Hello,

I'm a BOINC user of some time, having contributed to World Community Grid and CERN a fair bit. I've been eyeing the climate emergency with increasing alarm for years now, and wanted to help out wherever I can. To that end, I've added this project to my Debian box today (June, 7, 2021), but I have yet to receive any work on my machine. I've installed the 32-bit libraries as mentioned in this sticky thread: https://www.cpdn.org/forum_thread.php?id=8916

I'm running Debian Testing on a machine that has a Ryzen 7 3700x and 32GB of RAM.

Any help would be appreciated.

What does it say in the event log after you request work?
If you have requested work once, don't hit update till an hour has past as there is an automatic time backoff on the project.

Also once you start getting work, almost certainly going above five or six cores will lead to tasks slowing down quite a bit. Certainly not worth using more than the eight real cores as opposed to vitual ones. I have the same cpu and amount of RAM

Ah, waiting for the hour to pass actually worked in this instance. Shouldn't have been so impatient, lol.

Thanks.

I'm a BOINC user of some time, having contributed to World Community Grid and CERN a fair bit. I've been eyeing the climate emergency with increasing alarm for years now, and wanted to help out wherever I can.

Congratulations: my sentiments exactly.

I'm running Debian Testing on a machine that has a Ryzen 7 3700x and 32GB of RAM.

In anticipation of ever-larger ClimatePrediction models in the future, I doubled my RAM from 32 GBytes to 64 GBytes. RAM these days is so cheap I could not resist. My motherboard will take, IIRC, 256 GBytes of RAM, but it is not that cheap.

My processor claims 16 cores but 8 are real and 8 are hyperthreaded, so I told the BoincClient to use a maximum of 8 cores. I also told it to run only 4 ClimatePrediction models at a time. Processor cache is 16896 KB The rest are mainly WCG and Rosetta. My processor is 1511241. Running like this, the N216 models take about 8 days each.

Initially, I tried running all 16 cores, but to do this, I had to run the fans so fast I cold not stand the noise.

$ sensors
coretemp-isa-0000
Adapter: ISA adapter
Package id 0:  +75.0°C  (high = +88.0°C, crit = +98.0°C)
Core 1:        +68.0°C  (high = +88.0°C, crit = +98.0°C)
Core 2:        +68.0°C  (high = +88.0°C, crit = +98.0°C)
Core 3:        +65.0°C  (high = +88.0°C, crit = +98.0°C)
Core 5:        +63.0°C  (high = +88.0°C, crit = +98.0°C)
Core 8:        +74.0°C  (high = +88.0°C, crit = +98.0°C)
Core 9:        +75.0°C  (high = +88.0°C, crit = +98.0°C)
Core 11:       +72.0°C  (high = +88.0°C, crit = +98.0°C)
Core 12:       +66.0°C  (high = +88.0°C, crit = +98.0°C)

amdgpu-pci-6500
Adapter: PCI adapter
vddgfx:       +0.69 V  
fan1:        2068 RPM  (min = 1800 RPM, max = 6000 RPM)
edge:         +38.0°C  (crit = +97.0°C, hyst = -273.1°C)
power1:        3.21 W  (cap =  25.00 W)

dell_smm-virtual-0
Adapter: Virtual device
fan1:        4084 RPM
fan2:         898 RPM
fan3:        2759 RPM

---

Initially, I tried running all 16 cores, but to do this, I had to run the fans so fast I cold not stand the noise.

I went for liquid cooling on mine :)

Initially, I tried running all 16 cores, but to do this, I had to run the fans so fast I cold not stand the noise.



I went for liquid cooling on mine :)

I do not have enough L3 cache to justify running that many Boinc processes, especially the N216 models.

---

I do not have enough L3 cache to justify running that many Boinc processes, especially the N216 models.

With the same CPU neither do I but thought it a useful bit of future proofing as well as keeping the noise down when i occasionally render video.

waiting for my task too, there is no work for 3 days, but seems cos I run several others on the same computer

The work here is mostly for Linux.
The recent batch of Windows work may be all there is for months.

There are posts by people who are using WSL to run the Linux version of BOINC with success. Another option is virtual box. (I use it the other way round with Linux as the host to enable Linux and Windows clients to run at the same time.

Hi Les,

I have been running BOINC+Climateprediction.net years ago and but kind of forgotten about it, after using mainly laptops and re-installing Windows and so on. But today I installed it on a desktop PC again running Windows 10, and it seems it is not fetching any tasks. Maybe there are no unsent Windows tasks? The problem is that on the server status website, it is not obvious at all which of the unsent work is for Windows/Mac/Linux. Instead of volunteers spending time trying to figure why no work is fetched, it would be much better if the "Tasks by Application" table would have another column with W/L/M to indicate which tasks are available for which OS.
If there is no Windows work, I might decide to run another BOINC project, even though I am mainly interested in Climateprediction.net.

Could you suggest that website improvement to the Admins? That would be great.

It's just a few clicks away on the Project status page:
Go to the line of options at the top of the page, click on Computing, then click on Applications in the drop down menu.

Great, thank you!

I do not have enough L3 cache to justify running that many Boinc processes, especially the N216 models.

How much L3 cache is required for each kind of project WU?

Just been looking and failing to find the stuff about how much level3 cache the N216 tasks take up. It is only those that really hammer it. The others, I can run certainly 8 tasks on my Ryzen7 without any slowdown. Going above 5 N216's things start to slow down but I can run a mix of tasks using all 8 real cores and 8 virtual ones without getting errors. It would take some weeks to work out where the sweet spot is in terms of maximum throughput.

Going above 5 N216's things start to slow down but I can run a mix of tasks using all 8 real cores and 8 virtual ones without getting errors.

Let me up mine to five for a while. But errors? From what? Slowdown understandable, but errors???

---

I do not have enough L3 cache to justify running that many Boinc processes, especially the N216 models.

How much L3 cache is required for each kind of project WU?

I started a thread somewhere, soon after the n216 models first appeared, near the start of last year.
One of the researchers had just posted on our private site, about how the models used a lot of L3 cache, so I passed that on.
Over the course of a dozen or so posts, others wrote about how they were limiting the number of models they ran, and about how something similar was happening on some other project.

The consensus was that 4 megs of L3 cache was good, so people started checking their processor specs to find the amount of cache each had, and limiting the number of models so as not to exceed that.

I do not have enough L3 cache to justify running that many Boinc processes, especially the N216 models.

How much L3 cache is required for each kind of project WU?

The project code was not written well and is incompatible with hyperthreading. I disabled HTing in my BIOS for any computer crunching CP. Since I've never found a single clue that anyone responsible for writing the project code or running the servers participate in this forum I doubt they'll ever do anything about it.

It's just a few clicks away on the Project status page:
Go to the line of options at the top of the page, click on Computing, then click on Applications in the drop down menu.

The https://www.cpdn.org/apps.php page does not provide the information they asked for. All it has is some historical averages of an undisclosed duration.

The project code was not written well and is incompatible with hyperthreading. I disabled HTing in my BIOS for any computer crunching CP. Since I've never found a single clue that anyone responsible for writing the project code or running the servers participate in this forum I doubt they'll ever do anything about it.

The project code is around one million lines of Fortran written by/for the Met office many years ago to run on their supercomputers. The CPDN programs work by sending various parameters to that compiled program. At some point there will be OpenIFS but don't hold your breath on that one. The met office code is proprietary and not open source and the CPDN licence does not allow them to modify it.

If CPDN is ever to have its own custom built code, it will require a lot more people to hit the donate button on the home page.