can see the same type of unit, both of us have an AMD 7970, but you have an Intel i7 8 core cpu, while I have an AMD 6 core cpu. You are also using a newer version of Boinc then I am, but we appear to be using the same gpu drivers. I have not tweaked mine at all, I just brought it back here yesterday in fact.
I am NOT trying to say mine is something special, it isn't! I am just trying to say that between you and I, yours is doing MUCH better then mine is!!
You should NEVER 'reset' the project just to get a change in prefs. Resetting throws away all current tasks, program files and data and then downloads all new copies of everything. In other words, a complete waste for a preference change.
In a normal project, perhaps yes, but not at Einstein@Home. As Einstein has the "resend lost tasks" function enabled on the server. This will compare the contents of the client_state.xml file with the contents of the database, to see if all tasks that the server appointed to the hostID are really on the hostID, and if there is a mismatch, to resend the tasks that the database thinks the hostID should have.
So all that a project reset will do here is that you dump all contents of the projects\einstein.phys.uwm.edu\ directory, to then get those same applications and tasks downloaded to your computer again.
Might still be a waste of bandwidth, but other than that, nothing.
The only ways to get new work here are:
1. run the work, upload, report, get new.
2. detach the project. This will delete the whole projects\einstein.phys.uwm.edu\ directory and any entry about the project in the client_state.xml file. So that when you reattached to the project, it has nothing to compare to, and will have to send you new work.
I am running 2 WU at once and it takes around 2 hours to crunch them (7000 - 7300 seconds. GPU clock is set @ 550 MHz and memory clock is set @ 825 MHz.
What motherboard, memory (and settings) are you using? Are you 100% sure you have no temperature / power issues?
You shouldn't have to underclock to that degree just to get stability on a card (with Einstein), that's so (potentially) capable.
If you are convinced that the 'Matrix' is in perfect working order (which I doubt) and the power supply is good (which should be, but maybe not). The problem must lie in the surrounding system components or system config...
Are you able to return the card to the supplier for replacement? The amount of underclock you need to apply really does worry me...
No, you don't understand me.
)
No, you don't understand me. I am not telling that I am some kind of superman or my PC is some kind or supercomputer. I was so confused with my E@H problem. In the next few days I will try to enable 2 WU at once.
Thank you Gavin!
RE: ... My GPU is 85% load,
)
An overclocked CPU could be a problem. In any case, if you have to run the GPU severely underclocked, there has got to be an unresolved issue somewhere. Overclocking will draw more power and underclocking the GPU will probably save some power so perhaps your PSU isn't quite up to running an overclocked CPU and the GPU at standard settings. What is the brand and rating of the PSU and how old is it? I don't suppose you have access to a decent alternative PSU that you could try?
Cheers,
Gary.
PSU is Corsair TX 750. It
)
PSU is Corsair TX 750. It deliveres good and stable voltage and amps. I am running Battlefield 4 @ Ultra settings without any problems. PC was tested with stock settings too but E@H only works with underclocking GPU and memory. It is stable right now and I dont wont to make any changes (It works - do not touch). I noticed delay of changed E@H preferences too. I have to wait couple of hours after changing of preferences to apply them (update and resetting project does not help).
RE: PSU is Corsair TX
)
OK, should be no problem.
There has got to be a problem with your setup somewhere. I don't know anyone else who is underclocking the GPU like you are just to get it to work. It would be really interesting to try a different GPU (same model) to see if there is some problem with the GPU itself. You don't have an accommodating parts supplier by any chance - just in case the GPU is actually faulty? I would imagine that E@H tasks probably put more stress on the card (or certain parts of it) than games. It's a significant investment and you want to make sure it's 100% functional.
If you make changes of the GPU utilization factor, you have to download new work in order for the changes to be sent to your host. The easiest way is to make a small change to your work cache settings locally so that a work fetch is triggered. After that you can disable local prefs in order to restore the website prefs. For other website pref changes, an 'update' should be all that is needed to have the new information sent to the client. You should NEVER 'reset' the project just to get a change in prefs. Resetting throws away all current tasks, program files and data and then downloads all new copies of everything. In other words, a complete waste for a preference change.
Cheers,
Gary.
Yes, from the beggining it
)
Yes, from the beggining it was very strange with E@H because all other BOINC projects are running smooth. Almost all errors were caused by my kids (shutting down loaded PC).
It is incompatible to cut in half raw power of Matrix HD 7970 to get the same results like other E@H members. The GPU itself has LED indicator for indication of GPU load. It is green colour and it indicates idle but MSI Afterburner indicates 85% load.
I created completly new E@H account without any improvement. I will try to find nVidia GPU to test it with my setup.
RE: Yes, from the beggining
)
Your gpu is doing better then mine is! One of my tasks is here:
2,625.18 504.96 3.32 1,000.00 Binary Radio Pulsar Search (Arecibo, GPU) v1.39 (BRP4G-opencl-ati)
While this is one of yours:
1,797.64 390.56 4.05 1,000.00 Binary Radio Pulsar Search (Arecibo, GPU) v1.39 (BRP4G-opencl-ati)
You can see the same type of unit, both of us have an AMD 7970, but you have an Intel i7 8 core cpu, while I have an AMD 6 core cpu. You are also using a newer version of Boinc then I am, but we appear to be using the same gpu drivers. I have not tweaked mine at all, I just brought it back here yesterday in fact.
I am NOT trying to say mine is something special, it isn't! I am just trying to say that between you and I, yours is doing MUCH better then mine is!!
Are you running one or two WU
)
Are you running one or two WU at once?
RE: You should NEVER
)
In a normal project, perhaps yes, but not at Einstein@Home. As Einstein has the "resend lost tasks" function enabled on the server. This will compare the contents of the client_state.xml file with the contents of the database, to see if all tasks that the server appointed to the hostID are really on the hostID, and if there is a mismatch, to resend the tasks that the database thinks the hostID should have.
So all that a project reset will do here is that you dump all contents of the projects\einstein.phys.uwm.edu\ directory, to then get those same applications and tasks downloaded to your computer again.
Might still be a waste of bandwidth, but other than that, nothing.
The only ways to get new work here are:
1. run the work, upload, report, get new.
2. detach the project. This will delete the whole projects\einstein.phys.uwm.edu\ directory and any entry about the project in the client_state.xml file. So that when you reattached to the project, it has nothing to compare to, and will have to send you new work.
I am running 2 WU at once and
)
I am running 2 WU at once and it takes around 2 hours to crunch them (7000 - 7300 seconds. GPU clock is set @ 550 MHz and memory clock is set @ 825 MHz.
Still something not right
)
Still something not right here.
What motherboard, memory (and settings) are you using? Are you 100% sure you have no temperature / power issues?
You shouldn't have to underclock to that degree just to get stability on a card (with Einstein), that's so (potentially) capable.
If you are convinced that the 'Matrix' is in perfect working order (which I doubt) and the power supply is good (which should be, but maybe not). The problem must lie in the surrounding system components or system config...
Are you able to return the card to the supplier for replacement? The amount of underclock you need to apply really does worry me...
Gavin.