BOINC Pentathlon 2023

[SkillzTA]

6 minutes ago, Justaphf said:

The hosts that were affected were the following two:

* https://einsteinathome.org/host/13140397

* https://einsteinathome.org/host/13140395

 

It seems that after I aborted some 700+ WU that had stacked up on the following two hosts the first two listed started getting huge amounts of tasks.  I ended up reducing the download amounts to 0.5/0.25 on each computer after they started downloading 200-500 at a time each (and aborted a bunch of them to free up for others).

* https://einsteinathome.org/host/13140172 - about 250 aborted, this has a single 3080ti and is concurrently running yoyo on a 7900x

* https://einsteinathome.org/host/13140198 - almost 600 aborted, this has a single 1080ti and is concurrently running yoyo on an 8th gen i7

 

The next time it happens, go to those pages and click the link that's Last time contacted server: [DATE]

 

In this log, look for this line:

 

2023-05-17 02:17:07.3103 [PID=3464089]    [send] CUDA: req 141321.92 sec, 0.00 instances; est delay 0.00

 

Mostly you need to search for req and find the line that looks similar to this.

 

If it looks more like this

 

2023-05-17 02:17:07.3103 [PID=3464089]    [send] CUDA: req 1 sec, 0.00 instances; est delay 0.00

 

Notice it's only requesting 1 sec, then it's the same problem we have been noticing as well. If this happens. Do what I suggested above and it will fix it temporarily. Well, it should fix it temporarily. 

[magnavoxfull]

I just managed to get my einstein working correctly so it looks like my last day and a half of einstein running were a waste.

It wasnt putting any real load on my gpu with ~30 minute task times and only 10w above idle power draw on a 7900xtx but when i set the einstein project preferences to use .5 gpu per project my power draw went up to 400 and its completing 2 tasks ever 3.5 minutes.

 

It also looks like this was affecting my arc a770 and my 6700xt to a certain extent but its not a big of a change with those ones and its only around 2x as fast instead of 10x as fast.

 

Had anyone else ran into this yet or was I just unlucky?

[Justaphf]

10 minutes ago, SkillzTA said:

 

The next time it happens, go to those pages and click the link that's Last time contacted server: [DATE]

 

In this log, look for this line:

 

2023-05-17 02:17:07.3103 [PID=3464089]    [send] CUDA: req 141321.92 sec, 0.00 instances; est delay 0.00

 

Mostly you need to search for req and find the line that looks similar to this.

 

If it looks more like this

 

2023-05-17 02:17:07.3103 [PID=3464089]    [send] CUDA: req 1 sec, 0.00 instances; est delay 0.00

 

Notice it's only requesting 1 sec, then it's the same problem we have been noticing as well. If this happens. Do what I suggested above and it will fix it temporarily. Well, it should fix it temporarily. 

I just checked and the two that had the problem are both requesting 1 second of work as you show, however, they have both already downloaded about ~600 WUs each so they are ok for a few hours.  I further reduced the amount to download to 0.1/0.01 as well to prevent too large a backlog from downloading.  Hopefully that gets them to a not-too-greedy stable state.

[SkillzTA]

8 minutes ago, magnavoxfull said:

I just managed to get my einstein working correctly so it looks like my last day and a half of einstein running were a waste.

It wasnt putting any real load on my gpu with ~30 minute task times and only 10w above idle power draw on a 7900xtx but when i set the einstein project preferences to use .5 gpu per project my power draw went up to 400 and its completing 2 tasks ever 3.5 minutes.

 

It also looks like this was affecting my arc a770 and my 6700xt to a certain extent but its not a big of a change with those ones and its only around 2x as fast instead of 10x as fast.

 

Had anyone else ran into this yet or was I just unlucky?

That sounds like no tasks were running on the GPU when you looked at the power draw. 1 task will still draw a decent amount of power.

[magnavoxfull]

5 minutes ago, SkillzTA said:

That sounds like no tasks were running on the GPU when you looked at the power draw. 1 task will still draw a decent amount of power.

I think it was using the gpu atleast somewhat but i wouldnt be suprised if it was mostly just using the 1 assigned cpu thread to do all of the work. the gpu was also saying around 25% utilization though it was mostly at idle clocks.

[magnavoxfull]

Also does anyone know if its possible to use the same App Config trick we used with Number fields on Einstein instead. I'm hopeful that if I was able to set it at .5 cpu and .5 gpu it would be more likely to run both tasks since I'm struggling to get my 6700xt to run 2 tasks if Yoyo is also running.

 

I tried changing the file myself but I'm guessing I did it wrong because the directory is much harder to read compared to number fields.

[Justaphf]

Looks troubling for Obstacle Run...anyone got any bunkers they forgot to release?

[leadeater]

9 minutes ago, Justaphf said:

Looks troubling for Obstacle Run...anyone got any bunkers they forgot to release?

Forgot? No. Haven't done so yet, yes 

[leadeater]

@SkillzTA I'll say it here because damn I hate when the shoutbox on the event page gets all salty and toxic like it is.

 

At least for our own forum run F@H events renting or putting in the effort to use free cloud spot instances etc etc is not only allowed but encouraged. If someone is willing to put in the effort to utilize free cloud resources then we encourage that. Put in the effort then you get the reward, simple.

 

And you're right, there is no difference between buying 30  Nvidia GPUs or renting them, money was exchanged.

 

Sighhhhh, I really hate the complaining. Like a lot.

 

[912falcon]

What are the best preferences for Einstein @ home and how to change them?

[justpoet]

28 minutes ago, leadeater said:

@SkillzTA I'll say it here because damn I hate when the shoutbox on the event page gets all salty and toxic like it is.

 

At least for our own forum run F@H events renting or putting in the effort to use free cloud spot instances etc etc is not only allowed but encouraged. If someone is willing to put in the effort to utilize free cloud resources then we encourage that. Put in the effort then you get the reward, simple.

 

And you're right, there is no difference between buying 30  Nvidia GPUs or renting them, money was exchanged.

 

Sighhhhh, I really hate the complaining. Like a lot.

 

I also dislike the salty complaining.  I can, however, see the "only use what you have" aspect of things too.  To me, this just sounds like the rules need to be more clearly written for the future.  It might be as simple as replacing "systems" with "teams", as it sounds like there was basically a banano like team running in boinc where you did it to get coins.  That's certainly not the same as using free (or even paying for) cloud services.  So, if cloud services shouldn't be used, that's not covered in the current rules, though I can see how somebody might misinterpret what is written that way.

[justpoet]

56 minutes ago, Justaphf said:

Looks troubling for Obstacle Run...anyone got any bunkers they forgot to release?

I just got home.  I was going to wait a bit longer on mine, but since it sounded like the server might not be happy, and I'm not going to have time to fiddle with it later, I just went for it.  No issues...though I'm also not a super heavy hitter.

[leadeater]

30 minutes ago, justpoet said:

I also dislike the salty complaining.  I can, however, see the "only use what you have" aspect of things too.  To me, this just sounds like the rules need to be more clearly written for the future.  It might be as simple as replacing "systems" with "teams", as it sounds like there was basically a banano like team running in boinc where you did it to get coins.  That's certainly not the same as using free (or even paying for) cloud services.  So, if cloud services shouldn't be used, that's not covered in the current rules, though I can see how somebody might misinterpret what is written that way.

Same, I'm empathic of the opinion but that's all it is. From what little I know yes there was such an instance of payout for work and that I don't agree with. If the rules need updating or being more specific then let that happen.

 

I for one like to see how I and we can do, how we improve each year etc. I don't mind if that's 1st or 20th 

 

Spoiler

Also it's completely within my permissions and ability to throw a bunch of BOINC workload at our national Cray supercomputer cluster, use tens of thousands of CPU cores and hundreds of A100 GPUs, as well as install BOINC client on ~10,000 desktop PCs with minimum xx700 Intel CPUs or equivalent AMD and higher. I don't because that's obviously wrong, not fair and I have moral sense of what is right. Plus I like to keep my job. But I can totally create something to actually legitimately complain about.

Every year there is event feedback. If someone has something to say or thinks there is a problem use that not the damn shoutbox. Aaaaaaaaaaaa

 

[leadeater]

2 hours ago, magnavoxfull said:

Also does anyone know if its possible to use the same App Config trick we used with Number fields on Einstein instead. I'm hopeful that if I was able to set it at .5 cpu and .5 gpu it would be more likely to run both tasks since I'm struggling to get my 6700xt to run 2 tasks if Yoyo is also running.

 

I tried changing the file myself but I'm guessing I did it wrong because the directory is much harder to read compared to number fields.

You can, but you can do it directly from the website in the preferences.

[Favebook]

9 hours ago, Chachunka said:

Not sure if it was mentioned earlier, but congrats @Favebook!

 

Thanks - didn't even know

[Ithanul]

7 hours ago, leadeater said:

@SkillzTA I'll say it here because damn I hate when the shoutbox on the event page gets all salty and toxic like it is.

 

At least for our own forum run F@H events renting or putting in the effort to use free cloud spot instances etc etc is not only allowed but encouraged. If someone is willing to put in the effort to utilize free cloud resources then we encourage that. Put in the effort then you get the reward, simple.

 

And you're right, there is no difference between buying 30  Nvidia GPUs or renting them, money was exchanged.

 

Sighhhhh, I really hate the complaining. Like a lot.

 

It ain't the Pent until some whining occurs in that shoutbox.  I think in everyone I ever been in, there is always some drama going to occur in that shoutbox.

 

Either someone going to say a team is cheating or someone going to take offense about what some folks can afford to have and run.  smh

[SkillzTA]

8 hours ago, magnavoxfull said:

Also does anyone know if its possible to use the same App Config trick we used with Number fields on Einstein instead. I'm hopeful that if I was able to set it at .5 cpu and .5 gpu it would be more likely to run both tasks since I'm struggling to get my 6700xt to run 2 tasks if Yoyo is also running.

 

I tried changing the file myself but I'm guessing I did it wrong because the directory is much harder to read compared to number fields.

 

You will need to limit the number of running yoyo tasks using max concurrent in an app_config.xml file in the yoyo@home projects directory. About to go to work or I'll write it up for you.

 

7 hours ago, leadeater said:

@SkillzTA I'll say it here because damn I hate when the shoutbox on the event page gets all salty and toxic like it is.

 

At least for our own forum run F@H events renting or putting in the effort to use free cloud spot instances etc etc is not only allowed but encouraged. If someone is willing to put in the effort to utilize free cloud resources then we encourage that. Put in the effort then you get the reward, simple.

 

And you're right, there is no difference between buying 30  Nvidia GPUs or renting them, money was exchanged.

 

Sighhhhh, I really hate the complaining. Like a lot.

 

 

People complain every year about everything in the Pentathlon. Nothing changes.

 

6 hours ago, 912falcon said:

What are the best preferences for Einstein @ home and how to change them?

 

Running two tasks at once on most Nvidia GPUs and using the special app under Linux. Only run gamma ray tasks.

6 hours ago, justpoet said:

I also dislike the salty complaining.  I can, however, see the "only use what you have" aspect of things too.  To me, this just sounds like the rules need to be more clearly written for the future.  It might be as simple as replacing "systems" with "teams", as it sounds like there was basically a banano like team running in boinc where you did it to get coins.  That's certainly not the same as using free (or even paying for) cloud services.  So, if cloud services shouldn't be used, that's not covered in the current rules, though I can see how somebody might misinterpret what is written that way.

What they don't realize is that our team has a ton of hardware at home that we own as well. I got multiple EPYC builds, over 20 GPUs, and I'm not even the biggest on my team with hardware owned at home. One guy even owns two Genona EPYC setups now.

[Ithanul]

Hmmmm, we may need to shore up in yoyo.  SETI.Germany really outputting over there.

 

[Idiot_icLoser55]

12 minutes ago, SkillzTA said:

What they don't realize is that our team has a ton of hardware at home that we own as well. I got multiple EPYC builds, over 20 GPUs, and I'm not even the biggest on my team with hardware owned at home. One guy even owns two Genona EPYC setups now.

What do you use them for? Is it all just boinc and folding or is there something else.

[leadeater]

15 minutes ago, Idiot_icLoser55 said:

What do you use them for? Is it all just boinc and folding or is there something else.

For mine I use it for development and training work. Building networks that are infrastructure equivalent to medium/large networks. So I have Fortigate firewalls, 48x SFP+ 4x QSFP+ switch stacks, big CPU heavy servers running ESXi and vCenter, AD Domain(s), SQL clusters, SCCM/SCOM, Foreman.... blah blah etc etc. All sorts of stuff.

 

It just so helps that hardware wise the servers work really well for BOINC when needed.

[SkillzTA]

23 minutes ago, Idiot_icLoser55 said:

What do you use them for? Is it all just boinc and folding or is there something else.

 

Dedicated BOINC hosts for the most part.

 

I also host my own Plex Media server at home, as well as storage and some other personal services that some of the hosts run.

 

Majority of them just run BOINC strictly though.

[Captainmarino]

9 hours ago, Justaphf said:

Looks troubling for Obstacle Run...anyone got any bunkers they forgot to release?

[912falcon]

For some reason einstein @ home is using my integrated graphics instead of my 3060. how do you fix this?

[Idiot_icLoser55]

10 minutes ago, 912falcon said:

For some reason einstein @ home is using my integrated graphics instead of my 3060. how do you fix this?

If you go to the Einstein webpage and the computing preferences you can select which device architecture is being used. In this case you want Nvidia checked and Intel GPU unchecked.

[nadoteg]

 

Holy mom's spaghetti! What a race to the end!