The App is in the All-In-One package at the same location http://www.arkayn.us/lunatics/BOINC.7z
Make sure to read the README_x41p_V0.98.txt file in BOINC/projects/setiathome.berkeley.edu/docs.
The 10.2 App will need driver 440.33 or Higher, Ubuntu 15.04 or Higher, and Pascal or Higher GPU.
The 440.33 driver is available in the CUDA 10.2 ToolKit, you don't need to install the ToolKit, just the driver.
In My experience, the easiest way to install the nVidia Downloaded driver on a fresh Ubuntu install is to first install the highest Repository nVidia driver from Additional Drivers. Then download the 10.2 ToolKit from nVidia, place it in your Home folder, and make sure the Execute bit is set. Next reboot into Recovery Mode and enable Networking. Switch to Root mode, cd to your home folder, then remove the Repository driver by running, apt remove --purge nvidia* and then run, apt autoremove. Next install the Driver by running the installer and setting the options to just install the driver, ./cuda_10.2.89_440.33.01_linux.run
Depending on your HDD it may take a while to read and launch the installer, it's not long if using an SSD. Once the installer is finished, reboot.
It seems to work better the first time if you use the driver installer from the ToolKit, after that, running the downloaded standalone driver should work without any trouble. If the driver fails to work, simply boot back to Recovery mode and remove the downloaded driver by running, nvidia-uninstall
Have fun.

Thanks for the new app TBar. I had already been updated to 440.36 drivers by the ppa and didn't have to do anything other than edit the new app into app_info.

---

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

A proud member of the OFA (Old Farts Association)

Nice. I just recently started using the cuda apps and it's amazing. Kinda hurts to know how much power I wasted previously though. :-)

I have some noob questions. Do I need to change plan_class to cuda102 if I want to use the 10.2 app? From searching, I kinda get what it means, but it's not clear to me if cuda102 is a valid value or whether I need to change it at first place.

I am also a bit curious why ncpus is set to 0.1 when -nobs is set. The readme always recommends leaving enough CPU headroom, and to me the easiest way seems to be setting ncpus to 1 instead of messing with the "at most xx% CPUs" setting. Setting ncpus to 1 seems to be working well for me, though the core was definitely quite idle when -nobs was not set. I wonder if it's just a personal preference or I am misunderstanding something.

---

The README_x41p_V0.98.txt says nothing about changing the plan class. It says, "If you have a CUDA 10.2 driver, and Pascal or Higher GPU, you can use the CUDA102 App, change the app_info.xml to name the CUDA 10.2 App in the Two locations, <name> & <file_name>" If you change Anything Else you run the risk of Trashing any existing tasks you have. If you Don't have any Existing tasks, then it's safe to change the plan class although it will accomplish absolutely nothing.

All you accomplish by setting ncpus to 1 is Lock your number of tasks to the number of cores you have. That means you can't run more than Two GPU tasks on a Dual core CPU although you can easily run 4 GPUs on a Dual core CPU. It means you can only run 4 GPUs on a 4 core CPU although you can easily run 8 GPUs on a 4 core CPU. If you leave it set to 0.1 you don't have that problem, and running -nobs will help when running more than 2 GPUs. If you are running CPU tasks as well, use the BOINC Preferences to limit your CPU use, that's why the Use at most __ % of the CPUs setting is there. That's how it's possible to run 3 GPUs on a Dual core CPU, 5 GPUs on a quad core CPU, and 14 GPUs on an 8 core CPU. You can't do that if you set ncpus to 1, that's why the Original nVidia CUDA Developers set the ncpus to 0.04. They understood ncpus = 1 artificially limited the number of GPU tasks you could run.

CUDA 10.2 drivers have been in the Ubuntu drivers PPA since driver 440.26 (added mid October). It is now up to date with driver 440.36. The PPA is far easier to use on Ubuntu systems.

If you’re already using the PPA for drivers, all you need to do is:

sudo apt update
sudo apt purge *nvidia*
sudo apt install nvidia-driver-440

Reboot.

---

Seti@Home classic workunits: 29,492 CPU time: 134,419 hours

The App is in the All-In-One package at the same location http://www.arkayn.us/lunatics/BOINC.7z
Make sure to read the README_x41p_V0.98.txt file in BOINC/projects/setiathome.berkeley.edu/docs.
The 10.2 App will need driver 440.33 or Higher, Ubuntu 15.04 or Higher, and Pascal or Higher GPU.
The 440.33 driver is available in the CUDA 10.2 ToolKit, you don't need to install the ToolKit, just the driver.
In My experience, the easiest way to install the nVidia Downloaded driver on a fresh Ubuntu install is to first install the highest Repository nVidia driver from Additional Drivers. Then download the 10.2 ToolKit from nVidia, place it in your Home folder, and make sure the Execute bit is set. Next reboot into Recovery Mode and enable Networking. Switch to Root mode, cd to your home folder, then remove the Repository driver by running, apt remove --purge nvidia* and then run, apt autoremove. Next install the Driver by running the installer and setting the options to just install the driver, ./cuda_10.2.89_440.33.01_linux.run
Depending on your HDD it may take a while to read and launch the installer, it's not long if using an SSD. Once the installer is finished, reboot.
It seems to work better the first time if you use the driver installer from the ToolKit, after that, running the downloaded standalone driver should work without any trouble. If the driver fails to work, simply boot back to Recovery mode and remove the downloaded driver by running, nvidia-uninstall
Have fun.

Yea! Just read this. Would have saved me the trouble of building it!!

Yea! Just read this. Would have saved me the trouble of building it!!

True, but at least you know how to build the special app. Good experience for your attempt at the Windows version. Good luck!

---

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

A proud member of the OFA (Old Farts Association)

The README_x41p_V0.98.txt says nothing about changing the plan class. It says, "If you have a CUDA 10.2 driver, and Pascal or Higher GPU, you can use the CUDA102 App, change the app_info.xml to name the CUDA 10.2 App in the Two locations, <name> & <file_name>" If you change Anything Else you run the risk of Trashing any existing tasks you have. If you Don't have any Existing tasks, then it's safe to change the plan class although it will accomplish absolutely nothing.

All you accomplish by setting ncpus to 1 is Lock your number of tasks to the number of cores you have. That means you can't run more than Two GPU tasks on a Dual core CPU although you can easily run 4 GPUs on a Dual core CPU. It means you can only run 4 GPUs on a 4 core CPU although you can easily run 8 GPUs on a 4 core CPU. If you leave it set to 0.1 you don't have that problem, and running -nobs will help when running more than 2 GPUs. If you are running CPU tasks as well, use the BOINC Preferences to limit your CPU use, that's why the Use at most __ % of the CPUs setting is there. That's how it's possible to run 3 GPUs on a Dual core CPU, 5 GPUs on a quad core CPU, and 14 GPUs on an 8 core CPU. You can't do that if you set ncpus to 1, that's why the Original nVidia CUDA Developers set the ncpus to 0.04. They understood ncpus = 1 artificially limited the number of GPU tasks you could run.

Both make sense. Thanks for the explanation. I have more cores than GPUs on all of my rigs, and I didn't think of the need to accommodate the other way around.

---

Does the 10.2 has the support for the mutex?

---

answered already in GPUUG

---

Ran benchmark test on 101 and 102 along with CPU comparison using that test data at
http://www.arkayn.us/lunatics/Test_WUs.7z

I put results into an excel book

https://stateson.net/images/cuda_101_102.xlsx

see very little difference between 10.1 and 10.2

Not sure what any of it means!

Ran benchmark test on 101 and 102 along with CPU comparison using that test data at
http://www.arkayn.us/lunatics/Test_WUs.7z

I put results into an excel book

https://stateson.net/images/cuda_101_102.xlsx

see very little difference between 10.1 and 10.2

Not sure what any of it means!

I didn't expect any difference. The apps only make use of the code possibilities of when Petri built the code. Most of the new capabilities of the CUDA 10.2 is in the realm of AI, machine learning and such. Didn't see any changes in the basic FFT libraries that possibly would have benefitted the app. Only if Petri revisited the code or the existing code can make use of the same calls to the API with the improved routines.

https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

Anyone considering moving to this new app, heed this part of the release notes. Maxwell users take notice.

Support for the following compute capabilities are deprecated in CUDA 10.2. Note that support for these compute capabilities may be removed in a future release of CUDA.

sm_35 (Kepler)
sm_37 (Kepler)
sm_50 (Maxwell)

---

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

A proud member of the OFA (Old Farts Association)

Anyone considering moving to this new app, heed this part of the release notes. Maxwell users take notice.

Keppler, OK. But Maxwell? Already?? It's only just over 5 years old (5 years already! Time isn't just flying, it's flying faster each year).

---

Grant
Darwin NT

Ran benchmark test on 101 and 102 along with CPU comparison using that test data at
http://www.arkayn.us/lunatics/Test_WUs.7z

I put results into an excel book

https://stateson.net/images/cuda_101_102.xlsx

see very little difference between 10.1 and 10.2

Not sure what any of it means!

Kinda hard to tell what's going on in that format. The one you should be concerned with is the one that contains "Weakly similar". That means the results didn't match. It would help if you ran that task in the Lunatics Benchmark App with a CPU App in REF_APPs and the GPU Apps in APPS. The Terminal readout will be much easier to understand and identify the App that doesn't match the CPU App. I'm running it on another machine right now, but the CPU tasks will take a little while.

As far as the Release Notes, this is All you need to read, "Note that support for these compute capabilities may be removed in a future release of CUDA." That means Maxwell will still work with 10.2 , but MAY be removed on Newer ToolKits. nVidia has been complaining about Clang 6.0 since ToolKit 10.0, but, it still works with ToolKit 10.2 even though it has been listed as 'deprecated' for some time now.
That doesn't matter for this CUDA 10.2 App though, if you Read the First Post you will see, "The 10.2 App will need driver 440.33 or Higher, Ubuntu 15.04 or Higher, and Pascal or Higher GPU." Maxwell users can use the CUDA 9.0 App, it works just about as well on Maxwell.

Kinda hard to tell what's going on in that format. The one you should be concerned with is the one that contains "Weakly similar". That means the results didn't match. It would help if you ran that task in the Lunatics Benchmark App with a CPU App in REF_APPs and the GPU Apps in APPS. The Terminal readout will be much easier to understand and identify the App that doesn't match the CPU App. I'm running it on another machine right now, but the CPU tasks will take a little while.

Thought I did that ..

jstateson@h110btc:~/Projects/benchMT$ ls
APPS_CPU  APPS_GPU  APPS_REF  BenchCFG  benchMT  docs  LICENSE  README.md  rescmpv5  rescmpv5_l  runtest.sh  testData  workdir  WU_std_signal  WU_std_signal_bu  WU_test
jstateson@h110btc:~/Projects/benchMT$ ls -l APPS_REF/
total 4504
-rw-rw-r-- 1 jstateson jstateson    1254 Dec  7 11:41 AUTHORS.txt
-rw-rw-r-- 1 jstateson jstateson     401 Dec  7 11:41 README.md
-rwxrwxr-x 1 jstateson jstateson 4588648 Dec  7 11:41 ref-cpu.setiathome_8.00_x86_64-pc-linux-gnu
drwxrwxr-x 2 jstateson jstateson   12288 Dec  7 15:49 REF_RESULTS
jstateson@h110btc:~/Projects/benchMT$ ls -l APPS_GPU/
total 418312
-rw-rw-r-- 1 jstateson jstateson      1254 Dec  7 11:41 AUTHORS.txt
-rw-rw-r-- 1 jstateson jstateson    253299 Dec  7 11:41 MultiBeam_Kernels_r3584.cl
-rw-rw-r-- 1 jstateson jstateson       401 Dec  7 11:41 README.md
-rw-rw-r-- 1 jstateson jstateson     13292 Dec  7 11:41 ReadMe_MultiBeam_OpenCL.txt
-rwxrwxr-x 1 jstateson jstateson   4409256 Dec  7 11:41 setiathome_8.22_x86_64-pc-linux-gnu__opencl_ati5_sah
-rwxrwxr-x 1 jstateson jstateson   4421576 Dec  7 11:41 setiathome_8.22_x86_64-pc-linux-gnu__opencl_ati5_SoG
-rwxrwxr-x 1 jstateson jstateson   4421576 Dec  7 11:41 setiathome_8.22_x86_64-pc-linux-gnu__opencl_nvidia_SoG
-rwxr-xr-x 1 root      root      194222696 Dec  7 11:44 setiathome_x41p_V0.98b1_x86_64-pc-linux-gnu_cuda101
-rwxr-xr-x 1 root      root      220581672 Dec  7 11:45 setiathome_x41p_V0.98b1_x86_64-pc-linux-gnu_cuda102
jstateson@h110btc:~/Projects/benchMT$

I set the config for only 101 and then for 102. The CPU app did not execute a second time which was nice. I copied the lunatic data into the WU_std_signals and ran

./benchMT --boinc_home /usr/bin --num_repetitions 1 --max_gpus 1 --gpu_devices 0 --std_signals

@ TBar

I asked before but not sure if the answer was right, since was you the one who build the new 10.2 version, could you tell me if the code used has the mutex or not capacity?

Thanks in advance

---

The App is in the All-In-One package at the same location http://www.arkayn.us/lunatics/BOINC.7z
Make sure to read the README_x41p_V0.98.txt file in BOINC/projects/setiathome.berkeley.edu/docs.
The 10.2 App will need driver 440.33 or Higher, Ubuntu 15.04 or Higher, and Pascal or Higher GPU.
The 440.33 driver is available in the CUDA 10.2 ToolKit, you don't need to install the ToolKit, just the driver.
In My experience, the easiest way to install the nVidia Downloaded driver on a fresh Ubuntu install is to first install the highest Repository nVidia driver from Additional Drivers. Then download the 10.2 ToolKit from nVidia, place it in your Home folder, and make sure the Execute bit is set. Next reboot into Recovery Mode and enable Networking. Switch to Root mode, cd to your home folder, then remove the Repository driver by running, apt remove --purge nvidia* and then run, apt autoremove. Next install the Driver by running the installer and setting the options to just install the driver, ./cuda_10.2.89_440.33.01_linux.run
Depending on your HDD it may take a while to read and launch the installer, it's not long if using an SSD. Once the installer is finished, reboot.
It seems to work better the first time if you use the driver installer from the ToolKit, after that, running the downloaded standalone driver should work without any trouble. If the driver fails to work, simply boot back to Recovery mode and remove the downloaded driver by running, nvidia-uninstall
Have fun.

Yea! Just read this. Would have saved me the trouble of building it!!

Here it is. You can build the latest and port it to any platform.
https://drive.google.com/open?id=1ydIV4gv86_OERo6zF2qP7LNKVTMsN9BQ

---

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

The App referenced in this thread is setiathome_x41p_V0.98b1_x86_64-pc-linux-gnu_cuda102, it doesn't contain Any Mutex code as that App would be named V0.99b1. App V098b1 is Cross-Platform, and allows users to run other projects, neither of those will work with 0.99b1. For Now I will just concentrate on 0.98b1 as that version is most likely to end up on the SETI Server at some point.

Kinda hard to tell what's going on in that format. The one you should be concerned with is the one that contains "Weakly similar". That means the results didn't match. It would help if you ran that task in the Lunatics Benchmark App with a CPU App in REF_APPs and the GPU Apps in APPS. The Terminal readout will be much easier to understand and identify the App that doesn't match the CPU App. I'm running it on another machine right now, but the CPU tasks will take a little while.

Thought I did that ..

jstateson@h110btc:~/Projects/benchMT$ ls
APPS_CPU  APPS_GPU  APPS_REF  BenchCFG  benchMT  docs  LICENSE  README.md  rescmpv5  rescmpv5_l  runtest.sh  testData  workdir  WU_std_signal  WU_std_signal_bu  WU_test
jstateson@h110btc:~/Projects/benchMT$ ls -l APPS_REF/
total 4504
-rw-rw-r-- 1 jstateson jstateson    1254 Dec  7 11:41 AUTHORS.txt
-rw-rw-r-- 1 jstateson jstateson     401 Dec  7 11:41 README.md
-rwxrwxr-x 1 jstateson jstateson 4588648 Dec  7 11:41 ref-cpu.setiathome_8.00_x86_64-pc-linux-gnu
drwxrwxr-x 2 jstateson jstateson   12288 Dec  7 15:49 REF_RESULTS
jstateson@h110btc:~/Projects/benchMT$ ls -l APPS_GPU/
total 418312
-rw-rw-r-- 1 jstateson jstateson      1254 Dec  7 11:41 AUTHORS.txt
-rw-rw-r-- 1 jstateson jstateson    253299 Dec  7 11:41 MultiBeam_Kernels_r3584.cl
-rw-rw-r-- 1 jstateson jstateson       401 Dec  7 11:41 README.md
-rw-rw-r-- 1 jstateson jstateson     13292 Dec  7 11:41 ReadMe_MultiBeam_OpenCL.txt
-rwxrwxr-x 1 jstateson jstateson   4409256 Dec  7 11:41 setiathome_8.22_x86_64-pc-linux-gnu__opencl_ati5_sah
-rwxrwxr-x 1 jstateson jstateson   4421576 Dec  7 11:41 setiathome_8.22_x86_64-pc-linux-gnu__opencl_ati5_SoG
-rwxrwxr-x 1 jstateson jstateson   4421576 Dec  7 11:41 setiathome_8.22_x86_64-pc-linux-gnu__opencl_nvidia_SoG
-rwxr-xr-x 1 root      root      194222696 Dec  7 11:44 setiathome_x41p_V0.98b1_x86_64-pc-linux-gnu_cuda101
-rwxr-xr-x 1 root      root      220581672 Dec  7 11:45 setiathome_x41p_V0.98b1_x86_64-pc-linux-gnu_cuda102
jstateson@h110btc:~/Projects/benchMT$

I set the config for only 101 and then for 102. The CPU app did not execute a second time which was nice. I copied the lunatic data into the WU_std_signals and ran

./benchMT --boinc_home /usr/bin --num_repetitions 1 --max_gpus 1 --gpu_devices 0 --std_signals

Let's see. From your spreadsheet it seems Both Apps failed on the same task? That's the best I can make out of it. My guess would be that task was blc14_2bit_guppi_58692_02937_HIP79792_0121.10280.409.21.44.21.vlar.wu? I got the same results, both 101 & 102 failed with that task. You can easily see that with the printout from the Terminal;

---------------------------------------------------
Done with blc14_2bit_guppi_58691_83520_HIP79781_0103.8969.0.22.45.117.vlar.wu.
Current WU: blc14_2bit_guppi_58692_02937_HIP79792_0121.10280.409.21.44.21.vlar.wu
---------------------------------------------------
Skipping default app MBv8_8.22r3605_avx2_x86_64-apple-darwin, displaying saved result(s)
Elapsed Time: ………………………………… 1794 seconds
---------------------------------------------------
Running app with command : setiathome_x41p_v0.98b1_x86_64-apple-darwin_cuda101 -nobs -device 0
      223.55 real       201.49 user        16.82 sys
Elapsed Time : ……………………………… 223 seconds
Speed compared to default : 804 %
-----------------
Comparing results
                ------------- R1:R2 ------------     ------------- R2:R1 ------------
                Exact  Super  Tight  Good    Bad     Exact  Super  Tight  Good    Bad
        Spike      0      0      0      0      0        0      0      0      0      0
     Autocorr      0      1      1      1      0        0      1      1      1      0
     Gaussian      0      0      0      0      0        0      0      0      0      0
        Pulse      0      4      4      4      1        0      4      4      4      0
      Triplet      0      1      1      1      0        0      1      1      1      0
   Best Spike      0      1      1      1      0        0      1      1      1      0
Best Autocorr      0      1      1      1      0        0      1      1      1      0
Best Gaussian      1      1      1      1      0        1      1      1      1      0
   Best Pulse      0      1      1      1      0        0      1      1      1      0
 Best Triplet      0      1      1      1      0        0      1      1      1      0
                ----   ----   ----   ----   ----     ----   ----   ----   ----   ----
                   1     11     11     11      1        1     11     11     11      0

Unmatched signal(s) in R1 at line(s) 370
For R1:R2 matched signals only, Q= 99.75%
Result      : Weakly similar.
---------------------------------------------------
Running app with command : setiathome_x41p_v0.98b1_x86_64-apple-darwin_cuda102 -nobs -device 0
      226.30 real       205.43 user        15.59 sys
Elapsed Time : ……………………………… 226 seconds
Speed compared to default : 793 %
-----------------
Comparing results
                ------------- R1:R2 ------------     ------------- R2:R1 ------------
                Exact  Super  Tight  Good    Bad     Exact  Super  Tight  Good    Bad
        Spike      0      0      0      0      0        0      0      0      0      0
     Autocorr      0      1      1      1      0        0      1      1      1      0
     Gaussian      0      0      0      0      0        0      0      0      0      0
        Pulse      0      4      4      4      1        0      4      4      4      0
      Triplet      0      1      1      1      0        0      1      1      1      0
   Best Spike      0      1      1      1      0        0      1      1      1      0
Best Autocorr      0      1      1      1      0        0      1      1      1      0
Best Gaussian      1      1      1      1      0        1      1      1      1      0
   Best Pulse      0      1      1      1      0        0      1      1      1      0
 Best Triplet      0      1      1      1      0        0      1      1      1      0
                ----   ----   ----   ----   ----     ----   ----   ----   ----   ----
                   1     11     11     11      1        1     11     11     11      0

Unmatched signal(s) in R1 at line(s) 370
For R1:R2 matched signals only, Q= 99.75%
Result      : Weakly similar.
---------------------------------------------------
Done with blc14_2bit_guppi_58692_02937_HIP79792_0121.10280.409.21.44.21.vlar.wu.

All the other tasks passed using version 102 on the Mac. I'm still waiting on the 2nd Linux machine.

Kinda hard to tell what's going on in that format. The one you should be concerned with is the one that contains "Weakly similar". That means the results didn't match. It would help if you ran that task in the Lunatics Benchmark App with a CPU App in REF_APPs and the GPU Apps in APPS. The Terminal readout will be much easier to understand and identify the App that doesn't match the CPU App. I'm running it on another machine right now, but the CPU tasks will take a little while.

Thought I did that ..

TBar is asking you to use the Lunatics bench application that he is familiar with.

http://lunatics.kwsn.info/index.php?action=downloads;sa=downfile&id=504
http://lunatics.kwsn.info/index.php?action=downloads;sa=downfile&id=502

---

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

A proud member of the OFA (Old Farts Association)