Posts by ProDigit

1) Message boards : Number crunching : Problems and Technical Issues with Rosetta@home (Message 103448)
Posted 21 Nov 2021 by ProDigit
Post:
I reattached Rosetta, after I no longer got any WUs.
Not only on my Android device, but also my intel x86 units.
2) Message boards : Number crunching : Rosetta on Android 10 not working? (Message 97713)
Posted 27 Jun 2020 by ProDigit
Post:
My old phone has Android 10, and Rosetta on it for quite a while now.

I just got a new phone, a pixel 3a XL (also with Android 10), and tried adding Rosetta, and the wizard says:
"Rosetta@home
Could not connect"

They discontinued support, or do they need to update their website/certificate on the Android 10 app?

Running boinc 7.4.53
3) Message boards : Number crunching : 12 CPU WUs (Message 97712)
Posted 27 Jun 2020 by ProDigit
Post:
Yes it is possible.
Some projects in the past, successfully created 8 CPU WUs that will really run on those amount of CPUs.

I could screen shot you the example, where I'm downloading 12 core WUs, and when running them, they run on only 1 thread each.

Possible or not, it clearly is a bug, I've seen in other projects in the past as well.

And no, Rosetta doesn't have any modified app_config setting. If it had, certainly I didn't create it!
4) Message boards : Number crunching : Problems and Technical Issues with Rosetta@home (Message 97698)
Posted 27 Jun 2020 by ProDigit
Post:
12 CPU WUs hogging up my PC, using only 1 cpu core.
I will for the time being disconnect from this project until the issue is resolved.
5) Message boards : Number crunching : 12 CPU WUs (Message 97697)
Posted 27 Jun 2020 by ProDigit
Post:
You guys seem to really have effed up again,
2x 12 thread WUs hogging my entire PC, running on a single CPU core.
Can you like NOT send BETA WUs over the regular releases please????
Remove the multi CPU WUs! They don't work!

For as long as the issue isn't fixed, I will ban Rosetta from sending any new WUs to me.
Please follow up when issue is resolved!
6) Questions and Answers : Unix/Linux : Rosetta hogging resources, CPU cores inactive! (Message 96698)
Posted 21 May 2020 by ProDigit
Post:
For now I just did some redundant lines, all meaning basically the same thing, in the hopes that at least 1 triggers the command:

<app_config>
<app>
<name>Rosetta</name>
<max_concurrent>12</max_concurrent>
<gpu_versions></gpu_versions>
</app>
<app_version>
<app_name>Rosetta</app_name>
<avg_ncpus>12</avg_ncpus>
</app_version>
<project_max_concurrent>12</project_max_concurrent>
</app_config>
7) Questions and Answers : Unix/Linux : Rosetta hogging resources, CPU cores inactive! (Message 96697)
Posted 21 May 2020 by ProDigit
Post:
I have a Ryzen 3900x system with 24 threads, and only 16GB of RAM.
I noticed when there are more than 15 projects of Rosetta running, that some (like 5) CPU cores are passive (not crunching) because the projects are waiting for memory.
This also causes my CPU to run at lower frequencies, hogging down the entire crunching process, huring other project's progress as well!
The Ryzen 9 3900x in my machine is set to run most optimal with all cores loaded at 100% utilization.

I want to limit Rosetta to 12 cores max on this PC.
I know the access is in /var/lib/boinc/projects/ (the rosetta folder), app_config.xml

I have it currently as:
                                                             
<app_config>
<app>
<name>Rosetta</name>
<max_concurrent>12</max_concurrent>
</app>
</app_config>


Boinc MGR doesn't complain about it being a wrong name, but it also doesn't adhere to the limitation.
The project WU name is 'Rosetta 4.20', but I know app_config doesn't allow a space in the name.
I haven't been able to make it work so far...
What else do I need to do?
8) Message boards : Number crunching : Newbie Question : Dif Points on the same Crunching Time (Message 95448)
Posted 27 Apr 2020 by ProDigit
Post:
Same results here, even worse on ARM.
ARM 1,9Ghz quad core PPD is like NOTHING compared to x86 at 3,5Ghz.
I mean, it's not half, not even a quarter, but PPD for ARM is like 20x lower PPD:

https://boinc.bakerlab.org/rosetta/results.php?userid=2031893
9) Message boards : Number crunching : Rosetta 4.15 (short time estimates) (Message 94466)
Posted 14 Apr 2020 by ProDigit
Post:
When you preload tasks increase the setting of 0.1 day, and 0.1 additional day in boinc mgr to 3 or 4.
you'll get Wu's with longer deadline, which also means a higher chance of getting Wu's that run longer.
Rosetta has mixed WUs, some of which run close to 10 hours.

The deadlines assigned have nothing to do with how much work your machine requests. And runtimes are based on the runtime preference, not some fixed WU size.

I know, but if I set 10 day + 10 day of additional work, I usually get projects with long deadlines, which many of them happen to be long units.

We used to have tasks with 8-day deadlines quite frequently, but since the recent changes I haven't see a single 8-day task. And certainly none longer for about 5 years.
It looks like we're now an exclusively 3-day-deadline project


I've noticed the Covid 19 WUs on mobile get a 2 day deadline!
Those WUs run several hours on Android. Tough to get more than 6 loaded at the same time.
10) Questions and Answers : Unix/Linux : Working Instructions for Installing BOINC Client on new Installation of Ubuntu 18.04.4 LTS (Message 94465)
Posted 14 Apr 2020 by ProDigit
Post:
All of my Ubuntu machines are running 18.04.4, and I installed boinc directly from the built-in repos -- no PPAs needed.

Granted, most of them are headless machines that aren't running boinc-manager (if I need to see status locally, I SSH in and run boinctui), but I do have a laptop with a full GUI install that is running boinc-manager, and I don't recall having to do anything special to get it working.


I'm running Lubuntu 18.04 on several machines, and Boinc isn't included there.
Lubuntu should use the same repositories as Ubuntu. Perhaps Lubuntu just haven't added them and Ubuntu has?
11) Message boards : Number crunching : Running Rosetta on Raspberry Pi 3B+ (how to guide) (Message 94464)
Posted 14 Apr 2020 by ProDigit
Post:
Not sure how compressing RAM/Swap would work out for you, however it's pretty easy to set up the swap partition on an external drive, so you don't have to worry about SD card wear as much.

However, not all Boinc projects use a lot of RAM.
You could run 2 threads running Rosetta, and 2 running other project(s).
12) Message boards : Number crunching : Threadripper and Ryzen (and EYPC) (Message 94463)
Posted 14 Apr 2020 by ProDigit
Post:
From the specs, a TR2900x has only 8 cores (16 threads), running at 3,1Ghz.
Not a lot of info online about it.
The Threadripper 2950x pretty much seems to be a Ryzen 9 3900x on negative steroids.. I mean, it's identical but at 12nm vs 7nm for the 3900x.
I presume the only difference is the power consumption.
The Ryzen 9 (rated at 105W TDP) actually runs closer to 125-145W, and under full load still doesn't hit anywhere near 4,2Ghz boost frequencies (more like 3,6-3,8Ghz). I think the same could be said about the TR 2950x at 180W (it probably runs closer to 225W).

Things to consider when you compare these 2 (and the 2950x is newer than the 2900x):
The 2950x does run up to 2960 or so Mhz Memory speed. The 3900x gets up to 3200Mhz, but people managed to make it work with overclocked to 3600Mhz memory as well.

The annual running cost of the TR2950x would be ~$60+ more than a 3900x. You'd have to factor this into your calculations.
13) Message boards : Number crunching : Task start optimisation suggestion (Message 94411)
Posted 13 Apr 2020 by ProDigit
Post:
A lot of the stuff remains in RAM, and doesn't fully get written until a checkpoint has reached.
I would recommend buying a cheap $18 SSD (64GB) that would still give you enough writes to do Boinc for the next 5 or 10 years 24/7 before wear leveling becomes an issue.
14) Questions and Answers : Unix/Linux : Working Instructions for Installing BOINC Client on new Installation of Ubuntu 18.04.4 LTS (Message 94408)
Posted 13 Apr 2020 by ProDigit
Post:
sudo apt install boinc


^ That didn't get you up and running?

No, boinc isn't included in the PPAs.
You'd have to find Boinc PPA install instructions, like these ones:

sudo add-apt-repository ppa:costamagnagianfranco/boinc
sudo apt-get update


before you can install boinc on Ubuntu variants with the above command.
15) Message boards : Number crunching : Running Rosetta on Raspberry Pi 3B+ (how to guide) (Message 94406)
Posted 13 Apr 2020 by ProDigit
Post:
If you go headless you can run with just 8MB of VRAM (I set my resolution to 480/512p headless).

If you only run Boinc on the desktop OS, you'll be able to run it with 24MB of VRAM, if you choose 720p resolution (and 16MB on 480p; but the Raspbian desktop OS isn't very sub-768pix friendly).

I would seriously advise against using a swap file.
Your swap file should read zero usage.
Any Boinc project uses a lot of disk reads or writes, so if you do want to run it, I would recommend to boot from a bootable SSD if possible (via USB3,0).
It not only speeds up the process, but it also will be much less susceptible to data corruption than SD cards are.
And, especially if you're not running an A1/A2 card from Sandisk, you probably are going to experience system lags on the desktop.

From HTOP it appears you're only running 3 out of 4 cores for Rosetta.
Each one seems to use about 300-400MB of RAM.
In most scenarios, I would recommend to run only 2 instances of Rosetta on a 1GB system, and 4 on 2GB of available RAM.
16) Message boards : Number crunching : COVID 19 FAH and Rosetta the same? (Message 94400)
Posted 13 Apr 2020 by ProDigit
Post:
No, different. Both are important. I guess journal entries are a good way to compare the two.

FaH is concerned with the process of how proteins fold. R@h is concerned with where they end up once they have folded.

The folded state is important because that shape determines the proteins function and what it will bind with.

R@H also works on designing new proteins that don't exist in nature, which may well lead to one of the big technological revolutions, like the transistor did.

Or another Corona or Ebola..
I'm always iffy about messing with nature.
If one dictator gets a hold of some scientist's family, and forces him to work on a new lethal bio gas...
I don't know...
Then again, if it's going to be done anyway, might as well be the good guys, that won't spread it out...
17) Message boards : Number crunching : The most efficient cruncher rig possible (Message 94389)
Posted 13 Apr 2020 by ProDigit
Post:
Gear:
DDR4 uses about the same power as DDR3, with XMP enabled. This speeds up the RAM from 2100 or 2133Mhz to whatever it's rated for (usually 2400/2699/2900/3200 or even 3600Mhz).
Meanwhile max DDR3 speeds are only 1600Mhz.
At the stock 2100/2133Mhz, a 3200Mhz module runs at 1,2V instead of 1,5V on DDR3, and uses less power.
However RAM consumption is very minimal. Just a little more than an SSD; A few watts at most.
For CPU usage, DDR speed matters a lot more than for GPU crunching, so I'd say yes.

PSU:
Gold PSUs usually have a 90% efficiency rating at 80% load, and a 80% efficiency rating at below 40% and above 85 (sometimes 90)% load, quite often using lower fan profiles (alllowing the PSU to run hotter) to hit that target. Hot PSUs is not recommended.
You'll be paying yourself blue on a gold rated PSU. Instead, try to aim for between 50-75% load on a PSU.
If your system is using 500W, aim to get anywhere from a 800-1000W PSU. They still run awesome efficiencies, and the electric cost difference between a gold PSU and a bronze 80+ , you can recuperate in about 10 years (far after the PSU has given up the ghost running 24/7).
In my opinion gold or platinum PSUs aren't worth it. Get a regular, run off the mill Bronze PSU.
As far as brands go, I've had 4 EVGA PSUs blow out on me, meanwhile similar priced Corsair, as well as (surprisingly) the Amazon brand Rosewill has lasted me a long time without fail!

CPU Cooler:
I would really urge you to get a closed watercooling system on any 65W or greater CPU, especially the Ryzen CPUs, as well as modern Intel CPUs that turbo up the frequency based on the CPU temperature!
They're not that much more expensive than an air cooler.
For reference, a cheap air cooler costs $8, a decent one costs $14-20. A good one (for 65+W TDP) can easily hit in the 30's to 40's of USD or more!
A good single fan closed loop water cooling circuit, like this one, with a 1 year warranty, but guaranteed to work for at least 2 years non stop, will cost you just $45. It is good enough for 65W CPUs, and in case you'd run a Ryzen 9 CPU, it will allow the Ryzen to run at higher boost frequencies. Newer CPUs are much more susceptible to temperatures.
I don't believe the 1% difference review, unless he compared it to a $45+ air cooler. Even then, CPUs would hit 60C easily with those coolers, while they only hit in the high 40s to middle 50C on water cooling.
And there are so many bios settings on those CPUs, that you could run a single core at only 2,5Ghz (when all core boost is selected). In such case, no cooler would probably give the same results as a water cooling system!
My stock air cooler was so crappy, I had to throw it out in the garbage!
It would actually thermal throttle the CPU that under load it was hitting only 700Mhz!
That's a good 80+% reduction in performance. Granted, the block didn't look shiny and even, and looked like they did a shoddy job. But that was the one delivered with the 3900x.

A dual 120mm liquid cooling systems are great for Ryzen 9 3000 series CPUs (running 105w or less).
What's even more amazing is that a ryzen 9, has an eco setting in the bios.
A 105W ryzen can run at 65W (1/3rd less power consumption), and loses only about 15% of performance at that wattage! (which is right now the epitomy of efficiency).
In this ECO setting you can even water-cool the CPU with the single fan cooler I mentionend above!

In my opinion, if you're going air cooling or small (sub $40 cheap) liquid cooling, a 3950x doesn't make sense over a 3900x.
Because even with water cooling, both of their turbo frequencies are still thermally throttled.
The 3900x has fewer cores, but has higher boost speeds than the 3950x (I have them both).
You do the math and the few 100Mhz I get extra on a 3900x makes that it actually performs within 2% equal to the 3950x, save for overclocking or when a decent cooler is installed.
So I would not recommend the 3950x for the price.

Oh, and don't read their performance ratings. They're only valid when you're running the Ryzen 9 3900 series overclocked to their 'rated' 4,2 on all cores boost frequencies (at like 160-200W or so), which no person ever seems to hit without exotic cooling!
At the stock 65W (95W system consumption), it only hits a boost on all cores of about 3,5Ghz. At 105W (145W system) it runs 3,8Ghz, and occasionally hits 4,1 depending on the load. Those benchmarks are all done at the rated 4,2Ghz on all cores (massive cooling, and lots of VRMs).

For a 3900/3950x to make sense, I think anyone should get a dual fan (dual 120mm) liquid cooling system.
They go for about $100.

ARM processors:
From my observation on X86 PPD scoring and ARM, it fluctuates all over the place.
ARM is from 2,5 to 10x slower, but hits more of an average of about 5-8x slower on most benchmarks I've seen (depending from project to project) than x86 for similar clock speeds.
And here's the catch,
If you can run a better PPD/W on ARM, but the score comes in way too slow (takes days, vs only hours would you do it?
X86 is performing way faster, and you'll really need something with plenty of cores!
With an average runtime of 30k seconds per unit (8 hours per Wu) on ARM, vs 2000 seconds (=0.5hrs) on an x86 CPU running almost double the frequency, ARM IS 16x slower on my Rosetta WUs (8x if you do clock for clock).
They're using about 1W power consumption per core, which includes EMMC, Wifi, HDMI outputs and power supply, vs the X86 which gets RAM, PSU efficiencies and GPUs as added load.
So performance per Watt, 7nm Ryzen 9 CPUs pretty much match 10-12nm ARM processors pretty well.
Not their 12/14nm CPUs, only their latest gen CPUs (built on 7nm).

Another thing, a similarly performing system than a 3900x (24 threads) is at least 24 (quadcore) ARM boards, which are also about 24 times more troublesome to maintain!
If a server goes offline with a multicore X86 CPU, all it takes is just rebooting it.
In my experience, an ARM board usually goes offline at least once a week, sometimes even multiple times a day.
It could be from no internet, nothing to do, so it turns itself into sleep mode.
Errors, whatever...
I gave up on ARM servers a while back.
Not until they come out with an affordable 64+ core ARM CPU. I won't keep busy with the quad boards like the Pi.
BTW, if you really want to go ARM, and are thinking of the Pi, why not go with AMLogic boards?
Those $35 TV boxes they sell on Ebay?
They are cheaper, faster, and support Android.

Another issue with ARM, is that Rosetta only supports it for COVID19.
We don't know if they'll going to keep supporting it once the COVID 19 batches are finished.
My call would be:
You're much safer with X86.
18) Message boards : Number crunching : Rosetta 4.15 (short time estimates) (Message 94378)
Posted 13 Apr 2020 by ProDigit
Post:
When you preload tasks increase the setting of 0.1 day, and 0.1 additional day in boinc mgr to 3 or 4.
you'll get Wu's with longer deadline, which also means a higher chance of getting Wu's that run longer.
Rosetta has mixed WUs, some of which run close to 10 hours.


The deadlines assigned have nothing to do with how much work your machine requests. And runtimes are based on the runtime preference, not some fixed WU size.

I know, but if I set 10 day + 10 day of additional work, I usually get projects with long deadlines, which many of them happen to be long units.
19) Message boards : Number crunching : COVID 19 FAH and Rosetta the same? (Message 94377)
Posted 13 Apr 2020 by ProDigit
Post:
Are you working with the same companies, complementing or duplicating the same work as FAH?
20) Message boards : Number crunching : Rosetta 4.15 (short time estimates) (Message 94277)
Posted 12 Apr 2020 by ProDigit
Post:
When you preload tasks increase the setting of 0.1 day, and 0.1 additional day in boinc mgr to 3 or 4.
you'll get Wu's with longer deadline, which also means a higher chance of getting Wu's that run longer.
Rosetta has mixed WUs, some of which run close to 10 hours.


Next 20



©2025 University of Washington
https://www.bakerlab.org