News:

Members can see all forum boards and posts. Non members can only see a few boards.  If you have forgotten your password use this link to change it.
https://forum.boinc-australia.net/index.php?action=reminder

Main Menu

PPS-Sieve app_config file

Started by chooka03, November 24, 2019, 05:01:39 PM

Previous topic - Next topic

chooka03

Sorry but does anyone have this? To enable more than 1 WU @ a time running.
I'm sure it's something simple but it's a file I surprisingly don't have.

Would it be this? This is my closest guess.

<app_config>
    <app>
        <name>PPS-Sieve</name>
        <gpu_versions>
            <gpu_usage>0.5</gpu_usage>
            <cpu_usage>0.2</cpu_usage>
        </gpu_versions>
    </app>
</app_config>

If I add this to my current Primegrid app_config, will it just work?

chooka03

I cracked it! (Well no I didn't...I found it :)

<app>
<name>pps_sr2sieve</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.2</cpu_usage>
</gpu_versions>

In case anyone else comes looking.
Where's the "BOO its nearly Monday" emoji  :cry

chooka03

What do others use when crunching PPS Sieve work? Do you run more than 1 wu at a time?
It looks like with 1 wu running, there's bugger all activity on the GPU.

Dingo

#3
I had a look at my app_config and I do not have an entry for any GPU work so I must run one at a time.  I ran one just to see in CAM what it was doing and it was using between 67 - 70% for one task om a 1080Ti and it ran for 3 minutes 43 seconds.   :BigGrin

If you want to run some long tasks give Manual Sieving a try on Primegrid.   You start off with small tasks but as you return work successfully you have the ability to increase the size of the work.  http://www.primegrid.com/show_sieving.php  There is a how to page that you should read"Introduction to Manual Sieving" and the work gets credited towards the PSA Badge.  You basically
1. create a folder. 
2. Download the work and the app into the folder. 
3. Run the work in a command line box. 
4. Send the results back on the web page provided.


Radioactive@home graph
Have a look at the BOINC@AUSTRALIA Facebook Page and join and also the Twitter Page.

Proud Founder and member of BOINC@AUSTRALIA

My Luck Prime 1,056,356 digits.
Have a look at my  Web Cam of Parliament House Ottawa, CANADA

chooka03

Thanks Dingo  :thumbsup:
Yeah the AMD cards aren't that great for PG. I briefly mucked around with multiple WU's and reserving cores. Times were anywhere from 14min 30 down to 10mins.
1 WU put bugger all load on the GPU. I did notice when I cut my cores in half (50% CPU usage) and ran x 3 WU's, the load was consistent. In the end I gave up. I said I wasn't going to do any GPU work over summer. I need to try and stick to that. Too hot and getting too expensive 24/7.

I'll revisit this at a later date but improving my PG rank is on the cards.

chooka03

Does anyone have a working app_config for AMD cards?
My old one doesn't seem to work any more?

Thx  :thumbsup:

chooka03

All good.
Found it

<app_config>
<app>
   <name>pps_sr2sieve</name>
   <report_results_immediately/>
   <fraction_done_exact/>
</app>
<app_version>
    <app_name>pps_sr2sieve</app_name>
    <plan_class>openclatiPPSsieve</plan_class>
    <avg_ncpus>0.0001</avg_ncpus>
    <ngpus>0.5</ngpus>
</app_version>

</app_config>

turbo0

Thanks chooka im stealing this. Will have to have a play around to see best utilisation settings I guess.
Desktop 1 - Ryzen 9 7950x, RTX 4070ti, 64GB DDR5 6000,
Desktop 2 - Ryzen 7 5700x, RTX 3060ti, 16gb DDR4-3200,
Laptop - Ryzen 7 4800H, GTX 1660ti, 16gb DDR4-3200
SFF Dell - i7 8700, RX 6400, 16GB DDR4,
SFF Dell - i7 7700, Tesla P4, 16GB DDR4,
SFF Dell - i5 7500, GT 1030, 16GB DDR4
SFF HP - i7 3770, GTX 1650, 16GB RAM
Micro PC - N100, 12GB 4800mhz Ram


Folding@Home Stats

chooka03

Yeah I find it hard because I have 2 GPU's the same but in different pc's. Because I'm also crunching an LLR task and multithreading, it can get complicated.
I think I will pause the LLR work during the 4 or 5 day challenge.
Which brings me to a question please..... Kashi may know this?

In the config file, what effect will changing this line do?

ncpus>0.0001</avg

If I'm not using my cpu cores, could I change this to say 0.5cpu cores. Would that have any impact?

Anyone know? :)

Thank you

Dingo

I added this to my app_config.xml to run 2 cuda tasks at a time for pps_sieve on my 1660Ti's.  Running two at a time runs the GPU at 98% and 68 degrees.

<app>
   <name>pps_sr2sieve</name>
   <report_results_immediately/>
   <fraction_done_exact/>
</app>
<app_version>
    <app_name>pps_sr2sieve</app_name>
    <plan_class>cudaPPSsieve</plan_class>
    <avg_ncpus>0.0149</avg_ncpus>
    <ngpus>0.5</ngpus>
</app_version>


Radioactive@home graph
Have a look at the BOINC@AUSTRALIA Facebook Page and join and also the Twitter Page.

Proud Founder and member of BOINC@AUSTRALIA

My Luck Prime 1,056,356 digits.
Have a look at my  Web Cam of Parliament House Ottawa, CANADA

kashi

Hey chooka.
Delayed reply, doc is heavily booked and I'm overdue for a top up of Saturn V rocket fuel, starting to get a bit hot and bothered, etc.
Back to full love power in a week's time though, yay!  :boom:

Yes avg_ncpus is defined as "the number of CPU instances (possibly fractional) used by the app version".

I used to use it to reserve a core or more to help support/speed up GPU processing. So if I wanted to allocate a CPU core to each GPU task and I was running 2 GPU tasks concurrently on the one GPU then I would use <avg_ncpus>2</avg_ncpus>. It had the advantage over just reducing your number of BOINC CPUs on GPU projects where supply of GPU tasks was intermittent or variable. If computer ran out of GPU tasks then extra CPU task(s) would automatically start up. At other times I used to also turn off hyperthreading so that contention losses were reduced as this would help speed up both GPU tasks and the reduced number of CPU tasks that were running.

Many people back when GPU computing was first starting to be used in BOINC were reluctant to allow any CPU resources to support GPU crunching or they were unaware of the need. Hence the speed of their GPU processing was reduced, sometimes very greatly. At one time nVidia GPUs automatically allocated a full CPU core/thread to support GPU crunching to prevent this slowdown, whereas ATI/AMD GPUs did not.

The choice of whether to speed up GPU processing or run more CPU tasks is personal and also affected by the relative efficiency of the GPU app. Some GPU apps use almost no CPU resources and others are very much affected by any running CPU tasks. When I ran GPUGrid for example I ended up running no CPU tasks at all. My reasoning was that the extra efficiency of the GPUGrid GPU app compared to CPU apps means that I was doing more work overall when it was running at its fastest. And also using less power and running cooler. And earning more credit too!

Haven't tried the newer GPUGrid applications though, when I tried to run GPUGrid recently for a challenge I couldn't get any tasks, think I need to update my GeForce drivers.

Other people are prepared to compromise on maximum GPU efficiency when running this kind of CPU resource hungry GPU app. It may be that they have goals they wish to reach on particular CPU projects or believe they are "wasting" CPU resources unless their computer is at high or full load. Currently many choose to prioritise SARS-CoV-2 research above all other research areas and I understand that completely. I choose to still participate in Team Challenges when I can, but revert to WCG OpenPandemics - COVID-19 in between. Well I will as soon as I get Ruby Africa Rainfall badge very soon.

For your specific question, I don't remember how much the speed of PrimeGrid Sieving GPU app is affected by the amount of CPU resources allocated. If you aren't crunching any CPU tasks at all, I believe changing the avg_ncpus value will not make any difference as all the CPU resources are already available to support GPU crunching. avg_ncpus is a minimum value of CPU resources allocated to GPU processing not a maximum. I think the idea of using those small fractional default values for avg_ncpus is to prevent extreme slowdown or stoppage of GPU processing if people are running their computers on full load on other tasks.

Thanks for the nVidia version of app_config.xml for 2 concurrent Dingo, +1. Saves me having to remember/look up the specifics.

Good crunching everyone.  v:

Dingo

I have taken the code out of my ap_config.xml as the individual work unit time increased by almost double on my machines.  From average 228.5 to average 420.4 so that was a fail running two at atime.


Radioactive@home graph
Have a look at the BOINC@AUSTRALIA Facebook Page and join and also the Twitter Page.

Proud Founder and member of BOINC@AUSTRALIA

My Luck Prime 1,056,356 digits.
Have a look at my  Web Cam of Parliament House Ottawa, CANADA