After a bit of a break, I decided to use some of the spare hardware I have. I thought a Titan V would be a monster for this type of number crunching, but the results I have seen aren't great. So I figure I may be doing something wrong.

Note, I haven't tweaked the config files, except for cc_config.xml to get all GPUs enabled.

My Titan V is taking roughly 19 minutes per work unit.
My Titan X is taking roughly 19 minutes as well.
My 1060s are taking roughly 7 minutes per work unit.

I have enabled the 'optimize for compute' in the nvidia control panel.

The only difference I can see... the Titans are running 8.0..Cuda50.exe and the 1060s are running 8.22...opencl_nvidia_S0G.exe

Any help?

Thanks

yes the problem is that you're using the very old cuda50 app on the titans.

i see you used to have this system on linux. is there a reason you switched to windows 10?

there exists a highly optimized application that is linux only, that will greatly improve run times of your titans (and also 1060s), something like 30 seconds per WU on the titan V based on run times from another member here who is running that.

if you are ok with switching back to linux, you can get the most out of your hardware.

---

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

Absolutely I can switch, I prefer Fedora anyway. I switched to Win10 just to troubleshoot.

I will say that I was not getting good performance with linux either, which is what prompted the switch. Perhaps I need some special config when running under linux? or is it supposed to work 'out of the box'?

---

I will preface this with the fact that most of us here running linux are running Ubuntu, so some instructions might be slightly different in Fedora, but if you're already well versed in Fedora you may be able to spot where the modifications are required.

Download the entire BOINC package here: http://www.arkayn.us/lunatics/BOINC.7z

unzip/extract it to your home folder.
you need some dependencies before running it. this might be where you could run into issues.
you also need the proprietary nvidia drivers. if you use the standard CUDA9.0 app, i think you can get by with the repository driver (at least 384+ according to the post)
if you elect to run a slightly more recent version running CUDA 9.2 you will need a more recent driver, at least 396+, and if you want to run the CUDA 10 version, you'll need i think 410+.

This is a Package to install the Updated Berkeley BOINC version 7.8.3 with the CUDA 9.0 Special App Preinstalled. You should check the Dependencies and Execute bits before running. Most distros will need to install libwebkitgtk, some libcurl3. Mint users should probably use the repository version of BOINC. Read the README.md and the README_x41p_V0.97.txt in the docs folders. See 'The Berkeley Installer' for outdated instructions. To 'install', simply expand the download to your Home folder, make sure Dependencies are installed, open the BOINC folder, and double click boincmgr. You will need to add the Project SETI@Home to BOINC.

To Install the Dependencies, In the terminal;
To install 7zip run, sudo apt-get install p7zip
To install libwebkitgtk run, sudo apt-get install libwebkitgtk-1.0-0
To install libcurl3 run, sudo apt-get install libcurl3

The CUDA 9.0 Special App Requires at least Ubuntu 14.04.1 and a CUDA 9.0 driver installed. The Repository Driver 384.x should work.

CUDA 9.2 app here: http://www.arkayn.us/lunatics/Linux_Pascal+0.97b2_Special.7z, requires nv driver 396+
CUDA 10 app here: http://www.arkayn.us/lunatics/setiathome_x41p_V0.97_x86_64-pc-linux-gnu_cuda100.7z, requires nv driver 410+

if using app version not included in first boinc CUDA 9.0 package, make sure you copy and paste all the files from the CUDA9.2 or CUDA 10 packages, most importantly the app_info.xml file since that tells BOINC which apps to use.

[EDIT]
Looks like the CUDA 10 package is the app only, so you would need to manually edit the app_info file with the correct app name to get it working. but really all of the apps CUDA9+ are pretty close to each other in performance.

[EDIT2]
If you have problems getting the all-in-one package running due to dependencies or something, you can also use the repository install of BOINC, just copy and paste the apps and supporting config files from the all in one package or later CUDA packages. it will work if you have the proper app_info file.

---

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

Absolutely I can switch, I prefer Fedora anyway. I switched to Win10 just to troubleshoot.

I will say that I was not getting good performance with linux either, which is what prompted the switch. Perhaps I need some special config when running under linux? or is it supposed to work 'out of the box'?

The linux special app has an autoconfig who works very well. Try it first. You will see the difference. Especially on your Titan V. it's like you turn on the afterburners. Just follow the instructions posted on the previous msg and not forget with this apps you must run only 1 WU at a time on each GPU.

A Titan V could crunch a BLC WU in just 30-35 secs compare that with you actual 1100 secs and you will feel the dark side of the real power of your GPU.

---

Awesome, thank you. I'll work on setting that up today. Will notify when complete

---

The issue was running the CUDA5.0 app in both Windows and Linux. That app is not at all well suited to current Arecibo or GBT work. The SoG app is best for Windows. OTOH, the Linux CUDA9.2 and CUDA10.0 apps are the absolute best with modern hardware for both Arecibo and GBT work. So as others have stated, a return to a Linux installation would be best for performance.

---

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

A proud member of the OFA (Old Farts Association)

Did Jason ever release His windows code for the Newer Cuda 9.x for Windows?

Methinks Jason is gone, he has another life to live.

---

Wow, that's fast...

If you wish to follow the amazing crunching speed of the Titan V with Linux special builds just follow this host

https://setiathome.berkeley.edu/show_host_detail.php?hostid=7475713

It's has 1 Titan V + a 1080Ti + 2 x 1080 IIRC

The only close match for a Titan V AFAIK is a 2080Ti but even it can't beat the Titan crunching speed performance... look at:

https://setiathome.berkeley.edu/show_host_detail.php?hostid=8600449

---

A Titan XP is very similar to a 1080Ti. The V just smokes it!
My XP is running at 65s for the current BLC tasks.

No, Jason departed to his alternate field before he released the Windows version of "Petri's special". Reading back he had all sorts of problems with getting it stable and consistent, and that together with some "inter-personal issues" he vanished.
All told this is a great shame.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Thanks much Ian@Steve and Juan.

After some playing around with the special build of boinc itself, it because clear that just the repo version of boinc with the substitute the project files as all I needed. I have it running the v100 version of Cuda now, and running much better :-) I am sure I'll be tweaking settings over the week, but for now, I just want to let it stabilize so I can establish a baseline.

The machine has 1 Titan V and 2 Titan X gpus in it. The TitanV looks to take around 35-50 seconds per work unit, and the Titan X take around 90 seconds to complete their work. I can tell my next concern will be the generated heat. The Titan V doesnt get nearly as warm as the Titan Xs do. I'll have to maybe get an open style case, or maybe look into liquid cooling all three. Suggestions welcome!

Thanks again guys. You're all awesome.

Tod

---

Put a power limit on the Titan X's.

nvidia-smi -pm 1
nvidia-smi -i 1 -pl 150
nvidia-smi -i 2 -pl 150

---

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

A proud member of the OFA (Old Farts Association)

Thanks much Ian@Steve and Juan.

After some playing around with the special build of boinc itself, it because clear that just the repo version of boinc with the substitute the project files as all I needed. I have it running the v100 version of Cuda now, and running much better :-) I am sure I'll be tweaking settings over the week, but for now, I just want to let it stabilize so I can establish a baseline.

The machine has 1 Titan V and 2 Titan X gpus in it. The TitanV looks to take around 35-50 seconds per work unit, and the Titan X take around 90 seconds to complete their work. I can tell my next concern will be the generated heat. The Titan V doesnt get nearly as warm as the Titan Xs do. I'll have to maybe get an open style case, or maybe look into liquid cooling all three. Suggestions welcome!

Thanks again guys. You're all awesome.

Tod

Glad you got it sorted out. you really don't have to do much tweaking with the special app. just let it do it's thing and only run 1 job per GPU.

Put a power limit on the Titan X's.

nvidia-smi -pm 1
nvidia-smi -i 1 -pl 150
nvidia-smi -i 2 -pl 150

i think 150w is a bit severe might see too much slow down, 200w is a better balance IMO.

---

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

i think 150w is a bit severe might see too much slow down, 200w is a better balance IMO.

I thought about suggesting 180 or 200 W at first, but he complained about too much in heat in a compromised computer case. I've rarely seen 176W power consumption on my 1080Ti's with no restrictions. Unsure where the Titan X falls in comparison to the 1080Ti.

---

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

A proud member of the OFA (Old Farts Association)

I've set the power limit at 200 for now. I'll see where that gets me. As far as tweaking, I was more referring to the clocks. I might play with the gpu and memory clocks a little to find the sweet spot, in relation to the power consumed. (i.e. is the seti app memory or gpu bound.)

Definitely looking forward to seeing numbers after a couple days.

---

Titan X (Maxwell) is slower than a 1080ti. but comparable power consumption i believe.

while average power consumption may not breach ~170ish W, peak can certainly hit the max.

unrestricted, my 1080tis will have fairly low power consumption in the early stages of the WU, but will ramp up to max power ~240w towards the later half of the WU.

limiting the ceiling still impacts the job even if your average power use is less. now it is limited to that 200W when i hit the later half of the job, which slows it down just a bit.

---

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

I've set the power limit at 200 for now. I'll see where that gets me. As far as tweaking, I was more referring to the clocks. I might play with the gpu and memory clocks a little to find the sweet spot, in relation to the power consumed. (i.e. is the seti app memory or gpu bound.)

Definitely looking forward to seeing numbers after a couple days.

We've had lots of arguments on that in the past. Depends on the project and the applications. My contention is that the Seti special app responds best to memory overclocks. I see the builtin Nvidia firmware GPUBoost 3.0 do the heavy lifting with respect to the core clocks. If the card has the temp and TDP headroom, the card will boost the core clocks on its own without any intervention on your part. The recent special app has made an effort to reduce the amount of memory calls while doing computations so the benefit of increasing the memory clocks has been reduced from the older CUDA8 zi3v app for example.

Also don't forget that Nvidia penalizes the card when the driver detects a compute load being run and reduces the power state to P2 so that automatically inflicts both a core and mainly a memory clock reduction. You need to run a memory overclock to get the memory back up to what it should be running in normal graphic mode P0 power state.

---

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

A proud member of the OFA (Old Farts Association)

+1, you'll see the most changes by playing with card clocks, rather than any changes to seti app configs.

the app has almost all of the optimization built in for you. you really only have one knob left to turn as far as the app is concerned, and that's to run the command line argument "-nobs" or not. doesn't make a huge difference, but it's noticeable. you're in a good spot either way.

---

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

+1, you'll see the most changes by playing with card clocks, rather than any changes to seti app configs.

the app has almost all of the optimization built in for you. you really only have one knob left to turn as far as the app is concerned, and that's to run the command line argument "-nobs" or not. doesn't make a huge difference, but it's noticeable. you're in a good spot either way.

I was running -nobs from the get go but I have changed to not using it because the improvement is minimal now with the CUDA92 and CUDA10 with the recent change to inline CUDA libraries into the app app itself and the large reduction in calls outside the compute kernel and the memory calls. Also, there is a noticeable reduction in core utilization which favorably reduces cpu temps.

---

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

A proud member of the OFA (Old Farts Association)