I was referring to Petri's latest App which requires a CUDA 9.2 driver. It Appears he has posted a version that works with the Pascal GPUs aready. I'm still working on the version for the other GPUs.
So, if you have a Pascal GPU, and the CUDA 9.2 driver, check it out. But beware, it's basically untested; https://setiathome.berkeley.edu/forum_thread.php?id=83246&postid=1950582#1950582
I think the one I'm working on will work with the CUDA 9.1 driver, and with the GTX 750 Ti as well as the others.

I was referring to Petri's latest App which requires a CUDA 9.2 driver. It Appears he has posted a version that works with the Pascal GPUs aready. I'm still working on the version for the other GPUs.
So, if you have a Pascal GPU, and the CUDA 9.2 driver, check it out. But beware, it's basically untested; https://setiathome.berkeley.edu/forum_thread.php?id=83246&postid=1950582#1950582
I think the one I'm working on will work with the CUDA 9.1 driver, and with the GTX 750 Ti as well as the others.

Yes. I grabbed Petri's new app. Installed the CUDA 9.2 Toolkit and didn't install the driver. Kept the existing 396.51 driver. Verified CUDA 9.2 installed. Modified app_info for the new app name and promptly threw away all my gpu cache. Each time I replace the special app in the folder, on restart BOINC deletes the app. First error was incorrect CUDA version. Then any subsequent restart is just missing application because BOINC deletes it.

Could use some suggestions about what I am doing wrong. I don't have any app_info examples from way back when I tried the xs2 alpha static linked app.

Question. Should I remove the file references to the the libcudart.so.9.0 and the libcufft.so.9.0 files that are needed in app_info for the dynamically linked zi3v app because those resources are now CUDA 9.2 incorporated in the new static linked setiV0.97.linux_x64_10x0 application?

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

If your CPU does not support avx then DO NOT try to run an app which has been complied using the avx extensions. Doing so would result in either the app not running or if it run trashing loads of tasks.

The reason that the special sauce appears to be returning a lower flops count than expected itthe poor way flops are calculated on gpus.

I have an i7-3770 that supports AVX and processed AVX CPU applications via the Lunatics Beta 6? distro. It had a much higher CPU G flops than another machine currently out there (not mine) that is running 16c/32t AVX motherboard.

It seems unreasonable that his G flops are that low. So I was wondering about installing the Lunatics AVX CPU app on my i7-3770.

I am absolutely impressed by the performance of my GTX 750Ti's on another non-AVX box. I want to install an AVX version of the CUDA90 under Lubuntu on my i7-3770 box.

Tom

---

A proud member of the OFA (Old Farts Association).

I want to install an AVX version of the CUDA90 under Lubuntu on my i7-3770 box.

AVX is a CPU instruction set, CUDA is for general purpose processing on NVidia video cards. They have nothing at all to do with each other.

---

Grant
Darwin NT

. . I have a dumb question. Is it possible to do a non-destructive install of a later Linux OS without trashing everything on a working Linux rig? I have only done installs from scratch with nothing to preserve.

Stephen

?

I have an i7-3770 that supports AVX and processed AVX CPU applications via the Lunatics Beta 6? distro. It had a much higher CPU G flops than another machine currently out there (not mine) that is running 16c/32t AVX motherboard.

It seems unreasonable that his G flops are that low. So I was wondering about installing the Lunatics AVX CPU app on my i7-3770.

I am absolutely impressed by the performance of my GTX 750Ti's on another non-AVX box. I want to install an AVX version of the CUDA90 under Lubuntu on my i7-3770 box.

Tom

. . Firstly, Lunatics is for Windows only, not for Linux. There is/are Linux version/s of the AVX app and yes, a CPU running AVX should have better performance. If in doubt whether or not a CPU supports AVX look at the stderr.txt for a running CPU task, and see what features it lists for that CPU. There are many such as MMX SSE3.0 SSE4.1 etc but AVX should be near the end of the line.

. . Secondly, AVX is a function of CPUs, CUDA is a function on Nvidia GPUs, there is NO relationship between the two.

. . So have a look on Arkayns site for a Linux AVX app.

Stephen

.

. . I have a dumb question. Is it possible to do a non-destructive install of a later Linux OS without trashing everything on a working Linux rig? I have only done installs from scratch with nothing to preserve.

Stephen

?

Is your request about Linux or BOINC? Just zip up your BOINC directory and move it someplace else other than where you want to install the new Linux OS distribution. Then unpack your BOINC zipped file back to its original location and continue on. Should check dependencies are still met on the new Linux installation of course.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

. . I have a dumb question. Is it possible to do a non-destructive install of a later Linux OS without trashing everything on a working Linux rig? I have only done installs from scratch with nothing to preserve.

Stephen

?

Is your request about Linux or BOINC? Just zip up your BOINC directory and move it someplace else other than where you want to install the new Linux OS distribution. Then unpack your BOINC zipped file back to its original location and continue on. Should check dependencies are still met on the new Linux installation of course.

. . Sadly on the machine I want to try it on I made the booboo of installing the repository version of BOINC so zipping it all up and hoping to get it right putting it all back with a different release of Ubuntu is a bit of a worry. The other machine with a nice, user friendly BOINC folder in the home directory is my biggest hitter and I would not like to trash that when feeling my way through this process. It is also running GTX970's and Petri's sneak preview is for Pascal only.

. . Maybe I'll just wait until I can solve the problem getting Linux to install properly on the new rig. Softly, softly ...

Stephen

?

. . Sadly on the machine I want to try it on I made the booboo of installing the repository version of BOINC so zipping it all up and hoping to get it right putting it all back with a different release of Ubuntu is a bit of a worry. The other machine with a nice, user friendly BOINC folder in the home directory is my biggest hitter and I would not like to trash that when feeling my way through this process. It is also running GTX970's and Petri's sneak preview is for Pascal only.

. . Maybe I'll just wait until I can solve the problem getting Linux to install properly on the new rig. Softly, softly ...

Stephen

?

Petri also posted the link to the multiversion 0.97 for sm_50, sm_52, sm 61 and sm_70 type cards like the 970. That is the one I am going to try on my test machine with 970's.

https://setiathome.berkeley.edu/forum_thread.php?id=83246&postid=1950636

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

So, how far along are you with your app_info? Have you tried removing all the Library references and just have the App names, or are you going to try the one I posted and just list the CUDA 9.2 libraries the way I listed the 9.1 libraries? You say you have the 9.2 libraries, so, it might be best to just use a similar app_info to what I posted for now.

So, how far along are you with your app_info? Have you tried removing all the Library references and just have the App names, or are you going to try the one I posted and just list the CUDA 9.2 libraries the way I listed the 9.1 libraries? You say you have the 9.2 libraries, so, it might be best to just use a similar app_info to what I posted for now.

Since I have CUDA9.2 installed, I thought I would go with it and see what happens. I like your naming conventions because they are more descriptive and show the ancestry of the application.

Only a half hour or so until I clear the ghosts from my last attempt and I still think I will work my cache down to only a dozen or so tasks so I don't have to recover so many ghosts if I ef up again. This is what I am going with:

<app_info>
  <app>
     <name>setiathome_v8</name>
  </app>
    <file_info>
      <name>setiathome_x41p_V0.97_x86_64-pc-linux-gnu_cuda92</name>
      <executable/>
    </file_info>
    <file_info>
      <name>libcudart.so.9.2</name>
    </file_info>
    <file_info>
      <name>libcufft.so.9.2</name>
    </file_info>
    <app_version>
      <app_name>setiathome_v8</app_name>
      <platform>x86_64-pc-linux-gnu</platform>
      <version_num>801</version_num>
      <plan_class>cuda90</plan_class>
      <cmdline>-nobs</cmdline>
      <coproc>
        <type>NVIDIA</type>
        <count>1</count>
      </coproc>
      <avg_ncpus>0.1</avg_ncpus>
      <max_ncpus>0.1</max_ncpus>
      <file_ref>
         <file_name>setiathome_x41p_V0.97_x86_64-pc-linux-gnu_cuda92</file_name>
          <main_program/>
      </file_ref>
      <file_ref>
         <file_name>libcudart.so.9.2</file_name>
      </file_ref>
      <file_ref>
         <file_name>libcufft.so.9.2</file_name>
      </file_ref>
    </app_version>
  <app>
     <name>astropulse_v7</name>
  </app>
     <file_info>
       <name>astropulse_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100</name>
        <executable/>
     </file_info>
     <file_info>
       <name>AstroPulse_Kernels_r2751.cl</name>
     </file_info>
     <file_info>
       <name>ap_cmdline_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100.txt</name>
     </file_info>
    <app_version>
      <app_name>astropulse_v7</app_name>
      <platform>x86_64-pc-linux-gnu</platform>
      <version_num>708</version_num>
      <plan_class>opencl_nvidia_100</plan_class>
      <coproc>
        <type>NVIDIA</type>
        <count>1</count>
      </coproc>
      <avg_ncpus>0.1</avg_ncpus>
      <max_ncpus>0.1</max_ncpus>
      <file_ref>
         <file_name>astropulse_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100</file_name>
          <main_program/>
      </file_ref>
      <file_ref>
         <file_name>AstroPulse_Kernels_r2751.cl</file_name>
      </file_ref>
      <file_ref>
         <file_name>ap_cmdline_7.08_x86_64-pc-linux-gnu__opencl_nvidia_100.txt</file_name>
         <open_name>ap_cmdline.txt</open_name>
      </file_ref>
    </app_version>
   <app>
      <name>setiathome_v8</name>
   </app>
      <file_info>
         <name>MBv8_8.22r3711_sse41_x86_64-pc-linux-gnu</name>
         <executable/>
      </file_info>
     <app_version>
     <app_name>setiathome_v8</app_name>
     <platform>x86_64-pc-linux-gnu</platform>
     <version_num>800</version_num>   
      <file_ref>
        <file_name>MBv8_8.22r3711_sse41_x86_64-pc-linux-gnu</file_name>
        <main_program/>
      </file_ref>
    </app_version>
   <app>
      <name>astropulse_v7</name>
   </app>
     <file_info>
       <name>ap_7.05r2728_sse3_linux64</name>
        <executable/>
     </file_info>
    <app_version>
       <app_name>astropulse_v7</app_name>
       <version_num>704</version_num>
       <platform>x86_64-pc-linux-gnu</platform>
       <plan_class></plan_class>
       <file_ref>
         <file_name>ap_7.05r2728_sse3_linux64</file_name>
          <main_program/>
       </file_ref>
    </app_version>
</app_info>

I didn't try and make a symbolic link to the cufft and cudart CUDA 9.2 files so I just copied them to the project directory like the ones included in your CUDA90 special app from CA. The 0.97 application is Petri's multiversion which is said to work better on 970's with their smaller memory. If this test is successful I will try the version made for Pascal on one of my original crunchrs.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

OK, so you haven't tried it yet. I just made a backup of the boinc folder and I'm going to try it with just the App names.
Give me about 10 minutes and I'll tell if it works.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Seems to work. The first task, https://setiathome.berkeley.edu/result.php?resultid=6901725590
Looks like the same run-times as the CUDA 9.1 App I made earlier.
There isn't any CUDA Libraries in the Slot folder, so, you don't need any CUDA libraries other than what is embedded in the App. That means All you need to run the App is the 9.2 version of the CUDA Driver, 396.24, 396.45, or 396.51.

This is what is working for Me. Before you try it, you might want to make a backup of your BOINC folder and turn off Networking just in case yours decides to Ghost more tasks;

  <app>
    <name>setiathome_v8</name>
  </app>
	<file_info>
	   <name>setiV0.97.multi_sm.linux_X86_64_cuda92</name>
	   <executable/>
	</file_info>
	<app_version>
          <app_name>setiathome_v8</app_name>
	  <platform>x86_64-pc-linux-gnu</platform>
          <version_num>801</version_num>
	  <plan_class>cuda90</plan_class>
          <cmdline>-nobs</cmdline>
	    <avg_ncpus>0.1</avg_ncpus>
	    <max_ncpus>0.1</max_ncpus>
	     <coproc>
	       <type>CUDA</type>
	       <count>1.0</count>
	     </coproc>
	<file_ref>
           <file_name>setiV0.97.multi_sm.linux_X86_64_cuda92</file_name>
           <main_program/>
	</file_ref>
     </app_version>

That's just the CUDA section of my app_info.

OK, thanks TBar. Looks good. Just remove the file references to the CUDA libraries and use the app_name. Were the cufft and cudart 9.2 libraries in the project directory? Or did you just let the setiV0.97.multi_sm.linux_X86_64_cuda92 static linked app find the system CUDA resources on its own in /usr/local/cuda-9.2?

[Edit] Helps if you continue reading further. Nevemind.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Looks like Chomsky is running the CUDA_9.2 special application successfully on the simplified app_info file. Running Petri's multiversion application suitable for 970's.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

The New Apps on my machines seem to be working well. Have you changed your Fleet over to the new 9.2 Apps? It would be interesting to see how much it gains.

Now we need to get some more Arecibo tasks and some of those blc02, 03, and 04s at the top of the splitters to see how it handles those. Those low numbered blc task left over at the top of the splitter gave the old V0.96 Mac App a hard time, need to try them again with the new App.

Working on it. All but one to go. I ran out of monitors so need to move a monitor back to what became a headless machine after the latest contest build.

I'm seeing some significant number improvements across the board. Those Arecibo VLARS were cut in half basically from 4 minutes down to 2 minutes. The normal BLC14 and BLC05 seems to have 50-60 seconds shaved off. Sub-minute run_times on a 1080Ti.

Been busy converting so haven't spent much time examining returns yet. No errors to speak of. Need to see if the inconclusive rate changes. I was at less than 2.1% system wide across the fleet on zi3v.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

The New Apps on my machines seem to be working well. Have you changed your Fleet over to the new 9.2 Apps? It would be interesting to see how much it gains.

Now we need to get some more Arecibo tasks and some of those blc02, 03, and 04s at the top of the splitters to see how it handles those. Those low numbered blc task left over at the top of the splitter gave the old V0.96 Mac App a hard time, need to try them again with the new App.

Arecibo VLAR early overflow matched up with the Linux cpu app.
http://=https://setiathome.berkeley.edu/workunit.php?wuid=3101219379
All good.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

Watch your overclocks on memory with the new app. App works the cards harder and you can't get away with the same overclock on memory that you were running on zi3v. Too many 30 pulses found.

---

Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)

. . OK guys, I am about to take a leap and try it out, but before I land on my sword and get blood everywhere ... I would like to know if/where I have gotten it wrong.

. . I am running Ubuntu 14.04 on a Core2 Duo with a GTX1050ti crunching under BOINC 7.2.42 (repository version), currently using zi3v Cuda 80. I have downloaded Nvidia drivers 396.45 and Petri's compilation V0.97.Linux_10x0.

. . I am about to stop BOINC after suspending everything and then use TBar's method to install the new video drivers in recovery mode. Then I will copy the 0.97 app into my BOINC project folder and replace the app name for CUDA80 in app_info.xml with the new app name. Not sure if I can leave the references to the existing library files or if I need to remove them. Then it should just run when I restart BOINC. It all sounds too easy so I feel I must have it wrong ...

Stephen

? ?

Sounds right. Tell us how it went.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones