Back to bring the PAIN!

[Voodoo Rufus]

I really gotta try this on my 2080Ti. The Windows performance has been underwhelming for crunching.

 

[Holdolin]

for anyone that cares, has an nvidia GPU with CC 5.0 or greater, and willing to run linux.

petri has released a new version of the setiathome v8 nvidia GPU application. version 0.98b1.
this comes in CUDA 9.0 and CUDA 10.1 flavors. (10.1 requires 418+ drivers, FYI)

on high end cards (GTX 1080ti, RTX 2070-2080ti)
expect 10-15% speed increase on Arecibo non-VLAR
expect 25-30% speed increase on Arecibo and BLC VLAR tasks

on lower cards (Maxwell cards, GTX 1050-1080)
expect 10-15% speed increase on all tasks

this is average speedup ON TOP of the already fast v0.97 application, which was about 2-2.5x faster than the windows SoG applications.

My RTX 2070s can do the BLC32 VLARs in about 50 seconds. My 1080tis do them in about 57-60s.

My best system with 10x RTX 2070s, is back to over 1,000,000 RAC

get the whole "All-in-one" package here: http://www.arkayn.us/lunatics/BOINC.7z and run it in your home folder, or cherry pick out the relevant apps and edit your app_info file to use them with your repository BOINC install if you prefer that.

Now that's just some sweetness. Imma have to give that a go. I still need to get poof switched from folding to crunching, guess that's the best place to start. Willing to run Linux? I guess I better get less blood in my coffee stream and get to it

- - - Auto-Merged Double Post - - -

I really gotta try this on my 2080Ti. The Windows performance has been underwhelming for crunching.

You totally should. For all that you've spent on your hardware, to me it just makes good sense to use software best suited to what you want to do. Push your numbers even higher

 

[Voodoo Rufus]

I built it to game with, but might build a living room SFF to stuff the 2080Ti in for linux / HT gaming. Or convert my 2600K rig to Linux. Twerp won't OC the CPU any more.

 

[gsrcrxsi]

just looked up another user running a 2080ti on this latest app.

hes doing the BLC32 VLARs in about 38-40 seconds so compare that speed to whatever you're getting now on windows.

EDIT: That's about 10x faster than your current production rate.

are you running more than one WU at a time? (you can do 2 or 3 at a time most likely)
did you use the lunatics installer?

it doesn't look like you are running any special command line optimizations either, which could at least speed up your current setup

 

[Voodoo Rufus]

If there's any tutorials out there, I'm all ears. Since Lunatics has gone dead, I have no clue where to look any more. The Seti boinc forums aren't very useful or user friendly.

 

[gsrcrxsi]

even though the Lunatics installer hasn't been updated in a while and there aren't any new apps, the lunatics installer is still good to bypass the "learning" period that SETI does. or for systems that for whatever reason don't settle on the fastest app (SoG in most cases). but it looks like you've already settled on SoG so all good there. still probably good to run it to configure your app_info file properly and get you setup.

what you want to do is this:
1. download the lunatics installer here: http://mikesworld.eu/download.html, get the one on the upper right, Win64 0.45 beta 6. this is the latest
2. stop BOINC completely.
3. run lunatics installer, make sure you select the SoG app for NV cards (its not the default), and make the proper selections for your CPU, AVX if it supports it. do not select anything for iGPU if you have one. it's not worth the hassle it will cause.

4. optimization - in your SETI project folder (C:\ProgramData\BOINC\projects\setiathome.berkeley.edu\) there should be a text file called "mb_cmdline_win_x86_SSE3_OpenCL_NV_SoG.txt", open this file and add the folowing:

-sbs 1024 -period_iterations_num 1 -tt 1500 -high_perf -high_prec_timer -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64

(this may cause some screen/mouse lag)

5. run multiple WU - start out with 2 at a time
edit (or create if it doesnt exit) your app_config.xml file, paste the following into it. this file should also be placed in the seti project folder, same location as above

<app_config>
 <app>
 <name>astropulse_v7</name>
 <gpu_versions>
 <gpu_usage>0.5</gpu_usage>
 <cpu_usage>1.0</cpu_usage>
 </gpu_versions>
 </app>
 <app>
 <name>setiathome_v8</name>
 <gpu_versions>
 <gpu_usage>0.5</gpu_usage>
 <cpu_usage>1.0</cpu_usage>
 </gpu_versions>
 </app>
</app_config>

for 3 WU at a time, change the gpu flag to 0.33

 

[Holdolin]

for anyone that cares, has an nvidia GPU with CC 5.0 or greater, and willing to run linux.

petri has released a new version of the setiathome v8 nvidia GPU application. version 0.98b1.
this comes in CUDA 9.0 and CUDA 10.1 flavors. (10.1 requires 418+ drivers, FYI)

on high end cards (GTX 1080ti, RTX 2070-2080ti)
expect 10-15% speed increase on Arecibo non-VLAR
expect 25-30% speed increase on Arecibo and BLC VLAR tasks

on lower cards (Maxwell cards, GTX 1050-1080)
expect 10-15% speed increase on all tasks

this is average speedup ON TOP of the already fast v0.97 application, which was about 2-2.5x faster than the windows SoG applications.

My RTX 2070s can do the BLC32 VLARs in about 50 seconds. My 1080tis do them in about 57-60s.

My best system with 10x RTX 2070s, is back to over 1,000,000 RAC

get the whole "All-in-one" package here: http://www.arkayn.us/lunatics/BOINC.7z and run it in your home folder, or cherry pick out the relevant apps and edit your app_info file to use them with your repository BOINC install if you prefer that.

Zoiks yo, this client is crazy. My dual 2080ti system is downing wu's at ~40sec each. Now to get this set up on all my rigs

 

[gsrcrxsi]

Glad to hear you got it working.

Mind posting your host detail page?

With 2 2080tis, if it’s a 24/7 cruncher, you’ll likely make it to the top 10 of the most powerful hosts rankings.

https://setiathome.berkeley.edu/top_hosts.php

 

[Holdolin]

This what ya lookin for? Sorry, I'm a nub at this.

https://setiathome.berkeley.edu/show_host_detail.php?hostid=8706563

 

[gsrcrxsi]

Yeah that’s it. I wanted to see that computer and any others you’re running. You have a good amount of power, convert them all to the special app and you’ll be amongst the top ranks of contributors to the project.

 

[Holdolin]

Here ya go:

https://setiathome.berkeley.edu/hosts_user.php?userid=10809798

 

[Voodoo Rufus]

200k so far with dual 2080Tis? Very nice. Can't wait to see what it stabilizes at.

 

[gsrcrxsi]

200k is his current user average. he has 6 different systems running.

the dual 2080ti system he has currently has a RAC of 67k and rising, but will probably get up to around 300k if he sticks with it for about 2 months.

 

[Voodoo Rufus]

I was looking at his initial daily output for that rig on boinc stats.

 

[gsrcrxsi]

i prefer FreeDC personally, but yeah.

his system on FreeDC = https://stats.free-dc.org/stats.php?page=hostbycpid&cpid=4895fbbd6a1610b0e6a115b4b97458cc
it takes a few days for your daily production to level off and build up that revolving cache of WUs that just take a long time to validate with slow wingmen.

i was estimating his production vs the 2nd fastest system, which has 4x 2080tis, and does about 650k here

and my 10x2070 machine is doing about 1.4 million per day now lol. (RAC only at 1.125 mil and climbing)

 

[Voodoo Rufus]

Using the windows optimizations and line additions, I'm doing units in about 5.5 minutes each on the GPU.

 

[gsrcrxsi]

nice, let me take a look at some of your completed WUs and see if i can recommend anything more.

are you running multiple WU per GPU? or just one WU per GPU now?

 

[Voodoo Rufus]

Two units on the GPU. Giving the card two dedicated threads from the CPU.

 

[gsrcrxsi]

awesome, that's probably best. makes for an effective 2:45 per WU or so.

after running if for a while, see if it can handle 3 at a time. but if individual WU run time goes above 8:15mins (this would be the same production you have now), then revert.

 

[Holdolin]

200k so far with dual 2080Tis? Very nice. Can't wait to see what it stabilizes at.

Hehe, me either. I now have all of my systems running the new client. Curious to see how much work I can get done.