I bought a new GTX 970 and fired up seti@home overnight. That thing is a beast and I now have several pages of uploaded results. However, only one task was validated. All the other results are either pending validation or came back invalid or inconclusive.

I'm not even sure what that means. Is there a problem with this setup? Should I keep running tasks or not? It doesn't seem to make sense to keep doing them if 99% of them turn out invalid or inconclusive.

I don't recall ever seeing that before with my older GTX 460s. Maybe I just never noticed before.

Any ideas what's going on?

Any ideas what's going on?

For some reason your system is using the CUDA32 application- that is for the oldest Nvidia cards around. Even with your GTX 460 the CUDA50 would have been the best application, and it is certainly the one you need for the GTX970.
I'd suggest running the Lunatics installer, select the appropriate application for your CPU, select CUDA50 for the GPU for MB (not sure which is best for AP- I don't run it) & then see if things improve.

---

Grant
Darwin NT

have a look here
http://setiathome.berkeley.edu/forum_thread.php?id=76479

---

I came down with a bad case of i don't give a crap

have a look here
http://setiathome.berkeley.edu/forum_thread.php?id=76479

Made the link active.

Easy to do, highlight the whole of the url, then hit the URL Button.

Claggy

Thanks! I'm giving that a shot now. I installed via Lunatics installer and then started up BOINC. It looks like it's still using CUDA32, but maybe that's a holdover for this particular task it's working, maybe? I'll see what happens after it finishes the one it's working on. I'm not entirely sure I have it setup correctly.

Thanks again!

Thnx Claggy i'm still a bit woosy just got home from the hospital after surgery hehe

---

I came down with a bad case of i don't give a crap

Okay, I clearly didn't do something correctly. It finished that work unit and started a new one, but it's still cuda32. Hmmm.... I haven't worked directly with cuda before, so I really have no idea what I'm doing. :)

Okay, I clearly didn't do something correctly. It finished that work unit and started a new one, but it's still cuda32. Hmmm.... I haven't worked directly with cuda before, so I really have no idea what I'm doing. :)

ok run the lunatics and read what it says on the installer it gives you options on what to use for nvidia cards

it has an option for pretty much any nvidia and ati and even intel gpus jut check the ones that come close to your cards

EDIT: oh make sure you stop boinc before you install then restart boinc

---

I came down with a bad case of i don't give a crap

Okay, I clearly didn't do something correctly. It finished that work unit and started a new one, but it's still cuda32. Hmmm.... I haven't worked directly with cuda before, so I really have no idea what I'm doing. :)

Sounds like the Lunatics installation didn't take.

Generally I manually stop BOINC from running, check with task manager that the science applications have stopped as well.
Then I do the Lunatics installation, making sure to select myself the best option. Once installed then I restart BOINC.

In the event log I get this,
8/01/2015 17:29:59 | | Starting BOINC client version 7.0.64 for windows_x86_64
8/01/2015 17:29:59 | | log flags: file_xfer, sched_ops, task
8/01/2015 17:29:59 | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
8/01/2015 17:29:59 | | Data directory: C:\ProgramData\BOINC
8/01/2015 17:29:59 | | Running under account Grant
8/01/2015 17:29:59 | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz [Family 6 Model 42 Stepping 7]
8/01/2015 17:29:59 | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 popcnt aes syscall nx lm vmx smx tm2 pbe
8/01/2015 17:29:59 | | OS: Microsoft Windows 7: Professional x64 Edition, Service Pack 1, (06.01.7601.00)
8/01/2015 17:29:59 | | Memory: 3.98 GB physical, 7.97 GB virtual
8/01/2015 17:29:59 | | Disk: 107.13 GB total, 30.53 GB free
8/01/2015 17:29:59 | | Local time is UTC +9 hours
8/01/2015 17:29:59 | | CUDA: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 344.11, CUDA version 6.50, compute capability 5.0, 2048MB, 1913MB available, 2132 GFLOPS peak)
8/01/2015 17:29:59 | | CUDA: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 344.11, CUDA version 6.50, compute capability 5.0, 2048MB, 1910MB available, 2132 GFLOPS peak)
8/01/2015 17:29:59 | | OpenCL: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 344.11, device version OpenCL 1.1 CUDA, 2048MB, 1913MB available, 2132 GFLOPS peak)
8/01/2015 17:29:59 | | OpenCL: NVIDIA GPU 1: GeForce GTX 750 Ti (driver version 344.11, device version OpenCL 1.1 CUDA, 2048MB, 1910MB available, 2132 GFLOPS peak)
8/01/2015 17:29:59 | SETI@home | Found app_info.xml; using anonymous platform

That last line is the one that shows that the Lunatics application has been installed & running.


It may be necessary to reset the project- that stops all current processing, discards the currently downloaded work, and then it will restart, the Scheduler should then re-send the tasks you had previously & they should now be crunched with CUDA50.

---

Grant
Darwin NT

Okay, I clearly didn't do something correctly. It finished that work unit and started a new one, but it's still cuda32.

Don't worry, that's because the Wu was originally Branded as a Stock Cuda32 Wu, the app_info has multiple <app_version> sections in it,
(for Cuda22, Cuda23, Cuda32, Cuda42 and Cuda50 Wu's) all pointing to the app you selected to use,
Check task manager to see the filename of the apps running,
Eventually when you get fresh anonymous platform branded Wu's you'll find that they are branded with the app you selected.

Claggy

I thought that was what I did. When I got to the nvidia page of the installer, I selected cuda50. The default was cuda 32, I believe. Should I select both AstroPulse v7 and Multibeam v7? I did originally. Not sure if that was correct or not. The AstroPulse v7 text mentions that it includes legacy v6.

Okay, so I think there were two issues. First, Avast was actually interfering. I disabled Avast file system shield during install and that seemed to help. Then I restarted BOINC and reset the project. It is now using cuda50.

Many thanks! I really appreciate the help.

Just looked at your account & the work in progress, pending etc. & things are now running as they should.
However it will take a while for things to settle down- all the work presently waiting for validation for your video card was done with CUDA32 & from the looks of it all of them will be invalid. For each invalid result the amount of work you can do each day is reduced. But as you return CUDA50 work & it is validated the amount you can do each day will increase, so there will be a bit of to and fro between the valid & invalid work.
Given the processing ability of your card it shouldn't take too long for valid work to outway the invalids.

---

Grant
Darwin NT

Just looked at your account & the work in progress, pending etc. & things are now running as they should.
However it will take a while for things to settle down- all the work presently waiting for validation for your video card was done with CUDA32 & from the looks of it all of them will be invalid. For each invalid result the amount of work you can do each day is reduced. But as you return CUDA50 work & it is validated the amount you can do each day will increase, so there will be a bit of to and fro between the valid & invalid work.
Given the processing ability of your card it shouldn't take too long for valid work to outway the invalids.

Yeah, this card is pretty amazing. It's blowing through these work units in just a few minutes. I guess gone are the days where it took a few hours to process a work unit!

Thanks again for the help.

But as you return CUDA50 work & it is validated the amount you can do each day will increase, so there will be a bit of to and fro between the valid & invalid work.

No, he's running Anonymous platform now, So as he returns 'SETI@home v7 Anonymous platform (NVIDIA GPU)' Branded work and that valdates, that'll increase the amount of work he's allowed to download,
the Cuda50 app_version is already sat at Max tasks per day '2195', But since he's not running Stock that's not applicable any more:

Application details for host 7119200

Claggy

I had the same problem and this seemed to fix it thanks!

Are the units now custom Lunitics units or is the installer just forcing boinc to only download these units?

Also any idea why Boinc isn't recodnising the card is a 970 and only downloading Cuda50 units? why is it even getting 32 and 42 units?, im thinking there may be a lot of people out there crunching the wrong units and not even realisng.

---

Are the units now custom Lunitics units or is the installer just forcing boinc to only download these units?

All Wu's (of an app type) can potentially run on any app, be it for CPU, GPU, or any other device, there is no such thing as a custom Lunatics unit, or GPU only Wu, etc,
(with the exception that the scheduler doesn't send VLAR Wu's to GPUs at present),
The scheduler will brand Wu's to an app_version, be it Stock 7.00, Cuda22, Cuda23, Cuda32, Cuda42, Cuda50, Anonymous platform CPU, Anonymous platform GPU, etc,
the duration results of those will go towards the APR of that app_version,

Also any idea why Boinc isn't recodnising the card is a 970 and only downloading Cuda50 units? why is it even getting 32 and 42 units?, im thinking there may be a lot of people out there crunching the wrong units and not even realisng.

The scheduler is basically naff at making decisions at what app to send, it isn't hard coded to send work of a particular app_version to a particular GPU,

What the schduler actually does is take in account the APR value, and send work for the fastest app most of the time,
But, the different Angle Ranges of the different Wu's cause the APR to rise and fall, High AR Wu's contain a large portion of Autocorrelation, for GPU apps that causes the APR to fall,
this can cause the preferred app to Change, then there is running Multiple instances, that causes the APR to fall too, which can change the preferred app too,
for Stock use, there's ways around this, you can run a non-prefered app with greater instances to drive it's APR lower, (that can be setup via app_config.xml), hardly ideal,
you can abort non-prefered Wu's to drive it's max task per day down, making the scheduler sent work for your preferred app, again non-ideal, need to combine that with the above,
Or run anonymous platform, there you specify which app you're going to run.

At present your Cuda50 APR is lower than your Cuda42 APR, While your Cuda32 APR is lower still:

Application details for host 7407076

The scheduler operation gets worse when running Stock CPU Astopulse, if your host manages to complete it's 11 validations of an app_version,
It won't necessarily complete the 11 validations of all the other app_versions, meaning it can get locked into running a slow app_version, not ideal at all,
the way round this is to run anonymous platform.

Claggy

Anonymous platform?

so running the CPU along with the GPU and also running multiple work units on the GPU can cause the GPU to get units that will fail?

Pretty much all of the cuda 32 units I got were not validating, I have used the lunitics installer now to force cuda 50 so hopefully that will sort things.

My thread is here if you want to have a look :)

http://setiathome.berkeley.edu/forum_thread.php?id=76511&postid=1627083#1627083

TBH I just want to crunch, the hassle of getting wrong units / setting up the client to do multiple work units to max out my GPU ect is a pain.

---

Anonymous platform?

http://boinc.berkeley.edu/wiki/Anonymous_platform

BOINC applications are native-mode programs, so different versions are required for each platform (a "platform" is the combination of an operating system and a CPU architecture: e.g., Linux/Intel32). Each BOINC-based project has application versions for one or more platforms. When the BOINC client requests work from the project's server, the client tells the server its platform, and the server gives it the appropriate version.

This addresses the needs of most BOINC participants, but it's inadequate if:
your computers have platforms not supported by BOINC or by the project;
for security reasons, you want to only run executables you have compiled yourself;
you want to optimize applications for particular architectures.

To handle these cases, BOINC offers a mechanism called anonymous platform. This lets you build applications yourself, or obtain them from a third party, rather than getting them from the project server. This can be used only for projects that make their source code available. As an example, see the instructions for SETI@home.

so running the CPU along with the GPU and also running multiple work units on the GPU can cause the GPU to get units that will fail?

Pretty much all of the cuda 32 units I got were not validating, I have used the lunitics installer now to force cuda 50 so hopefully that will sort things.

My thread is here if you want to have a look :)

http://setiathome.berkeley.edu/forum_thread.php?id=76511&postid=1627083#1627083

TBH I just want to crunch, the hassle of getting wrong units / setting up the client to do multiple work units to max out my GPU ect is a pain.

All the Cuda32, Cuda42 and Cuda50 apps should run and validate on all GPUs, subject to driver requirements,
If they're not, either one or more Cuda versions has a Bug, or GPU isn't producing correct results at a particular GPU clock, or GPU memory Clock, etc.

Claggy

Thanks, it only seems to be this machine with this 970 that was failing, can you have a quick look?

When this 970 was in my other machine (Media_box) it seemed to be fine and looking at the stats it only seemed to grab Cuda_50 units?

---