9th BOINC Pentathlon hosted by SETI.Germany

[GOTSpectrum]

27 minutes ago, palespartan said:

For my three gpus would they bet better put to work on primegrid or on astroids I don't mind having to put in the work to optimize for either one but wont be able to have access to my computers to change them out after they are all nvidia so its just a matter of what the better strategy is.

I think we are better off trying to push high on just one, lets try Primegrid.

 

Then CPUs on the other projects you feel best on our chances. me I'm throwing at marathon and sprint. 

[porina]

To my thinking, Asteroids isn't ideal for a GPU in that you could be competitive on CPUs alone. PrimeGrid is GPU all the way if you're after credit. Of course, everyone will be in the same boat so it will still be a case of how much power we can throw at it relative to other teams. The GPU credit is high as credit was originally set using a reference CPU. It turned out the GPU app when it became available was a LOT faster than CPU, hence the massive credit.

 

While PrimeGrid is my pet project, the GPU ones are not my focus as you can tell from my CPU farm. I think it prefers nvidia, but don't quote me on that, and I think AMD is still good too. I need to check my driver status at some point before it starts so will use that opportunity to get some benching done.

 

BTW I'm with another team on PrimeGrid. After discussion it seems the easiest way for me to participate is to move to the LTT team temporarily for the duration of the event, then move back to my old team. 

[Ithanul]

13 hours ago, porina said:

Ooh, just saw the cross county event will be PrimeGrid. GPUs on standby! PPS-sieve sub-project on GPU is relatively high credits-per-time so will probably be the focus for most. Possibly so much faster it doesn't make much sense to run anything else. CPUs might be best put on other events. You will want to check the project selection and other settings as appropriate. I think it is default for new joiners, but if you joined in the distant past other sub-projects may be selected.

 

If you only run BOINC the following doesn't matter, but they are implementing anti-bunkering on PSA. If you don't know what it is, don't worry about it.

Yep PPS-sieve units, fair warning though: toasty, toasty, toasty.  Last time they had PrimeGrid for the Pent, my room was sauna level heat.

[porina]

3 minutes ago, Ithanul said:

Yep PPS-sieve units, fair warning though: toasty, toasty, toasty.  Last time they had PrimeGrid for the Pent, my room was sauna level heat.

As far as I'm concerned, that's just GPUs normally! And in part why I like to crunch with CPUs.

[Ithanul]

Just now, porina said:

As far as I'm concerned, that's just GPUs normally! And in part why I like to crunch with CPUs.

My GPUs to tend run hotter doing PrimeGrid sieve units.  E@H and F@H don't put near that heat level out.

 

Though, usually my room is near sauna or sauna level.  Lucky enough I am very tolerant of heat.

[porina]

3 minutes ago, Ithanul said:

My GPUs to tend run hotter doing PrimeGrid sieve units.  E@H and F@H don't put near that heat level out.

 

Though, usually my room is near sauna or sauna level.  Lucky enough I am very tolerant of heat.

You're not going to like PG CPU tasks much either then Most of them that run on LLR are essentially like running Prime95. 

 

I'm definitely not tolerant of heat. Above 25C ambient or so, I go into thermal throttling.

[Ithanul]

7 minutes ago, porina said:

You're not going to like PG CPU tasks much either then Most of them that run on LLR are essentially like running Prime95. 

 

I'm definitely not tolerant of heat. Above 25C ambient or so, I go into thermal throttling.

Lmao, you probably best to never come down to the south lands of the US in Summer then.

It is double whammy with the humidity levels here.

 

25C?  77F?  Darn, that cool weather here.  I usually only get a bit uncomfortable around 85-90F/29C-32C.

Sometimes outside temps can go beyond 100-115F/37-46C.

 

They don't call it sub-tropic down here for nothing.

[GOTSpectrum]

4 hours ago, zzrhardy said:

Yes. 4 concurrent work units per card seems to be the sweet spot for both the Titan V and the Vega 64's.Testing suggests Kuroneko and Chibineko will be able to spit out 2.5 PPS-sieve work units per minute between them, not counting CPU chump change.

 

At 24 hours before PrimeGrid start time I will stop other events as the CPU only projects like to cannibalise CPU capacity from GPU work units and stop them running.

Coupla hints in the spoiler below, but you guys will likely know this stuff already.

  Hide contents

 

- Suggested app_config.xml to use as a starting point attached to this post if anyone else wants to run multiple GPU work units through each cards, as per the instructions HERE.

- As porina mentioned mentioned, make sure to just select the "PPS-sieve (Proth Prime Search Sieve (PPS-Sieve)" sub project in the Edit Primegrid preferences as they are overpowered for credit awarded compared to other sub projects.

 

 

app_config.xml

I'm highly confused on how to get my Vega to run multiple WUs at the same time 

[GOTSpectrum]

Also does anyone know how to make boinc stockpile more WUs?

[Uzume]

3 hours ago, Ben Quigley said:

I'm highly confused on how to get my Vega to run multiple WUs at the same time 

Exactly the same way as getting the nVidia cards to process multiple work units. You need to have the app_config.xml file in your project directory. The default location for your project directory is as follows (of course drives may vary and you may have specified an alternative when installing BOINC):

 

c:/ -> ProgramData (this directory hidden by default) -> BOINC -> Data -> Projects -> www.primegrid.com

 

Pop that sample app_config.xml file in that I attached to the last post or make a new file with that name. The sample settings I had in that file are as follows, but they can be changed with text editor:

 

<app_config>
<app>
<name>pps_sr2sieve</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.5</cpu_usage>
</gpu_versions>
</app>
</app_config>

 

In the above example setting the gpu_usage tag to 0.25 means 4 units will run concurrently. Setting it to 0.5 will mean 2 units will run concurrently and so on. 1 equals 1 unit.

 

Setting the cpu_usage tag to 0.5 will mean each concurrent GPU work unit will have that amount of reserved CPU thread out of the total percentage you have allowed the entire project. The amount required changes from GPU to GPU but you should enter the figure the BOINC client normally assigns when processing a work unit. Less will gimp the GPU processing whilst more will just be wasted. BOINC assigns 0.162 CPU to my Vega work units so I set this setting to 0.25 on this box. BOINC assigns 0.58 CPU to my Titan V work units so I set it to 0.75 on that box (I set a little leeway).

 

After your app_config.xml file is set as desired, restart your BOINC client to make it read the config file and you should see it processing multiple work units per GPU. If not, make sure the file type for app_config.xml file is XML and not Text. The example I attached is already correct, it will just need text editing before using. 

 

The above settings are identical whether using nVidia or Radeon but different cards will have different CPU % needs and have limitations on how many work units they can process at a time based on VRAM capacity. Also once you saturate your card compute power completely, you get no further gain. Further increasing concurrent WU's just increases time per work unit in a linear fashion.

 

it is possible to get real fancy with your app_config.xml to account for differing GPU's, but I am a trogladyte and keep my cards uniform and file basic. You will need a lot of google fu and messing around to get a mixed GPU setup working with different settings per GPU. Ie, GPU0 and GPU1 tagged sections. Messy.

 

By the way the above app_config.xml file can be used for other projects but the name tag needs to be correct. pps_sr2sieve is the correct name for what we are planning to do with pps-sieve, but if you wanted to process another project or sub project then you need to source the correct application name.

 

BOINC will stockpile the maximum work units allowed per project automatically as long as you have your "store at least X days of work" and "store at least an additional X days of work" preference set high. I have mine set to 10 days each but I think this is well above what BOINC needs to grab the max amount of work units. To be honest I am not even sure BOINC checks this preference anymore. Most projects allow 100 work units and PrimeGrid is normally the same, but the admin has temp raised the limit to 1000 for the duration of the event to allow bunkering.

 

For others reading, the bunkering tactic means setting up the project and download all your assigned work units, then in BOINC under "activity" choosing "suspend network activity". This means you work through your entire queue without trying to submit results as you go. Once the contest opens, then un-suspend the network activity and dump all your pre-processed work units within the contest window.

 

Expect everyone to be doing this and the project's backend server to prolapse in the opening hours of the event....

 

Let m know if you have issues or I have written stuff in a confusing way.

[Uzume]

5 hours ago, Ithanul said:

My GPUs to tend run hotter doing PrimeGrid sieve units.  E@H and F@H don't put near that heat level out.

 

Though, usually my room is near sauna or sauna level.  Lucky enough I am very tolerant of heat.

 

Ditto. My two boxes normally pull 1300 watts under Seti load but they are pulling 1700 watts under PrimeGrid sieve unit load.

[GOTSpectrum]

3 hours ago, zzrhardy said:

 

Ditto. My two boxes normally pull 1300 watts under Seti load but they are pulling 1700 watts under PrimeGrid sieve unit load.

Do you know what I'm doing wrong here?

app_config.xml

 

EDIT: The file doesn't seem right, not the configuration but the actual filetype. 

[porina]

8 hours ago, Ithanul said:

25C?  77F?  Darn, that cool weather here.  I usually only get a bit uncomfortable around 85-90F/29C-32C.

Sometimes outside temps can go beyond 100-115F/37-46C.

25C is my thermal throttle point. 30C and it is emergency shutdown. Not used to it, as we don't often encounter such. Ok, 25C in daytime isn't unusual on a clear day in summer, but 30C is uncommon, and outside of unusual events you don't get much above that.

 

It's all relative I guess. When I was working in Japan a long time ago, there was snow on the ground from previous fall but the air temp wasn't cold during the day. I was walking around in a T-shirt quite happily. I had observed the locals were all wearing thick winter coats. Didn't think much of it. As I was walking back to the apartment block I was staying in I bumped into the landlady. She didn't speak English, I didn't speak Japanese. We still somehow had a conversation that concluded, no, I wasn't cold. It was nice for me at least.

[Uzume]

2 hours ago, Ben Quigley said:

Do you know what I'm doing wrong here?

app_config.xml

 

EDIT: The file doesn't seem right, not the configuration but the actual filetype. 

 

First line of the configuration opens with <app_config> so last line needs to close with </app_config>

 

The file works when I edit in the closing tag, so the type is right. Attached the corrected file.

app_config.xml

[Uzume]

7 hours ago, porina said:

We still somehow had a conversation that concluded, no, I wasn't cold. It was nice for me at least.

I am guessing she included the word "genki" quite a lot.

 

Walked around Japan in shorts and t-shirt in winter also and lot of people would say "genki!" to me. I don't speak Japanese but knew it meant "healthy" so I would shake my head, point to myself and say "aho" (idiot). 

[GOTSpectrum]

2 hours ago, zzrhardy said:

 

First line of the configuration opens with <app_config> so last line needs to close with </app_config>

 

The file works when I edit in the closing tag, so the type is right. Attached the corrected file.

app_config.xml

Damn I feel stupid, butt TBF it was around 5am I was working on it. The file does exactly what I needed, The vega crunching three at a time while the NV cards are just doing one each. As if often the case with computers if you're not exactly sure how to fix the problem there is almost always a quick and dirty way to go it. 

 

Thanks man for seeing my blunder! Now its time to crunch!

[GOTSpectrum]

Guys can you transfer 25-50% of your swimming threads to sprint? 

[Nostremitus]

2 hours ago, Ben Quigley said:

Guys can you transfer 25-50% of your swimming threads to sprint? 

I started my wife's 16 threads last night, but she may have forgotten and shut her computer down after her morning browsing session. I'll check when I get off work.

[porina]

2 hours ago, Ben Quigley said:

Guys can you transfer 25-50% of your swimming threads to sprint? 

Swimming ends if under 6 hours at time of writing so should be looking at switching anyway. We seem to be doing reasonably on sprint so let's see if that momentum keeps up. Good number of points there.

[Ithanul]

16 hours ago, zzrhardy said:

 

Ditto. My two boxes normally pull 1300 watts under Seti load but they are pulling 1700 watts under PrimeGrid sieve unit load.

Darn, what the heck you have pulling 1700 watts?  Here I thought my 1200 and 1300 watt PSUs where big.

I'm probably going to have to get a bigger PSU for my 1950X build.  Not sure if the current 1300W unit will handle a tuned in 1950X with tuned in Titan Xps, once I get it all setup.  I know on my next parts buying is investing into a bigger UPS to handle my systems.

[Nostremitus]

I have enough Swimming waiting to validate, I just switched everything I have over to Sprint.

 

That's 54 threads, but 10 of them are ARM.

[GOTSpectrum]

2 hours ago, porina said:

Swimming ends if under 6 hours at time of writing so should be looking at switching anyway. We seem to be doing reasonably on sprint so let's see if that momentum keeps up. Good number of points there.

I meant add that to sprint then distribute the rest as you see fit. 

[christian1122]

hi all, im new to Boinc

any recommendations on what is a good project for CPU?

i use my GPU for F@H

[GOTSpectrum]

31 minutes ago, christian1122 said:

hi all, im new to Boinc

any recommendations on what is a good project for CPU?

i use my GPU for F@H

What CPU do you have? And if you're looking to support the team we could do with more helping out on numberfields@home. 

[Uzume]

du

3 hours ago, Ithanul said:

Darn, what the heck you have pulling 1700 watts?  Here I thought my 1200 and 1300 watt PSUs where big.

I'm probably going to have to get a bigger PSU for my 1950X build.  Not sure if the current 1300W unit will handle a tuned in 1950X with tuned in Titan Xps, once I get it all setup.  I know on my next parts buying is investing into a bigger UPS to handle my systems.

 

Ahh, 1700 watts combined, not individually. I have a wattmeter on the power board feeding both systems so that is what I read off.