Hello!

Finally I realized that because of the errors occur in the GPU countable applications.
To blame Firefox. If you use it the job often fails, but if it does not run, and the use of Chrome is no problem.

Tell me, please, can somebody faced with similar and what can help.

Link to host job http://www.gpugrid.net/results.php?hostid=170640

Did the development team can offer what is the debug version of the job and together we will be able to fix the error?

P.S. I do not want to stop participating in the project because of errors :(

---

I'm no expert on GPUGrid but I think there is no connection between running it and which Web browser you are using.

I suspect your problem lies with your GPU. Is it running too hot? Is it overclocked too much? Are the contacts not perfectly clean?

Also, are you running the latest versions of BOINC and the Nvidia driver?

Hello!

Finally I realized that because of the errors occur in the GPU countable applications.
To blame Firefox. If you use it the job often fails, but if it does not run, and the use of Chrome is no problem.

Tell me, please, can somebody faced with similar and what can help.

Link to host job http://www.gpugrid.net/results.php?hostid=170640

Did the development team can offer what is the debug version of the job and together we will be able to fix the error?

P.S. I do not want to stop participating in the project because of errors :(

You are getting "Simulation Has Become Unstable" messages in your output files which usually means you or factory have overclocked the GPU or Memory or both too much. Reset it to stock speeds and see how it goes.

Finally I realized that because of the errors occur in the GPU countable applications.
To blame Firefox. If you use it the job often fails, but if it does not run, and the use of Chrome is no problem.

Both browsers use graphics acceleration for displaying contents of webpages; dynamic content (video, flash, java, html5, vr or 3d engines) could push your card over its limits.

Tell me, please, can somebody faced with similar and what can help.

Link to host job http://www.gpugrid.net/results.php?hostid=170640

Did the development team can offer what is the debug version of the job and together we will be able to fix the error?

P.S. I do not want to stop participating in the project because of errors :(

You are getting "Simulation Has Become Unstable" messages in your output files which usually means you or factory have overclocked the GPU or Memory or both too much. Reset it to stock speeds and see how it goes.

According to NVidia specifications, the GPU of this GTX 660 is overclocked:

GPU Engine Specs:
CUDA Cores:              960
Base Clock:              980 MHz
Boost Clock:            1033 MHz
Texture Fill Rate:      78.4 billion/sec
Memory Specs:
Memory Clock:            6.0 Gbps
Standard Memory Config: 2048 MB
GDDR5Memory Interface
Memory Interface Width   192 bit GDDR5
Memory Bandwidth       144.2 GB/sec

Here's an excerpt from the stderr.txt of a successful task:

# GPU [GeForce GTX 660] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 0	:
#	Name		: GeForce GTX 660
#	ECC		: Disabled
#	Global mem	: 2048MB
#	Capability	: 3.0
#	PCI ID		: 0000:01:00.0
#	Device clock	: 1097MHz
#	Memory clock	: 3004MHz
#	Memory width	: 192bit
#	Driver version	: r355_00 : 35582
# GPU 0 : 69C
# GPU 0 : 70C
# Simulation unstable. Flag 10 value 773
# The simulation has become unstable. Terminating to avoid lock-up
# The simulation has become unstable. Terminating to avoid lock-up (2)
# Attempting restart (step 2765000)

The GPU clock of this card is 64 MHz higher than the boost clock of the standard, so it should be "underclocked" of its factory settings to get NVidia's original settings. Use MSI Afterburner, or NVidia Inspector to do that.

Thank you all!

You are getting "Simulation Has Become Unstable" messages in your output files which usually means you or factory have overclocked the GPU or Memory or both too much. Reset it to stock speeds and see how it goes.

The device is sold with factory overclocked ( http://www.gigabyte.com/products/product-page.aspx?pid=4361#sp ).

In addition the project Einstain at home works fine (with the settings in the three parallel working tasks) on the graphics card it is loaded by the kernel of 95% and the memory controller to 77%. Ie even more, while the temperature 3-4 degrees higher than GPUgrid (75-76 degrees).

Both browsers use graphics acceleration for displaying contents of webpages; dynamic content (video, flash, java, html5, vr or 3d engines) could push your card over its limits.

At that point chrome running, there are no problems. However, when running firefox fall off the job.

The GPU clock of this card is 64 MHz higher than the boost clock of the standard, so it should be "underclocked" of its factory settings to get NVidia's original settings. Use MSI Afterburner, or NVidia Inspector to do that.

Well, using NVidia Inspector reduced the core frequency up to 1032 MHz, are now testing.

With NVidia Inspector I found that the driver has a profile for Chrome. For Firefox there is no profile. it is possible that the problem is this.

---

It seems that the reduction of the nominal frequency to not help :( The truth is not clear why in the log says that the frequency is still the same, but the utility NVidia Inspector and GPU-Z shows that the video card running at the nominal frequencies. What should I do to reflash the BIOS? :( I have all the same strong feeling that the blame is Firefox. Really no one has similar problems in the combination of Firefox + GPUGrid? Maybe but I exclusively use to BOINC background on machines with the users, rather than dedicated?

---

It seems that the reduction of the nominal frequency to not help :(

I would reduce the GPU clock a little more, and also the RAM frequency (by 100MHz). At some point it will get stable.
Your GPU is a bit hot, so perhaps cleaning the fans and heatsinks with compressed air would help.

The truth is not clear why in the log says that the frequency is still the same, but the utility NVidia Inspector and GPU-Z shows that the video card running at the nominal frequencies. What should I do to reflash the BIOS? :(

That's simply because this info comes from the BIOS. You don't have to reflash the BIOS for testing.

I have all the same strong feeling that the blame is Firefox. Really no one has similar problems in the combination of Firefox + GPUGrid?

I use Chrome most of the time, but I haven't noticed that Firefox could make GPUGrid tasks to crash.

I use Firefox all the time with GPUGrid and don't have any problems. Having several "users" using the same computer with user switching maybe the problem. Why don't you try only having one user on the machine and see if a WU completes ok. You could do that during the night.

The GPUGrid work units tend to be very demanding on the GPU. Using the GPU both for crunching and driving your display is not recommended. One way or another, the two duties will get in the way of one-another.

You may not have noticed these errors with Einstein, simply because those tasks are not so demanding on the GPU as the GPUGrid ones.

If you are using your crunching PC for "normal" desktop use, I propose that you do one of the following:

Do not run GPU-demanding programs (games, video, etc)
Suspend GPUGrid tasks when running other GPU-demanding programs
Use another GPU for driving your display, one that will not be crunching GPU tasks.

Personally, I am following the last proposition from above, I am using the GPU embedded in my Intel CPU to drive the display. I have excluded it from doing any Boinc work, which also allows me to connect to my PC remotely. Quite handy!

---

My computers have to do all the usual tasks asscociated with computers and still run GPUGrid 24/7 with no problem including watching HD video streams etc.

The only caveat is if you are 3D gaming you must suspend/turn off BOINC because as every gamer knows you need as much CPU/GPU power as you can get for 3D games. The same applies for Video Rendering.

The more likely reason for these errors once everything else has been discounted is "User Switching".

Thanks all!!

I would reduce the GPU clock a little more, and also the RAM frequency (by 100MHz). At some point it will get stable.
Your GPU is a bit hot, so perhaps cleaning the fans and heatsinks with compressed air would help.

Thank, relatively high temperature caused by a fairly compact case of the machine, the ventilation system is clean and everything is fine with her.


Maybe I found a problem. Firefox takes up memory on the graphics card 3 times (> 500 MB) more than Chrome. It is likely, very likely that the problem is just that. Now try to reset the video card to the factory and to experiment with running Firefox and control memory. It also would explain why Einstain, PrimeGrid work without errors, they require significantly less memory than GPUGrid.

Also, I updated Firefox to Firefox ESR in order to exclude the effect of different experiments in the browser to the test.

---

He looked at Firefox ESR - depression, there is even obvious mistakes functionality is not correct. Updated back to 41.0.1. :)

Now waiting for the GPUGrid jobs will ...

---

Good day! It seems to be completely corrected the problem of reducing the memory frequency on 100 MHz.I put to autostart NVidia Inspector through a task scheduler, that he would upon login by reduced memory frequency on 100 MHz from the nominal, one month finally shall see to result and a maybe reflash the BIOS, and may be kept as a it is now.

---

:) I have 3 GPU's. One of them needs underclocked for complete compatibility, while the other 2 can be overclocked higher than factory clocks while maintaining complete compatibility (tested via overnight run of Heaven 4.0 and overnight run of GPUGrid work units).

I have a .bat file, set to startup by placing a shortcut in:
C:\ProgramData\Microsoft\Windows\Start Menu\Programs\StartUp

The .bat file sets: gpu 0 gets power set to max 123 with clock +32, gpu 1 gets power set to max 140 with clock -46, gpu 2 gets power set to max 110 with temp target 80 and clock +8

The .bat file contents are:
"c:\Program Files\NVIDIA Inspector\nvidiaInspector.exe" -setPowerTarget:0,123 -setBaseClockOffset:0,0,32 -setPowerTarget:1,140 -setBaseClockOffset:1,0,-46 -setPowerTarget:2,110 -setTempTarget:2,1,80 -setBaseClockOffset:2,0,8

You can always run the following to get the list of command parameters:
"c:\Program Files\NVIDIA Inspector\nvidiaInspector.exe" /h

Sure, every time I restart Windows, I have to say "Yes" to the User Account Control popup, but it sets the frequencies then it's done. I prefer this .bat file, instead of Precision X (an awful program lately) or MSI Afterburner (which I use, but I only use it for fan control and monitoring, not for setting clocks).

Fun.

You can always run the following to get the list of command parameters:
"c:\Program Files\NVIDIA Inspector\nvidiaInspector.exe" /h

Sure, every time I restart Windows, I have to say "Yes" to the User Account Control popup, but it sets the frequencies then it's done.

If you want to avoid pressing "Yes" every time you restart Windows, you should make a scheduled task to start "c:\Program Files\NVIDIA Inspector\nvidiaInspector.exe" /h every time a user logs on, and set the task properties to "Run at highest privileges".

Interesting idea - I'll try that later. Thanks!