Optimizing the apps

Message boards : Number crunching : Optimizing the apps
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 6 · 7 · 8 · 9 · 10 · 11 · Next

AuthorMessage
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 769
Credit: 22,251,971,800
RAC: 6,978,709
Message 1669 - Posted: 13 Apr 2019, 12:35:52 UTC - in response to Message 1668.  
Last modified: 13 Apr 2019, 12:42:20 UTC

Hi Mikey, I used the same as a 1080 and seemed to it work ok.


Thank you very much!!! I tried the 480 numbers and they are exactly the same and it's taking 725 seconds for each unit. Are you running just one unit at a time too?
ID: 1669 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ian

Send message
Joined: 6 Sep 14
Posts: 3
Credit: 76,904,042,169
RAC: 110,701,008
Message 1672 - Posted: 14 Apr 2019, 11:35:22 UTC - in response to Message 1669.  

Hi, Yes only 1 work unit per card
ID: 1672 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 769
Credit: 22,251,971,800
RAC: 6,978,709
Message 1673 - Posted: 14 Apr 2019, 12:38:09 UTC - in response to Message 1672.  

Hi, Yes only 1 work unit per card


Yup that's what I'm doing too, thank you very much for your help!!
ID: 1673 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Arnulf

Send message
Joined: 30 Oct 17
Posts: 7
Credit: 6,211,530,570
RAC: 1,220
Message 1795 - Posted: 10 Jun 2019, 9:06:23 UTC

Hi!

Since my last posting I have gotten a Radeon VII card.
I used the Vega 64 config as a starting point and saw times around 4 minutes, down from around 5:25 on the Vega 64.

Some experimenting with the settings has the times down to around 3:25, so here is my settings for the ASUS Radeon VII:

verbose=1
kernels_per_reduction=64
threads=8
lut_size=19
sieve_size=30
cache_sieve=1
sleep=0
reduce_cpu=0

I have increased the lut_size from 17 to 19 and the kernels_per_reduction from 56 to 64.

Notes:
- Increasing threads above 8 makes the WU error out.
- Increasing sieve_size above 30 makes the WU error out
ID: 1795 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bluestang

Send message
Joined: 30 Jul 18
Posts: 5
Credit: 1,482,500,380
RAC: 14
Message 1856 - Posted: 25 Jun 2019, 17:18:59 UTC - in response to Message 1795.  

Hi Arnulf,

Great info thanks. What driver version you running?
ID: 1856 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Arnulf

Send message
Joined: 30 Oct 17
Posts: 7
Credit: 6,211,530,570
RAC: 1,220
Message 1866 - Posted: 27 Jun 2019, 7:11:41 UTC - in response to Message 1856.  

I had to check .. currently running 19.6.1 but I can see a 19.6.2 optional update waiting for me.
ID: 1866 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Robert Klein

Send message
Joined: 11 Jun 17
Posts: 4
Credit: 12,812,877,342
RAC: 10,459,581
Message 1878 - Posted: 2 Jul 2019, 8:33:41 UTC

Hi!
Are there any suggestions for the parameters of the Quadro M4000?
Thanks and happy crunching!
Robert
ID: 1878 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
VietOZ

Send message
Joined: 24 Jul 18
Posts: 1
Credit: 9,751,299,335
RAC: 1,619
Message 1879 - Posted: 6 Jul 2019, 18:23:23 UTC - in response to Message 1878.  

Hi!
Are there any suggestions for the parameters of the Quadro M4000?
Thanks and happy crunching!
Robert


Try:
verbose=1
kernels_per_reduction=48
threads=8
lut_size=16
sieve_size=30
cache_sieve=1
sleep=0
reduce_cpu=0


Runs a few tasks then change lut_size=16 to lut_size=17 to see which one is faster then go with that one
ID: 1879 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Robert Klein

Send message
Joined: 11 Jun 17
Posts: 4
Credit: 12,812,877,342
RAC: 10,459,581
Message 1880 - Posted: 8 Jul 2019, 12:52:18 UTC - in response to Message 1879.  

Quadro M4000


Try:
verbose=1
kernels_per_reduction=48
threads=8
lut_size=16
sieve_size=30
cache_sieve=1
sleep=0
reduce_cpu=0


Runs a few tasks then change lut_size=16 to lut_size=17 to see which one is faster then go with that one


Thank you very much! Looks very good. 40 percent more credits per GPU time in both variants. Also no errors with both.
https://boinc.thesonntags.com/collatz/show_host_detail.php?hostid=822306
ID: 1880 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
sir sant

Send message
Joined: 17 Nov 10
Posts: 6
Credit: 4,565,763,909
RAC: 0
Message 1973 - Posted: 7 Oct 2019, 19:21:05 UTC - in response to Message 1615.  
Last modified: 7 Oct 2019, 19:22:19 UTC

Hi

I wondered if anyone had got the optimsed settings for the GTX1660 Ti please?


Next thread would have given you this:

verbose=1
kernels_per_reduction=48
threads=7
lut_size=17
sieve_size=30
cache_sieve=1
sleep=0

at least it's something to start with.

Played around with mine:
<app_config>
verbose=1
kernels_per_reduction=48
threads=6
lut_size=17
sleep=1
reduce_cpu=0
sieve_size=30
</app_config>

I get around 8mill a day with my 1660ti using that config and mild oc.
I run 2wu-s together.
ID: 1973 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 769
Credit: 22,251,971,800
RAC: 6,978,709
Message 1974 - Posted: 8 Oct 2019, 11:34:04 UTC - in response to Message 155.  

this is where i have my 1080 ti set currently.verbose=1
kernels_per_reduction=48
threads=8
lut_size=18
sleep=0
cache_sieve=1
reduce_cpu=0
sieve_size=30 there running between 235 and 245 sec.not making anywhere near the credit i was making before the issues started.was getting around 9-10 million a day,now down to little over 3 million.oh well.


These are my own1080Ti settings:
AMD 1080Ti
verbose=1
kernels_per_reduction=48
sleep=1
threads=9
lut_size=18
reduce_CPU=0
sieve_size=30
cache_sieve=1

My last unit: ran in 244.41 but I currently run Collatz as my backup project not my primary one so don't know how much it can get per day if it just ran Collatz.
ID: 1974 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Skivelitis2
Avatar

Send message
Joined: 28 Mar 15
Posts: 8
Credit: 3,173,151,463
RAC: 12,542,704
Message 1981 - Posted: 10 Oct 2019, 10:50:01 UTC

Could someone please suggest settings for both a GTX 1650 and a GTX 950? I used the settings mikey provided for my GTX 1060 and am getting the expected 13min run times. Tried the same settings on the 1650 and got comp errors. Not sure what or in which direction to adjust.
ID: 1981 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 769
Credit: 22,251,971,800
RAC: 6,978,709
Message 1983 - Posted: 10 Oct 2019, 11:08:24 UTC - in response to Message 1981.  

Could someone please suggest settings for both a GTX 1650 and a GTX 950? I used the settings mikey provided for my GTX 1060 and am getting the expected 13min run times. Tried the same settings on the 1650 and got comp errors. Not sure what or in which direction to adjust.


To start for the 950 try the 980 ones, as for the 1650 try the 1660 ones and then adjust the Lut size up or down and see if the errors go away.
ID: 1983 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Skivelitis2
Avatar

Send message
Joined: 28 Mar 15
Posts: 8
Credit: 3,173,151,463
RAC: 12,542,704
Message 1984 - Posted: 10 Oct 2019, 12:13:15 UTC

Well, now I am truly baffled! I used the settings for the 1660 as suggested on one of my 1650s and the work unit did not error out. However, whereas I was getting 16/17min run times with the stock app (no dedicated CPU core), the opti app was only 42% complete after 20 minutes. Promptly aborted that wu and deleted the opti file and it appears run times are back close to where they were after 50% completion. I have to get out and help pay for the cost of running these things soon but I will tinker around this evening or by the weekend at the latest and at least get the drivers updated to the same version especially on the identical CPUs.
ID: 1984 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 769
Credit: 22,251,971,800
RAC: 6,978,709
Message 1985 - Posted: 10 Oct 2019, 19:52:37 UTC - in response to Message 1984.  

Well, now I am truly baffled! I used the settings for the 1660 as suggested on one of my 1650s and the work unit did not error out. However, whereas I was getting 16/17min run times with the stock app (no dedicated CPU core), the opti app was only 42% complete after 20 minutes. Promptly aborted that wu and deleted the opti file and it appears run times are back close to where they were after 50% completion. I have to get out and help pay for the cost of running these things soon but I will tinker around this evening or by the weekend at the latest and at least get the drivers updated to the same version especially on the identical CPUs.


Don't look at the run times as it crunches as under the opti app they will bounce all over the place, you only care about the finish time. Crunch one unit under the opti app then make changes, abort the unit it's working on and the next unit to start will use the new settings. Keep a chart of your changes and the results, that way once you get going the right way and find the sweet spot you will know it. Don't worry about the aborted units as someone else will get them, it's one less the Server has to create.
ID: 1985 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Skivelitis2
Avatar

Send message
Joined: 28 Mar 15
Posts: 8
Credit: 3,173,151,463
RAC: 12,542,704
Message 1987 - Posted: 11 Oct 2019, 18:47:27 UTC - in response to Message 1985.  

Thanks for the help and advice mikey. What alarmed me was the fact that the actual clock time was longer on the very first optimized work unit compared to the stock app. I will be a little more patient this weekend and fiddle with the settings some. Have other cards to optimize as well.....

ID: 1987 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 769
Credit: 22,251,971,800
RAC: 6,978,709
Message 1988 - Posted: 12 Oct 2019, 11:39:04 UTC - in response to Message 1987.  

Thanks for the help and advice mikey. What alarmed me was the fact that the actual clock time was longer on the very first optimized work unit compared to the stock app. I will be a little more patient this weekend and fiddle with the settings some. Have other cards to optimize as well.....


It takes time for it to settle down, it can even start out saying it will take longer but the closer to the end it gets the faster it goes. One thing mine do is bounce high and then low, the time left I'm talking about, and it seems to do it for every workunit but in the end they are much faster. I only run one workunit at a time and I always leave one cpu core free just for the gpu to use.
ID: 1988 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Skivelitis2
Avatar

Send message
Joined: 28 Mar 15
Posts: 8
Credit: 3,173,151,463
RAC: 12,542,704
Message 1992 - Posted: 14 Oct 2019, 12:51:21 UTC

Just upgraded a box from Mint 18.3 to 19.2. In 18.3, I always used the gksudo xed command in a terminal to modify files and in fact used it successfully yesterday to install the optimized app. Mint 19.2 no longer uses that command and trying with sudo returns a command not found message. What is the proper command in Mint 19.2?
ID: 1992 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 769
Credit: 22,251,971,800
RAC: 6,978,709
Message 1994 - Posted: 15 Oct 2019, 10:35:29 UTC - in response to Message 1992.  

Just upgraded a box from Mint 18.3 to 19.2. In 18.3, I always used the gksudo xed command in a terminal to modify files and in fact used it successfully yesterday to install the optimized app. Mint 19.2 no longer uses that command and trying with sudo returns a command not found message. What is the proper command in Mint 19.2?


I right click on the folder to edit as root then edit it with the text editor, I do not use the command line
ID: 1994 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Padanian

Send message
Joined: 28 May 10
Posts: 13
Credit: 2,392,304,259
RAC: 2
Message 1996 - Posted: 17 Oct 2019, 9:37:29 UTC

Just updated my PC with a Gigabyte GTX1660Ti and config'ed it with the setup recommended in post #1973

I'm getting an average of 340-360s on a single WU for an average of 29k credit/WU. In 20hours/day that would make approx 200WU/day for a gross credit of 5.8M/day. Tomorrow I will check if my math is correct.
Is it reasonable?

I would probably throttle it down a bit: power input and average temperature are rather significant.
ID: 1996 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 6 · 7 · 8 · 9 · 10 · 11 · Next

Message boards : Number crunching : Optimizing the apps


©2020 Jon Sonntag; All rights reserved