BOINC Community Board

[leadeater]

8 minutes ago, Windows7ge said:

NGL you alone were probably the reason we pulled off 6th place last year. I love that we gave SETI.Germany a run for their money. (or I think it was SETI.Germany anyways).

 

Oh well. We'll manage.

A little bit, but the GPU projects are highly important and I contribute next to nothing to those. For that event I'd still try and wrangle something for the duration of it. I might have to more officialize the request to use the resources though, rather than just hiding it lol.

 

Edit:

Also TBH I'm not sure how much the other teams would actually like that I did/doing that, I don't think any of them really understood how much I was utilizing work resources. It's not like I was hiding it, but still kind of dickish in a way to just steamroll everyone in that way. If it becomes I real problem I'll just run the work stuff as it's own team, participate but request zero points for each event.

[Windows7ge]

Oh WTH I have the most posts in this thread!?!

 

 

Do I need to attend BOINC's Anonymous?

 

3 minutes ago, leadeater said:

A little bit, but the GPU projects are highly important and I contribute next to nothing to those. For that event I'd still try and wrangle something for the duration of it. I might have to more officialize the request to use the resources though, rather than just hiding it lol.

I'm looking forward to this years event. I swear I better have this EPYC CPU ready to go. Last year I swore I'd have a HPE DL560 Gen 8...that never happened. But who needs it if I have an EPYC 32C/64T processor.

[Gorgon]

On 3/19/2021 at 1:40 PM, Windows7ge said:

...

In all seriousness after this years BOINC Pentathlon you will have a very real shot at passing me in subsequent months. I experience all four seasons where I live and the summers get HOT. I can't run the servers/nodes full-tilt all year round as you seem to be able to. Expect my current output to be cut in less than half until the end of October~mid November.

...

I hear you. I also get wild extremes where I live. Summers are 25-35C and winters -25-40C.

 

I run my GPUs 18 hours a day just stopping them when the time of use electricity rate doubles during the week and full-time on the weekends. In the summer I have to offset the heat load of the GPUs with the air conditioning so at least the time of use periods align with the hottest periods so it’s not so bad but even then I drop all the GPUs down to minimum power targets.

 

In the winter I raise the power targets about 20-30% above the minimums and open a basement window about 1” at bedtime when I have to close the door to the unfinished portion of the basement where the systems live. This keeps the room from going up to over 30C by morning but I have to be careful as too much freezing air will freeze the pipes if the power goes out. During the day I run a 24” box fan suspended from the ceiling joists to push the heated air into the rest of the house.

 

I really need to install a fresh air intake with a blower to cycle based on temperature. That way if the power goes out the blower stops and there’s less risk.

 

The CPUs I run 24x7x365 as their power consumption is negligible compared to the GPUs.

 

When I was running Einstein on GPUs I did have a Cron job that suspended that project during the expensive electricity periods.

 

I was planning on running all my older Pascal GPUs on Einstein but with the Pandemic I decided to focus the GPUs on F@H and the CPUs on OpenPandemics and I sold all my 1060s for $150 Canadian each a few months ago so a few lucky people got a good deal.

[Windows7ge]

@Gorgon

To manage both the heat and electricity usage I plan on running my units 12 hours a day during the hours of the day the space isn't typically occupied.

 

I have two soon to be three nodes dedicated to BOINC. I plan to do something similar where these will be set to shutdown fully during the day then turn themselves on via RTC when they're due to run. I want to explore how I can improve the ease of scaling these nodes so I plan to construct a SAN where-in these nodes will boot off the network to iSCSI LUNs via iPXE. From here the physical nodes will just be containers for pre-constructed OS's that I can duplicate and configure on the side. As an added perk if a node catastrophically fails and takes hardware with it no work is lost.

 

Right now I've submitted the RMA papers for the supermicro motherboard and I"m awaiting my RMA number. It appears my only option is Advanced RMA which is a double edged sword. It will get me the answer faster but I'm out a lot of money if the return gets lost.

[mattheginger]

 Has anyone dabbled with boinc on the new mac M1's? Thinking of upgrading my macbook and just curious of its capabilities. I pretty much only do WCG on boinc.

[WhitetailAni]

I have joined!
With my Pentium III!

Because it's not going to do anything else.

Once I fix my iMac G5 I'll add that too.

[Windows7ge]

@Gorgon Not going to lie. Supermicro is seriously letting me down right now...

Quote

Supermicro’s limited warranty doesn't cover any failure(s) or defect(s) caused by misuse, accident, abnormal or unusual heavy use, improper packaging or handling, neglect, abuse, alteration, improper installation, unauthorized repair or modification, improper testing or causes external to the product such as, but not limited to, excessive heat or humidity, power failures, power surges or acts of God.

Because of the very small blemish on one of the corners (which is how I received it) Supermicro isn't even willing to test the board let-alone attempt a repair (although given the CPU situation the board is most likely fine).

 

This isn't actually my first time working with Supermicro. As a business I'm sure they're a fantastic supplier, but as an individual their tech support/RMA services are abysmal.

 

This is definitely the last motherboard I'm buying from them.

[Gorgon]

Couldn’t resist those bundles on Ryzen CPUs and motherboards. Picked up a 3600x and a Gigabyte Aorus Pro WiFi b450m. The 3600x went in one of the kids motherboards and the 2700 that fell out of the motherboard went in the b450 in a micro ATX case that was sitting empty. So another 16 threads for WCG!

 

Not too bad at 90W at the wall.

 

Upped all my systems to 1GB/thread while memory was still cheap and that’s allowed me to re-purpose a i5-6400 motherboard in another empty micro ATX case with some 4GB DDR4-2400 sticks that fell off the bottom of the RAM upgrade. Only another 4 threads but every bit helps so that puts me up to 102 threads.

 

I see @miker07 is back on WCG with his army of Threadrippers, 3950 and he seems to have won the lottery and added a 5950x as well. It will still take him a little while to roll past me though 

 

I'm regretting not picking up a another Aorus Master and 3950x while that sale was on though but I want to keeps some $ around for when my number comes up in the queue at EVGA for a 3080.

[Windows7ge]

@GorgonThe heat has gotten to me. Damn local weather. The least I can do is estimate that with the addition of my 32-core EPYC I'd be in the neighborhood of 120,000PPD.

 

Let's see if I can manage to keep my position until fall when things cool down again. 

[Gorgon]

11 hours ago, Windows7ge said:

@GorgonThe heat has gotten to me. Damn local weather. The least I can do is estimate that with the addition of my 32-core EPYC I'd be in the neighborhood of 120,000PPD.

 

Let's see if I can manage to keep my position until fall when things cool down again. 

Yeah, warm here too in Eastern Ontario today.  Almost switched from my jammies to shorts. Still managing the heat load for 10GPUs @ ~1.4kW and ~600W of CPU.

 

Were in lockdown here again but the government hasn’t reduced the electricity rates like they did in the past yet 

[Windows7ge]

35 minutes ago, Gorgon said:

Yeah, warm here too in Eastern Ontario today.  Almost switched from my jammies to shorts. Still managing the heat load for 10GPUs @ ~1.4kW and ~600W of CPU.

 

Were in lockdown here again but the government hasn’t reduced the electricity rates like they did in the past yet 

Electricity costs where I live aren't cheap but I mentally accept that as part of my operating costs for having all these servers. I think I'm looking at somewhere around 2kW worth of CPU. :3

 

Speaking of, EPYC 7601 up and operational.

 

I'm impressed by it's full load temps and the performance of this little NH-U9. Thing isn't exceeding 60°C with NF-A9 fans running at 1200RPM.

 

On a somewhat related note. I made a status update tonight and threw a little mention in there about the upcoming pentathlon. Do a little early recruiting before things get going. :3

[Windows7ge]

OK, my desktop is NOT happy for some reason...

 

 

First all it's existing jobs just froze then when I tried removing/re-adding the project to fix that it acted like WCG couldn't be reached.

 

System restarts did little to nothing. To get it going again I had to full on:

sudo apt remove boinc-client boinc-manager
shutdown now
sudo apt install boinc-client boinc-manager

Now I was able to add the project back and I'm downloading jobs.

 

I have no idea why this happens periodically and it only happens on my desktop.

 

On a light note this showed me something I never though I'd see from WCG:

1CPU + 1ATI GPU

That explains why for a couple of days my desktops productivity went up six fold. :3 I never thought I'd see the day WCG would offer anything GPU computational.

[WhitetailAni]

Hmm...

I may not be able to finish any tasks with my P3...

Spoiler

Yes, that is accurate. No, the "time to completion" hasn't stopped going up.

No, I won't wait 46 minutes 45 seconds so it will say 69 hours of CPU time.

[da na]

Yo how do i get the boinc badge? I was a member of the boinc team like 3 years ago, before I knew the forum exists. I was on the LTT team.

[Gorgon]

8 hours ago, wat3rmelon_man2 said:

Yo how do i get the boinc badge? I was a member of the boinc team like 3 years ago, before I knew the forum exists. I was on the LTT team.

Here:

 

[Windows7ge]

@GorgonI've been observing the behavior of this EPYC 7601 for the past few days and it's been doing something rather peculiar...I'm not certain if it's a performance issue or just HTOP not reporting correctly. Anyways you mentioned you wanted to compare one of your CPU's to this. In what metric did you want to test? Tasks per day? Points? Run time?

 

[Gorgon]

43 minutes ago, Windows7ge said:

@GorgonI've been observing the behavior of this EPYC 7601 for the past few days and it's been doing something rather peculiar...I'm not certain if it's a performance issue or just HTOP not reporting correctly. Anyways you mentioned you wanted to compare one of your CPU's to this. In what metric did you want to test? Tasks per day? Points? Run time?

 

I’m running 29 threads on my 3950x currently 24x7 and 21 threads on my 3900x. Both are running about 4.1GHz/thread using stock settings (PBO etc) on higher end Gigabyte x570 motherboards.

 

Tell me what data you want to see and how I can generate it.

[Gorgon]

On 4/13/2021 at 9:03 PM, Windows7ge said:

OK, my desktop is NOT happy for some reason...

 

 

First all it's existing jobs just froze then when I tried removing/re-adding the project to fix that it acted like WCG couldn't be reached.

 

System restarts did little to nothing. To get it going again I had to full on:

sudo apt remove boinc-client boinc-manager
shutdown now
sudo apt install boinc-client boinc-manager

Now I was able to add the project back and I'm downloading jobs.

 

I have no idea why this happens periodically and it only happens on my desktop.

 

On a light note this showed me something I never though I'd see from WCG:

1CPU + 1ATI GPU

That explains why for a couple of days my desktops productivity went up six fold. :3 I never thought I'd see the day WCG would offer anything GPU computational.

Damn, I didn’t realize the GPU experiment was out of beta.  There seems to be a bunch of frustrated ex SETI folks chomping at the bit for all the GPU WUs.

 

Maybe I’ll put the GTX 1060 back in one of my BOINC systems instead of selling it.

[Windows7ge]

21 minutes ago, Gorgon said:

I’m running 29 threads on my 3950x currently 24x7 and 21 threads on my 3900x. Both are running about 4.1GHz/thread using stock settings (PBO etc) on higher end Gigabyte x570 motherboards.

 

Tell me what data you want to see and how I can generate it.

...you said you wanted to compare, or something. I'm all good, don't have any questions.  Was there some statistic you wanted me to run/test on the CPU?

 

22 minutes ago, Gorgon said:

Damn, I didn’t realize the GPU experiment was out of beta.  There seems to be a bunch of frustrated ex SETI folks chomping at the bit for all the GPU WUs.

 

Maybe I’ll put the GTX 1060 back in one of my BOINC systems instead of selling it.

The magic #5 that I haven't seen on any WCG project since I joined. OPN is the only one that supports it right now.

[Gorgon]

11 hours ago, Windows7ge said:

...you said you wanted to compare, or something. I'm all good, don't have any questions.  Was there some statistic you wanted me to run/test on the CPU?

Mostly just curious about the differences in performance of a Ryzen to Thread Ripper and/or Epyc.

 

Heres the last few weeks of mine:

 

My estimate is about 8294 and 7221 points per thread per day for the 3900x and 3950x respectively.  Both at stock settings, the 3900x on a NH-D15 and the 3950x is wearing a EVGA 240 CLC. Both just a ‘titch under 4.1GHz

[Windows7ge]

9 hours ago, Gorgon said:

My estimate is about 8095 points per thread for the 3900x and 7241 points per thread for the 3950x.  Both at stock settings, the 3900x on a NH-D15 and the 3950x is wearing a EVGA 240 CLC. Both just a ‘titch under 4.1GHz

Hmn, because of the weird thing the CPU is doing I think I'm going to need to collect a couple weeks worth of data before I can give you anything conclusive.

[Windows7ge]

On 4/15/2021 at 11:25 PM, Gorgon said:

Damn, I didn’t realize the GPU experiment was out of beta.  There seems to be a bunch of frustrated ex SETI folks chomping at the bit for all the GPU WUs.

At over 300K PPD I take it you got GPU jobs working?

 

For reasons unknown I'm not getting any more.  I can't compete with GPU compute.

 

How could you WCG. You were my ace in the hole but with GPU jobs available I have nothing.

[Gorgon]

1 hour ago, Windows7ge said:

At over 300K PPD I take it you got GPU jobs working?

 

For reasons unknown I'm not getting any more.  I can't compete with GPU compute.

 

How could you WCG. You were my ace in the hole but with GPU jobs available I have nothing.

The GPU jobs appear to have dried up yesterday afternoon.  I only got one batch on one GPU around 22:00 EST but that’s it.

 

I had a 1060 6GB initially on it to dip my toe in the water but they’re only releasing 1700 jobs every 1/2 hour pseudo randomly and a bunch of the old SETI folks appear to be all over them so my GPU was only active 25-30% maximum. I added a 1070ti the evening before yesterday but the the work fizzled out.

 

They’ve been having issues with invalids so they’re still working the kinks out and I expect they’ll return but though you get nice points for them you don’t get double the time contributed for 0.9 CPU and 1.0 GPU.

 

The jobs aren’t very well optimized judging by the fact that though the GPU is at 100% utilization the 1060 is only drawing 60-80/120W and the 1070ti showed a similar lower power.

 

Still might be a nice boot with the A/C season coming.

 

I’ve moved the 1070ti back to F@H for now.

[Windows7ge]

7 hours ago, Gorgon said:

 

I had a 1060 6GB initially on it to dip my toe in the water but they’re only releasing 1700 jobs every 1/2 hour pseudo randomly

I was thinking to move one of my un-utilized R9 290X's over to one of my nodes that has a working Linux OpenCL driver but I think the GPU is borked so I don't know if it'd do any good or just kernel panic the system.

[Gorgon]

2 hours ago, Windows7ge said:

I was thinking to move one of my un-utilized R9 290X's over to one of my nodes that has a working Linux OpenCL driver but I think the GPU is borked so I don't know if it'd do any good or just kernel panic the system.

As long as the driver supports OpenCL 1.2 it should work assuming the hardware is good. Apparently issues popped up in Beta with older cards. GTX 700-series don't work under Windows but do under Linux and some older AMD Cards will work if you spoof the reporting so they report as supporting OpenCL 1.2 so OpenCL 1.1 appears to work fine for the container/toolchain they are using (AutoDock) but has been restricted by the project.

 

Still not much movement on the GPU tasks. I've only seen a few today so far. OpenPandemics is limiting the number of tasks currently as they are still assessing the impact on their "pipeline" presumably to ensure the rapid rate of returns from GPUs doesn't break their collection and validation servers.

 

So though their "Beta" period completed successfully and they went into production April 6th, with the limited amount of tasks available they are still, in essence, in Beta.