You are running the CUDA9.0 version. Maybe it´s better if you try to run with: setiathome v8 enhanced x41p_V0.97b2, Cuda 9.20 special who was the latest avaliable for the CUDA 9 builds.

If you can´t find it PM and i will send it for you.

I have this one: "setiathome_x41p_V0.97b2_Linux-Pascal+_cuda92"

Will it run with 396 driver ?

From Ian & Steve post:

CUDA90/91 i believe can use the the normal repository driver (version 380 something?)
CUDA92 will need at least nvidia driver 396 or higher. 
CUDA100 will need at least nvidia driver 410 or higher. 

So i believe yes.

---

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND                                                                                                      
 5996 r         39  19   52788  35956   4536 R  99,7  0,4  22:08.57 MBv8_8.22r3711_                                                                                              
 5939 r         39  19   52756  35928   4536 R  97,7  0,4  35:56.98 MBv8_8.22r3711_                                                                                              
 5937 r         39  19   53264  43464   4536 R  95,7  0,5  35:58.28 MBv8_8.22r3711_                                                                                              
 6010 r         39  19   53264  43460   4536 R  91,7  0,5  18:53.41 MBv8_8.22r3711_                                                                                              
 6069 r         30  10 14,901g 650504 340880 R   8,6  8,0   0:17.73 setiathome_x41p                                                                                              
  973 root     -51   0       0      0      0 S   2,7  0,0  24:23.97 irq/30-nvidia                                                                                                
 1105 r         20   0 4011528 332288  93460 S   0,7  4,1   1:08.97 gnome-shell 

The way I see it, you aren't taking you're own advice and freeing a CPU core to run the GPU. The biggest user after the CPUs is the GPU. Since the CPUs are trying to use every bit of your 4 cores, the GPU has to use part of what the CPUs are trying to use. It's a typical SETI Cluster, both the CPUs and GPU run slower because CPU time isn't allotted to the GPU. Happens a Lot around here.

I would suggest you read your own ReadMe and free a CPU core for the GPU before changing anything else.
;-)

Following the suggestion of Tbar, i suggest you to do the test first without running any CPU WU. Do the tests on the GPU only and when you be satisfied with the test slowly increase some CPU tasks. Take some time to test with or without -nobs. In my host that makes a difference of a pair of seconds. Probably the sweet spot on your 4 core will be 2 CPU + 1 GPU. But that needs to be tested as usual.

---

dont think you can run the monitor off the iGPU while crunching on the GPUs.

Some lot of water has flowed since this message, but anyway I reply, hoping it is still useful.

I do on one of my crunchers.
You just need to prepare an xorg.conf file.
I put the relevant links to information I followed here: http://setiathome.berkeley.edu/forum_thread.php?id=81271&postid=1952070.
You can skip all the driver installation things and just rely on the xorg.conf information. That is all that it takes, after you are are already running the right NVidia drivers. I realised that after some time.
It works as a charm.

If you need more help I am here.
If I can give back just 0.00001 % what Raistmer hase been giving to the community I will be happy to.

Can you elaborate? You say all you have to do is prepare the xorg config file, but the post you linked to says otherwise

- I found no need to edit xorg.conf.

---

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

dont think you can run the monitor off the iGPU while crunching on the GPUs.

Some lot of water has flowed since this message, but anyway I reply, hoping it is still useful.

I do on one of my crunchers.
You just need to prepare an xorg.conf file.
I put the relevant links to information I followed here: http://setiathome.berkeley.edu/forum_thread.php?id=81271&postid=1952070.
You can skip all the driver installation things and just rely on the xorg.conf information. That is all that it takes, after you are are already running the right NVidia drivers. I realised that after some time.
It works as a charm.

If you need more help I am here.
If I can give back just 0.00001 % what Raistmer hase been giving to the community I will be happy to.

Thanks! I'll just keep this info for now and return to multi-GPU configs later when single one will be tuned enough.

Regarding CPU stealing cycles from GPU - yep, I'll test it on this setup thoroughly.

Unfortunately, there are inconclusives started to develop, This one, for example:
http://setiathome.berkeley.edu/result.php?resultid=7308235024
GPU found additional Triplet. CPU missed it. As one can see from stderr, there are many triplets in single PoT detected so app switched to re-processing on CPU. SoG does the same with triplets.
That is, CPU processing somehow different. To ensure correct operation of CUDA app is the main target, optimal performance on particula host is nice but less important.
I'll try to catch that WU and check it with different apps.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Please advise some tools like Windows's GPU-Z/CPU-Z for Linux.
For CPU usage I could use top command, what about GPU?

TOP covers just about everything related to cpu.

You have a couple of tools for the gpu. You have the Nvidia X Server Settings app installed by the drivers in your applications. You can always look at the gpu/s for their temps, fan speeds and clock frequencies.

nvidia-smi on the command line shows utilization, temps and power usage.

---

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

A proud member of the OFA (Old Farts Association)

Here https://cloud.mail.ru/public/HzTH/FczfHokkg I will put TestCases for CUDA Special app along with testing results.
For now excess Triplet under investigation.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

If I remember right, upgrading Special app to version x41p_V0.97b2 should help with missing pulses
You can find it here https://drive.google.com/file/d/1y6u53aJ_1nKqWdi0pngmAWKTB3hK4QWO/view?usp=sharing

The version I gave Raistmer is the latest version. It is the version I recommended to Eric for BETA.
Of course, I haven't heard anything from Eric since...and then Raistmer showed up, finally willing to run Linux. If Eric gets a version from Me it will be the version I gave Raistmer, which works with the repository drivers for 16.04 & 18.04.
<shrugs>

Only tried to help. Let me explain:
I was earlier looking some Raistmers validated tasks and stderr output indicates use of "setiathome v8 enhanced x41p_V0.97, Cuda 9.00 special", not "x41p_V0.97b2", which I found on my tasks.
So I thought he is running earlier version of Special Sauce.

BTW, I was trying to find success with my new ASUS mining board which seems to suffer stalled tasks when more than 8 gpus are used. I decided to try the older OSes since the newer ones weren't working very well. First try was with Ubuntu 14.04.5, where I found they have NV driver 384.130 in the repository. 384.130 works with the cuda90 App, so, that means the App will work with the repository drivers for 14.04, 16.04, and 18.04. 5 Years of Linux system repository drivers will work with the App Raistmer is running, which is what a SETI App needs to do. Now that he has corrected his setup, the App is just as fast as any of the others even without running nobs, which is what my tests show. Looks good to me, in fact, it seems the server decided his inconclusive results were far too low and introduced him to Mr. Bad Wingman.

In other news, 14.04.5 worked fine with 8 gpus but wouldn't use more than 8. 15.04 on the other hand seems to work fine with 10 gpus, without any stalls. It's a shame the ASUS board seems a little slower than the Gigabyte mining board. I'll have to try the Gigabyte one again when the replacement arrives. Now to try 12 on the ASUS board, I had 12 working on the Gigabyte board just before it died.

Here https://cloud.mail.ru/public/HzTH/FczfHokkg I will put TestCases for CUDA Special app along with testing results.
For now excess Triplet under investigation.

It seems all these Arecibo tasks have the Inconclusive rates much higher than normal. As I mentioned to Eric, the App does much better on the BLC tasks. Looking at your current Inconclusive results it appears there are 4 BLC GPU tasks, the rest are Arecibos, with most being Bad Best Pulses on the shorter tasks. The Bad Best pulses are another one of those problems that has been present for a Very long time, about as long as the resuming suspended tasks problem. At least we know it's not associated with the Unroll function, I believe the Unroll is only used on the VLARs, and most of the Inconclusive tasks are not VLARs. Oh well...