Optimizing the apps

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

To post messages, you must log in.

Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · Next

AuthorMessage
Speedy51

Send message
Joined: 9 Oct 10
Posts: 8
Credit: 84,882,990
RAC: 473
Message 2687 - Posted: 18 May 2020, 3:53:27 UTC - in response to Message 2669.  

Thanks, I get very similar run times to you.
ID: 2687 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Carlos
Avatar

Send message
Joined: 12 Jul 09
Posts: 10
Credit: 1,510,678,202
RAC: 0
Message 2723 - Posted: 26 May 2020, 2:07:44 UTC - in response to Message 312.  

Well it's time for me to throw my hat into the ring. I am running 3 RX2070 and it appears that with standard setting my results were

around 389 to about 399 sec run time and 38,812 credits. Total just shy of 20 million credits per day.

I just put this in my config file and will see what happens.
<app_config>
verbose=1
kernels_per_reduction=64
threads=8
lut_size=18
sleep=0
reduce_cpu=0
sieve_size=30
cache_sieve=1
</app_config>

Any suggestions greatly appreciated.
ID: 2723 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
TimeLord04
Avatar

Send message
Joined: 30 Aug 18
Posts: 398
Credit: 344,280,772
RAC: 0
Message 2725 - Posted: 26 May 2020, 9:25:33 UTC - in response to Message 2723.  

Well it's time for me to throw my hat into the ring. I am running 3 RX2070 and it appears that with standard setting my results were

around 389 to about 399 sec run time and 38,812 credits. Total just shy of 20 million credits per day.

I just put this in my config file and will see what happens.
<app_config>
verbose=1
kernels_per_reduction=64
threads=8
lut_size=18
sleep=0
reduce_cpu=0
sieve_size=30
cache_sieve=1
</app_config>

Any suggestions greatly appreciated.

I'm NOT sure IF that works, Carlos... NOT app_config.xml BUT instead there's a ".config" File
wherein these parameters are Set.

The app_config.xml CAN be Set to increase the Number of Tasks per Card... (Depending on the
available VRAM on the Cards in question.) ie: On my MacVidCards' GTX-1070 8GB Card,
originally I was able to do 2 Units at a time. IF your 2070 Cards also have 8GB VRAM, you
'should' be able to do AT LEAST 2 Units at a time per GPU.


TL
Have TARDIS will travel!!!
Come along, K-9!
ID: 2725 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 927
Credit: 24,523,632,110
RAC: 0
Message 2727 - Posted: 26 May 2020, 9:49:41 UTC - in response to Message 2725.  

Well it's time for me to throw my hat into the ring. I am running 3 RX2070 and it appears that with standard setting my results were

around 389 to about 399 sec run time and 38,812 credits. Total just shy of 20 million credits per day.

I just put this in my config file and will see what happens.
<app_config>
verbose=1
kernels_per_reduction=64
threads=8
lut_size=18
sleep=0
reduce_cpu=0
sieve_size=30
cache_sieve=1
</app_config>

Any suggestions greatly appreciated.

I'm NOT sure IF that works, Carlos... NOT app_config.xml BUT instead there's a ".config" File
wherein these parameters are Set.
TL


If you look at his valid workunits it looks like he he did put it in the right place he just didn't need the <app_config> or </app_config> parts,
by that I mean the Project is using the commands but is saying the </app_config> is an unknown command.His run time is about 4 min per workunit.

And that's where the second part of your answer comes in...trying running multiple units at one time to see if the speed stays up there.

The app_config.xml CAN be Set to increase the Number of Tasks per Card... (Depending on the
available VRAM on the Cards in question.) ie: On my MacVidCards' GTX-1070 8GB Card,
originally I was able to do 2 Units at a time. IF your 2070 Cards also have 8GB VRAM, you
'should' be able to do AT LEAST 2 Units at a time per GPU.


The app_config.xml file would go into the Collatz project directory and I think examples have been posted in here in the past.
ID: 2727 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Carlos
Avatar

Send message
Joined: 12 Jul 09
Posts: 10
Credit: 1,510,678,202
RAC: 0
Message 2729 - Posted: 26 May 2020, 13:31:32 UTC - in response to Message 2727.  

Well it looks like I did put it in the right place. My run times are down to about 240 sec but the credit per WU is also down to 29-30,000 per WU. A drop of about 10,000 credits per WU. Time to tinker.
ID: 2729 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Slicker
Project administrator

Send message
Joined: 11 Jun 09
Posts: 78
Credit: 943,644,517
RAC: 0
Message 2730 - Posted: 26 May 2020, 18:19:59 UTC - in response to Message 844.  

I have no idea what that means and haven't tried using a one instead.


The 980 config you posted says "verbose=0".

Was it a mistake or is there a reason to set verbose to 0?

I'd guess mistake, so maybe it should be corrected to 1?


It is a copy from the old website before Collatz switched to the current one, so no it wasn't a typo or mistake on my part.
I run 2 GTX980's and they are doing just fine with the zero.

My point was I have no clue what "verbose" even means in this context so a 1 or a zero wasn't tested by me any more than the settings for my GTX1080Ti gpu's were tested by me. Jon posted the original post and then people said changing this or that setting to this or that was faster so I copied the new settings and pasted them here.


Verbose just logs more information which helps with debugging if the apps aren't behaving as expected.
ID: 2730 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
TimeLord04
Avatar

Send message
Joined: 30 Aug 18
Posts: 398
Credit: 344,280,772
RAC: 0
Message 2731 - Posted: 26 May 2020, 21:37:35 UTC - in response to Message 2729.  
Last modified: 26 May 2020, 21:41:16 UTC

Well it looks like I did put it in the right place. My run times are down to about 240 sec but the credit per WU is also down to 29-30,000 per WU. A drop of about 10,000 credits per WU. Time to tinker.

Yes, you have it in the right place. I just reacted to the <app_config> and </app_config> Lines,
they don't need to be there in this ".config" File. (As pointed out by Mikey.)

As I mentioned, though, you CAN create an "app_config.xml" File to Crunch 2 or more Units
at a time per GPU.

Welcome aboard, Carlos!!! 😀😃😄😁

[EDIT:] As to the Total Credits per Unit dropping while Crunching faster, the
deficit is made up by Total Work Done which GOES UP by Crunching faster.

Have fun tweaking your Settings. Took me about a week to get things locked in.
I've had to STOP Crunching; however, due to LACK of GPU Fan Control in MacOS.


TL
Have TARDIS will travel!!!
Come along, K-9!
ID: 2731 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Farscape
Avatar

Send message
Joined: 21 Sep 09
Posts: 5
Credit: 35,217,840,774
RAC: 58,404,285
Message 2742 - Posted: 27 May 2020, 17:24:39 UTC
Last modified: 27 May 2020, 17:29:11 UTC

Carlos - As to faster times / lower credit issue you talked about.....All of my RTX 2070s turn out 10-11 million credits per day day. the faster times more than offset the lower credit per WU.
None are overclocked.
ID: 2742 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 927
Credit: 24,523,632,110
RAC: 0
Message 2743 - Posted: 27 May 2020, 22:34:47 UTC - in response to Message 2742.  

Carlos - As to faster times / lower credit issue you talked about.....All of my RTX 2070s turn out 10-11 million credits per day day. the faster times more than offset the lower credit per WU.
None are overclocked.


And that's what's supposed to happen
ID: 2743 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Carlos
Avatar

Send message
Joined: 12 Jul 09
Posts: 10
Credit: 1,510,678,202
RAC: 0
Message 2744 - Posted: 28 May 2020, 12:52:56 UTC - in response to Message 2743.  

Carlos - As to faster times / lower credit issue you talked about.....All of my RTX 2070s turn out 10-11 million credits per day day. the faster times more than offset the lower credit per WU.
None are overclocked.


And that's what's supposed to happen


That is what I am seeing. My rac is climbing. I am not getting 10-11 per card. I am around 22 million credits per day with all 3. Still need to tinker.
ID: 2744 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 927
Credit: 24,523,632,110
RAC: 0
Message 2745 - Posted: 28 May 2020, 20:17:27 UTC - in response to Message 2744.  
Last modified: 28 May 2020, 20:29:03 UTC

Carlos - As to faster times / lower credit issue you talked about.....All of my RTX 2070s turn out 10-11 million credits per day day. the faster times more than offset the lower credit per WU.
None are overclocked.


And that's what's supposed to happen


That is what I am seeing. My rac is climbing. I am not getting 10-11 per card. I am around 22 million credits per day with all 3. Still need to tinker.


I see it's an 8 core cpu how many of those cores do you leave free for each gpu to use? I would suggest letting it use 1 full cpu core per gpu and see what if any change there is, then slowly ratcheting it back until you are happy.

Farscape, the other guy that posted with the11 million per day, is using almost double the cpu time on his units and they are all at the low end of your completion range, both on 8 cpu core pc's. That suggests he's giving it more cpu time and you are giving it less, or yours are on 4x pcie-slots and his are on 8x or 16x pcie-slots. Anything from 8x and up is faster than Boinc needs but 4x slows things down.

Do you have any SLI cables attached? If so removing them will make things faster as Boinc has no clue about them so checking them slows things down. You want each card to be in it's own independant World as much as possible but still have cpu access whenever it needs it. And NO an SSD drive will not help at all, Boinc doesn't write that much that often to matter unless the drive is nearly full then it can be a slowdown.
ID: 2745 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProDigit

Send message
Joined: 23 Nov 19
Posts: 13
Credit: 6,318,341,604
RAC: 37,029
Message 2747 - Posted: 31 May 2020, 14:03:26 UTC

Hi guys,
My RTX2080Ti runs the whole thing just fine without tuning, in about 3:30 min runs per wu.
Can I improve on this?

Also, app_config.xml needs to be edited, as the CPU usage reported to Boincmgr is incorrect.
Currently all my collatz WUs are reported to use 0.97CPU, but in htop I only see 2.5% CPU used.
My app_config.xml looks like this:
<app_config>
<app>
<name>collatz_sieve</name>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>0.03</cpu_usage>
</gpu_versions>
</app>
</app_config>

But it doesn't seem to affect CPU or GPU usage,
ID: 2747 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 927
Credit: 24,523,632,110
RAC: 0
Message 2748 - Posted: 31 May 2020, 19:33:31 UTC - in response to Message 2747.  

Hi guys,
My RTX2080Ti runs the whole thing just fine without tuning, in about 3:30 min runs per wu.
Can I improve on this?

Also, app_config.xml needs to be edited, as the CPU usage reported to Boincmgr is incorrect.
Currently all my collatz WUs are reported to use 0.97CPU, but in htop I only see 2.5% CPU used.
My app_config.xml looks like this:
<app_config>
<app>
<name>collatz_sieve</name>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>0.03</cpu_usage>
</gpu_versions>
</app>
</app_config>

But it doesn't seem to affect CPU or GPU usage,


The 'tuning' being discussed here has a GTX 2070 running units in 3:45 per run, I think it would help your gpu ALOT!!!

Read this thread: https://boinc.thesonntags.com/collatz/forum_thread.php?id=123
One of the latest messages says the person is doing units iin 176.6 seconds, or under 3 minutes on their own 2080
ID: 2748 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Carlos
Avatar

Send message
Joined: 12 Jul 09
Posts: 10
Credit: 1,510,678,202
RAC: 0
Message 2749 - Posted: 1 Jun 2020, 14:54:28 UTC

Looks like I have found a good spot. WU in around 220 range. I did try to push a little harder but that crashed the system. This looks good.
ID: 2749 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile mikey
Avatar

Send message
Joined: 11 Aug 09
Posts: 927
Credit: 24,523,632,110
RAC: 0
Message 2750 - Posted: 1 Jun 2020, 22:29:13 UTC - in response to Message 2749.  
Last modified: 1 Jun 2020, 22:29:30 UTC

Looks like I have found a good spot. WU in around 220 range. I did try to push a little harder but that crashed the system. This looks good.


Yup just let it settle in for a bit and watch your RAC go up a bit as you do workunits faster.
ID: 2750 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Fardringle

Send message
Joined: 21 Jul 09
Posts: 6
Credit: 178,017,712
RAC: 203
Message 3147 - Posted: 9 Feb 2021, 1:09:51 UTC

I'm not sure if it's best to post here or start a new thread, so I'll go here first..

I decided to give the optimization settings a try on a couple of older GPUs to see what happens.

The recommended settings for the R9 290X are great for my R9 280X. They cut task run times in half, resulting in about 2.8M points per day, which is not bad at all for a really old card.

Unfortunately, I can't seem to get the recommended settings to work for my GTX 1060Ti 3GB. In fact, no matter what settings I put in there or how I change the numbers, if there is ANYTHING in the Nvidia config file, all tasks fail after just a few seconds of run time. This GPU runs other projects without any trouble and isn't overheating or throttling, and it runs Collatz OK (but slow) without the optimization settings so I'm not sure what is causing the failures. These are the settings I started with, and have tried raising and lowering the kernels, threads, lut_size, and sieve_size individually but haven't managed to get a single successful work unit completion unless the config file is completely blank. I also tried it with sleep and cache_sieve at 0 and 1 with no change...

verbose=0
kernels_per_reduction=48
sleep=0
threads=10
lut_size=17
reduce_CPU=0
sieve_size=30
cache_sieve=1
ID: 3147 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile Farscape
Avatar

Send message
Joined: 21 Sep 09
Posts: 5
Credit: 35,217,840,774
RAC: 58,404,285
Message 3148 - Posted: 9 Feb 2021, 3:58:42 UTC - in response to Message 3147.  

I had a similiar problem with a 1660 Super - I fixed the issue (by accident) by playing with threads and lut size.

This is what I am running in the 1660S and getting (computer ASUS-P5E-WS-Pro) and getting 375 sec average without overclocking:

verbose=0
kernels_per_reduction=48
sleep=0
threads=7
lut_size=17
reduce_CPU=0
sieve_size=30
cache_sieve=1
ID: 3148 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Fardringle

Send message
Joined: 21 Jul 09
Posts: 6
Credit: 178,017,712
RAC: 203
Message 3149 - Posted: 9 Feb 2021, 4:05:52 UTC - in response to Message 3148.  

Thanks for the response. I didn't keep track of every exact setting I tried so I'm not sure if I used those, but I'll give them a try and see what happens. :)
ID: 3149 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Fardringle

Send message
Joined: 21 Jul 09
Posts: 6
Credit: 178,017,712
RAC: 203
Message 3150 - Posted: 9 Feb 2021, 6:58:16 UTC - in response to Message 3149.  

The tasks do actually run with those settings, so I either didn't try those specific numbers, or something else changed since the last time I tried. However, they don't seem to have any real effect on the task run times for this GPU. Still getting the same 13-15 minutes range (780-900 seconds) as it gets with a blank .config file. Maybe that's just the best that this card can do. It's not a big deal, but I don't like mysteries that don't make sense.. ;)
ID: 3150 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile marsinph

Send message
Joined: 18 Feb 18
Posts: 19
Credit: 2,456,322,513
RAC: 5,987,579
Message 3151 - Posted: 9 Feb 2021, 9:58:05 UTC - in response to Message 3150.  

The tasks do actually run with those settings, so I either didn't try those specific numbers, or something else changed since the last time I tried. However, they don't seem to have any real effect on the task run times for this GPU. Still getting the same 13-15 minutes range (780-900 seconds) as it gets with a blank .config file. Maybe that's just the best that this card can do. It's not a big deal, but I don't like mysteries that don't make sense.. ;)




Hello,
No, you not run the optimozed !
Look the detail from one of the latest valid WU.

Yes you card can do much better. I have the same GPU. It runs in 550 sec.
Try the following settings :
verbose=1
kernels/reduction=48
threads=8
lut_size=17
sieve_size=30
sleep=1
cache_sieve=1
reduce_cpu=0

Be very carefull : copy it in a notepad (not word)
Save ti under : collatz_sieve_1.30_windows_x86_64__opencl_nvidia_gpu.config
Yes the extension is "config" (the extension "txt" from notepad need to be removed)
Of course in folder C:\ProgramData\BOINC\projects\boinc.thesonntags.com_collatz
Not need to restart Boinc. It will run at the next WU.
Enjoy
ID: 3151 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 . . . 8 · 9 · 10 · 11 · 12 · Next

Message boards : Number crunching : Optimizing the apps


©2022 Jon Sonntag; All rights reserved