BOINC Community Board

[Gorgon]

and this is what the production looks like when you do maintenance:

picked up a couple of AMD R9 5900x CPUs on a killer sale at a local bricks and mortar store and 2 Noctua NH-U12S coolers from Amazon.

Did a massive re-arranging of CPUs and Coolers to match the motherboards all along the “food chain” with a Asus h170m motherboard with a i5-6100 and a Ryzen 3 1200 and Wraith Spire falling out the other end into the parts collection along with my venerable Athlon 64 xII that’s been in service for over 10 years and an i3 NUC that are both just too inefficient to run BOINC anymore.

 

Took the opportunity from the down time to update the BIOS on the motherboards, clean Dust Filters and swapped a couple of GPUs between systems to better balance power loading on UPSes and branch circuits.

 

A net gain of 48 threads with a loss of 6 yielding 42 additional threads to the cause. The current lineup is a 5950x, 2 x 5900x, a 3950x, 3900x, 2700x,  2 x 2700, a  i9-990K and the Xeon e3-1231v3 in my Daily Driver. So 208 threads total with 11 reserved for overhead and 8 for driving GPUs that are folding so 189 threads total now on BOINC.

 

Most are on WCG Mapping Cancer Markers with 3 systems running Open Pandemics on CPU and each of these has an older Pascal GPU running WCG Open Pandemics GPU WUs when they come along and Einstein WUs when they’d be otherwise idle.

[rkv_2401]

12 hours ago, Gorgon said:

and this is what the production looks like when you do maintenance.

 

Most are on WCG Mapping Cancer Markers with 3 systems running Open Pandemics on CPU and each of these has an older Pascal GPU running WCG Open Pandemics GPU WUs when they come along and Einstein WUs when they’d be otherwise idle.

Impressive.

 

Is WCG your favorite BOINC project?

[danwat1234]

Gorgon, impressive! What software are you using to query GPU power consumption across multiple machines? What a pain it will be to dust out the machines in my garage when the time comes.

[Gorgon]

3 hours ago, rkv_2401 said:

Impressive.

 

Is WCG your favorite BOINC project?

Thanks, I prefer to support research into Cancer, COVID and Neurological diseases so WCG fits the bill. Rosetta is another option. I also run Einstein as the WCG Open Panedemics GPU work units are few and far between and looking for Binary Neutron Stars is a lot more cool, IMHO, than finding Prime Numbers or solving esoteric Mathematical conjectures. That's the beauty of BOINC as it's an independent platform for many different types of research.

1 hour ago, danwat1234 said:

Gorgon, impressive! What software are you using to query GPU power consumption across multiple machines? What a pain it will be to dust out the machines in my garage when the time comes.

Dan, If your going to respond please either respond using the link in the post or use @user_id so the original poster gets a notification.

 

I have a Intel i7 NUC running a free ESXi 6.7 instance and have a Virtual Machine on that running Ubuntu 18.04.3 LTS that is my Zabbix Server which queries Zabbix Agents installed on my Distributed Computing Nodes. This is a bit of a Black Art but I've been doing System Management a work for over 25 years and so is second nature to me.

 

I can tell at a glance if any of the Nodes are acting up. I have a number of scripts that the Zabbix Agents run to collect relevant data. The GPU power figures are queried using the Nvidia System Management utility nvidia-smi and the output parsed using standard Linux command line utilities (sed, awk, grep etc.).

 

I prefer Linux for the Nodes as it's free, you can update when you want, and you can use SSH to fully manage the systems remotely and aren't reliant on a GUI (VNC, RDP etc.). Learning the Command Line Interface (CLI) can be a steep curve but the flexibility it provides is well worth it.

 

I haven't figured out a consistent method yet to programmatically query CPU power consumption from /proc/sys so I'm reliant on estimating it from UPS data on total system power. Querying the UPS data is problematic as though my older UPSes have network interfaces that I can query the System Power with via SNMP the newer UPSes only have USB and though it works somewhat reliably for APC UPSes I've replaced most of those with Cyber Power units due to a high failure rate on the APC Back UPS series and the Cyber Power USB implementation is buggy AF and the connections are thus problematic.

 

Ideally, as I have at work, a networked metered Power Distribution Unit (PDU) would be the solution but the cost of these start at $250 and quickly go up and I just can't justify that cost for home use. I'd rather just take those $ and use them to fund more hardware that can do actual work

 

My systems are all in the unfinished portion of my basement and so accumulate concrete dust which can be conductive so every three months I vacuum the dust filters and once a year I do a complete system inspection and more thorough cleaning. I've invested in cases with mesh dust filters and try to use Noctua Industrial iPPC fans as they have better seals on the bearings to prevent ingress of dust.

 

Likely your approach using second hand eBay systems and just replacing them as they die is more cost effective but my wife told me I needed a hobby and I like the challenge of finding the best mix of hardware at the best price-point to fit my needs and keep my OCD humming along.

[rkv_2401]

18 hours ago, Gorgon said:

Thanks, I prefer to support research into Cancer, COVID and Neurological diseases so WCG fits the bill. Rosetta is another option. I also run Einstein as the WCG Open Panedemics GPU work units are few and far between and looking for Binary Neutron Stars is a lot more cool, IMHO, than finding Prime Numbers or solving esoteric Mathematical conjectures. That's the beauty of BOINC as it's an independent platform for many different types of research.

Agreed, I think Rosetta is the most ambitious of the projects (maybe even more so than Folding@home, which is the biggest to my knowledge) and would love to contribute to them, but looks like they've been having problems with the Work Units since September or even earlier. I don't like WCG's new UI/UX changes, I prefer the old BOINC method of just showing what computer produced how many credits and said computer's hardware specs - but they seem to be the most active healthcare research project at the moment, so I've decided to put 6 to 8 cores on that and I may put the other one on SiDock@home. SiDock doesn't seem to be ambitious at all compared to Rosetta's goal of designing new proteins and Folding@home's goal of finding cures for as-of-now incurable diseases, but they're enthusiastic, so I'll give them the benefit of my doubt. 1070's on Folding@home for the moment. I think this stint from now till around April/start of May will be my last serious one until I find a way to power these PCs with renewable energy. Really wish it could have been with Rosetta.

 

Ibercivis's BOINC division has just vanished, showing no trace of my credits accumulated with them or me ever having run their project. I understand there are difficulties around keeping the website of a defunct project running, but that's still a bit rude, IMO.

[Gorgon]

5 hours ago, rkv_2401 said:

... I don't like WCG's new UI/UX changes, I prefer the old BOINC method of just showing what computer produced how many credits and said computer's hardware specs ...

I didn't like it at first and I agree that the old Interface from IBM was better suited but I suspect the new UI is designed to be mobile compliant so some compromises were likely needed. The information is all there in the new interface just in different places.

 

I usually use BOINCstats anyway as I can see at-a-glance my progress across all the projects I've participated in.

[Gorgon]

Noticed a couple of issues after the hardware swap.

 

The 3900x was running really hot, approaching Tdie of 90C in the Gigabyte ax370 gaming 7 with a Vcore of 1.34V it was pushing 3.9GHz all-core and the VRM was reading 80C which is within spec but a little hotter than I like.

 

First I Decreased the Processor Package Target (PPT) from the default to 95W in the BIOS and applied a Dynamic Vcore Adjustment (Dvid) of -0.100V. This yielded Vcore of 1.1V, 3.75GHz all-core Tdie 71C and the VRM settling at 69C after about 1/2 hour of run time.

 

Next I tried a simpler approach just enabling PBO then enabling ECO mode which sets the PPT to 87W. This resulted in a Tdie of 63C, the VRM at 61C with an all-core of 3.6GHz at a Vcore of 1.07V.

 

I decided to leave it in ECO mode even though setting a manual PPT with a Dvid yielded better results that could change over time resulting in instability and so the ECO mode is likely the better option.

 

The 2700x I moved from the Gigabyte ax370 Gaming 7 was installed in a Gigabyte Aorus b450m. While the Tdie was at acceptable levels the VRM creeped up to 115C over 24 hours. I've had issues with the VRMs failing on lower-end Gigabyte b450 boards due mostly to their 4-phase design with doubled Low-side MOSFETs combined with inadequate heat sinks. In this instance the solution was to change the PPT to 95W which dropped the all-core from 3.9 to 3.7GHz and Vcore from 1.3 to 1.1V but the VRM appears to settle at about 77C with Tdie at 60C. So though the CPU could run comfortably at stock the VRM on the motherboard isn't up to running a 105W TDP CPU 24x7x365.

 

Lastly I noticed the 2700x removed from the Gigabyte ax370 Gaming K7 which installed in a Spare Gigabyte Aorus b450m, replacing the h170 and i5 that was in that system, was taking 4 1/2 hours to complete a MCM WU whereas similar WUs on an identical system were taking just over 2 hours. The two systems were running at a similar all-core clock but the less performant one was running at lower temperatures so I doubted I had a bad mount on the stock cooler.

 

After some thought I began to suspect that the OS was confused by the architecture change from Intel to AMD. So I ran a "apt upgrade" followed by an "apt update" to get update packages and force a kernel re-compile which would also optimize the kernel for the new architecture. This appears to have solved the issue and I'm now seeing the WUs that were in progress complete in increasingly shorter times approaching 2 hours.

 

The moral of the story. After making changes monitor the systems for a couple of days checking thermals and performance to catch any issues at the start.

[rkv_2401]

Final set up complete -

6 cores of an R5 3600 @ 3.6Ghz running SiDock@home and configured to get Rosetta WUs when they arrive (personally given up on that, though) - 50W reported package power draw

6 cores of an i5 10400 @ 2.9Ghz running WCG Africa Rainfall and MCM, and also configured to get Rosetta WUs (hasn't gotten a task in 2 weeks) - 30W reported power draw

2 cores of an i7 10750H @ 2.6Ghz - running WCG MCM for now, will take it offline in a month or so, once the heat gets unbearable - 15W reported power draw

 

 

[Gorgon]

PSA The Transfer of the Server Infrastructure from IBM to Krembil is underway. Yesterday was the last day Work Units (WUs) for World Community Grid (WCG) were served and any WUs still underway have until February 27th to be completed.

 

The new site will start stress testing February 28th. It is unclear when the new site will start distributing WUs, likely sometime after February 28th and before April 22nd when the new project site and forums are scheduled to be on-line.

 

I've moved my rigs over to Einstein for CPU and GPU work and will keep them there until WCG is back up and running.

[MrFinn741]

That explains why I have no tasks running. I did read there was a move underway, but not that there would be a pause in WUs 

[Gorgon]

1 hour ago, MrFinn741 said:

That explains why I have no tasks running. I did read there was a move underway, but not that there would be a pause in WUs 

Yeah, I have one system still with some MCM and OpenPandemic tasks and I'm trying to finish the WCG tasks so I can tune the Einstein tasks but I've noticed a few additional tasks sneaked in a couple of hours ago so I've set the Project to "No New Tasks" so it will finish the ones currently in the queue and I can move forward.

[danwat1234]

Last week upon hearing of the big switchover, I set WorldCC to 90% resource share, Rosetta and SiDock each to 5% and added SiDock because Rosetta often isn't very active. Einstein at 1% but I have it suspended currently. So, for the newbs that means that if World Community Grid is offline, Rosetta and/or SiDock can still use all cores on all my machines 24/7, but if World comes back online it gets priority once again. Although BOINC sometimes doesn't behave.. but really i'm happy to contribute to any of those 3. 

[GOTSpectrum]

 

[Gorgon]

@Windows7ge Well I would have taken 1st place on Team LTT for Einstein today but we had a new member "magne56" join the team this weekend with 2,125,099,591 points so I have a little bit more work to do

 

So stop whining about me and my GPUs in WCG!

 

Something about karma being a bitch ...

[Windows7ge]

22 hours ago, Gorgon said:

@Windows7ge Well I would have taken 1st place on Team LTT for Einstein today but we had a new member "magne56" join the team this weekend with 2,125,099,591 points so I have a little bit more work to do

 

So stop whining about me and my GPUs in WCG!

 

Something about karma being a bitch ...

It won't be happening particularly soon but I have roadmap plans to build a GPU compute server. At the moment the most appealing candidate is the NVIDIA Tesla M40 12GB. If I can pack 5 or 6 of them into a box it'd be a fantastic F@H rig. Looking to get some forum badges for folding.

[Gorgon]

1 hour ago, Windows7ge said:

It won't be happening particularly soon but I have roadmap plans to build a GPU compute server. At the moment the most appealing candidate is the NVIDIA Tesla M40 12GB. If I can pack 5 or 6 of them into a box it'd be a fantastic F@H rig. Looking to get some forum badges for folding.

Boy if you thought your electricity bill was big now those toasters are 250W TDP or there about and yield about as much (1.2-1.7MPPD) for F@H as a 1660 or 3050 at half the power.

 

Might be good for some BOINC Compute depending on their DP performance though. (Nope - that sucks too)

 

Still, if you want to play with some DC stuff in a 4U chassis that screams you've got the right project.

[BondiBlue]

1 hour ago, Windows7ge said:

It won't be happening particularly soon but I have roadmap plans to build a GPU compute server. At the moment the most appealing candidate is the NVIDIA Tesla M40 12GB. If I can pack 5 or 6 of them into a box it'd be a fantastic F@H rig. Looking to get some forum badges for folding.

Several of those M40s would be one hell of a power draw considering their PPD. My 1660 Ti (super low end OEM model with a terrible cooler) gets around 1.4-1.5 million PPD while only pulling 110W, and that's at completely stock settings too. Are you sure those Teslas would be a good idea? 

[Windows7ge]

1 hour ago, Gorgon said:

Boy if you thought your electricity bill was big now those toasters are 250W TDP or there about and yield about as much (1.2-1.7MPPD) for F@H as a 1660 or 3050 at half the power.

 

Might be good for some BOINC Compute depending on their DP performance though. (Nope - that sucks too)

 

Still, if you want to play with some DC stuff in a 4U chassis that screams you've got the right project.

 

25 minutes ago, BondiBlue said:

Several of those M40s would be one hell of a power draw considering their PPD. My 1660 Ti (super low end OEM model with a terrible cooler) gets around 1.4-1.5 million PPD while only pulling 110W, and that's at completely stock settings too. Are you sure those Teslas would be a good idea? 

 

I'm more into the DC stuff that screams TBH. F@H is just making it useful for something. I'm not married to the M40 though. If you know of a better bang for buck card that's not for desktop I'm open to suggestions.

[BondiBlue]

1 hour ago, Windows7ge said:

I'm more into the DC stuff that screams TBH. F@H is just making it useful for something. I'm not married to the M40 though. If you know of a better bang for buck card that's not for desktop I'm open to suggestions.

Ah, I see. I still fold on a GTX 680 even though it uses more power than my 1660 Ti or RX 570 while only getting a fraction of the points. It keeps the house warm though, so I haven't had to run the heat at all lol. 

 

I can't think of anything DC wise that would be interesting right off, but if I do I'll try to remember to mention it. 

[Windows7ge]

7 hours ago, BondiBlue said:

Ah, I see. I still fold on a GTX 680 even though it uses more power than my 1660 Ti or RX 570 while only getting a fraction of the points. It keeps the house warm though, so I haven't had to run the heat at all lol. 

 

I can't think of anything DC wise that would be interesting right off, but if I do I'll try to remember to mention it. 

Some day off in the future I hope to setup a solar system on my property. We have a nice chunky side yard which sees sunlight almost the entire day. I can offset my electric bill with that.

[Ithanul]

Oi, so I'm alive, just been busy.

 

In mean time, this weekend shall be enjoyable one.  Big main boi is finally getting the full refresh I been working on.

Finally piece is gained.

Some proper crunching in F@H/BOINC shall occur to get this chap tune.

 

[rkv_2401]

19 hours ago, Windows7ge said:

Some day off in the future I hope to setup a solar system on my property. We have a nice chunky side yard which sees sunlight almost the entire day. I can offset my electric bill with that.

Renewable energy + distributed computing is the ultimate combination

[Gorgon]

World Community Grid has missed their target re-launch date today and is now estimating they will be back in business on May 9th.

[GOTSpectrum]

Who is registering the team for the pent this year? 

 

@Ithanul @Gorgon , one of you two? 

[Gorgon]

Muwahahaha - I have all the Powers

I have achieved monitoring nirvana.