Hi. My PC (AMD Ryzen 9 3950X) downloads only x86 tasks. How can I set Rosetta@Home to download x86_64 tasks?
https://boinc.bakerlab.org/rosetta/results.php?hostid=3778294

---

ASUS X570 E-Gaming
AMD Ryzen 9 3950X, 16 core / 32 thread 4.4 GHz
AMD Radeon Sapphire RX 480 4GB Nitro+
Nvidia GTX 1080 Ti Gaming X Trio
4x16 GB Corsair Vengeance RGB 3466 MHz

Last I knew, the 64bit was just a 64bit wrapper over the 32bit application. So, shouldn't make a difference either way.

---

Rosetta Moderator: Mod.Sense

This has me confused also. Why is it downloading an INTEL i686 4.07 application and tasks when the cpu is an AMD x86-64 cpu which should be running the 4.08 x86 application.
Is the 32 threads of my Ryzen 9 3950X confusing the scheduler into thinking I have an INTEL cpu?

I see every one of my teammates running the 4.08 x64 application and tasks. The 4.07 app is exceedingly slow on my cpu.

This has me confused also. Why is it downloading an INTEL i686 4.07 application and tasks when the cpu is an AMD x86-64 cpu which should be running the 4.08 x86 application.

I expect it's similar to Seti & the "Windows/x86 running on an AMD x86_64 or Intel EM64T CPU" application, except in this case they left of the AMD x86 bit.

The 4.07 app is exceedingly slow on my cpu.

Here the WUs run for a specific time, the default is 8 hours.
In the Rosetta preferences, is "Target CPU run time" from 1 hour to 1 day.

---

Grant
Darwin NT

This has me confused also. Why is it downloading an INTEL i686 4.07 application and tasks when the cpu is an AMD x86-64 cpu which should be running the 4.08 x86 application.

I expect it's similar to Seti & the "Windows/x86 running on an AMD x86_64 or Intel EM64T CPU" application, except in this case they left of the AMD x86 bit.

The 4.07 app is exceedingly slow on my cpu.

Here the WUs run for a specific time, the default is 8 hours.
In the Rosetta preferences, is "Target CPU run time" from 1 hour to 1 day.

I saw that in Preferences but have no clue what it means.

Can you explain what a "Target cpu run time" does.

You can choose the length of WUs on Rosetta. A WU just crunches decoys for as long as you choose (don't worry, it will always complete the latest decoy it is crunching, it's not like after 6 hours it abruptly stops interrupting the decoy running!) and then it is completed and sent and another WU start. At a minimum a decoy per WU is completed, if doing only 1 decoy takes more than the time limit you choose.
If your PC is on only a few hours per day then having a short time limit can help you complete WUs in a short time instead of taking a week to crunch 24 hours for a WU.

The Rosetta project crunches a lot of decoys for each protein to find the best shape. If you go check in your account in the task section you can see how many decoys you completed for each WU. For example:

DONE ::     4 starting structures  25972.6 cpu seconds
This process generated      4 decoys from       4 attempts

---

The 4.07 app is exceedingly slow on my cpu.

Here the WUs run for a specific time, the default is 8 hours.
In the Rosetta preferences, is "Target CPU run time" from 1 hour to 1 day.[/quote]

That's right, so you should see fairly consistent runtime, inline with your runtime preference. So if you were expecting to complete the tasks faster than other types of CPUs, perhaps that is why you say they are slow?

R@h sorta works like your teacher in school. If you finish the assignment early, the teacher gives you another one. With R@h, each WU can be used to compute an infinite number of models. Each time a model is completed, it pops up, looks around, and decides if it should go run another model, or cut it off and report the results.

Beware, if you change your WU runtime preference, the BOINC Manager work requests take a while to adjust. So, I typically recommend people ensure their network settings are set to only have low amounts of work around when changing the runtime preference. Otherwise, if you change the 8hr default time to 24 hours, the BOINC Manager ends up requesting three times more work than it really needs. If you combine that with a 3 days cache of work requested, it ends up taking 9 days to complete it all.

So, either small work cache in the network preferences, or change work unit runtime preference incrementally. Going a notch or two in the direction of your desired value each day.

---

Rosetta Moderator: Mod.Sense

Last I knew, the 64bit was just a 64bit wrapper over the 32bit application.

In 2020, we have not a Windows 64 native app.
It's a pity.

---

The 4.07 app is exceedingly slow on my cpu.

Here the WUs run for a specific time, the default is 8 hours.
In the Rosetta preferences, is "Target CPU run time" from 1 hour to 1 day.

That's right, so you should see fairly consistent runtime, inline with your runtime preference. So if you were expecting to complete the tasks faster than other types of CPUs, perhaps that is why you say they are slow?

R@h sorta works like your teacher in school. If you finish the assignment early, the teacher gives you another one. With R@h, each WU can be used to compute an infinite number of models. Each time a model is completed, it pops up, looks around, and decides if it should go run another model, or cut it off and report the results.

Beware, if you change your WU runtime preference, the BOINC Manager work requests take a while to adjust. So, I typically recommend people ensure their network settings are set to only have low amounts of work around when changing the runtime preference. Otherwise, if you change the 8hr default time to 24 hours, the BOINC Manager ends up requesting three times more work than it really needs. If you combine that with a 3 days cache of work requested, it ends up taking 9 days to complete it all.

So, either small work cache in the network preferences, or change work unit runtime preference incrementally. Going a notch or two in the direction of your desired value each day.

Well based on how my 3950X runs other cpu project tasks, it is not running very well on Rosetta. I run a locked 42 multiplier on all cores and 3.6Ghz CL14 on the memory. I normally process a cpu task in 30-60 minutes.
What I did not like was that the very first download of work netted way too much, 260 tasks in fact. No way that is going to finish in time before deadline.

You can set Rosetta@Home to work for a specific time. For example, I set the project to work for 6 hours, so my PC will complete the tasks in time.

---

ASUS X570 E-Gaming
AMD Ryzen 9 3950X, 16 core / 32 thread 4.4 GHz
AMD Radeon Sapphire RX 480 4GB Nitro+
Nvidia GTX 1080 Ti Gaming X Trio
4x16 GB Corsair Vengeance RGB 3466 MHz

Does Rosetta obey using a max_concurrent statement in an app_config? I am having issue with out of memory issues preventing my gpu tasks from running and I am not able to well control just using the %cpu setting in Preferences.

Does Rosetta obey using a max_concurrent statement in an app_config? I am having issue with out of memory issues preventing my gpu tasks from running and I am not able to well control just using the %cpu setting in Preferences.

Keith Myers,

If you were to choose 8 CPU hours on the prefs page, the WU would run for 8 CPU hours.
If you choose 24 hours, then it's 24 hours.
If a WU does go on for longer than the specified CPU target runtime, a cut-off happens 4 hours later.

Yes, I have an app_config limiting Rosetta to 4 concurrent work units due to memory issues.

Does Rosetta obey using a max_concurrent statement in an app_config? I am having issue with out of memory issues preventing my gpu tasks from running and I am not able to well control just using the %cpu setting in Preferences.

Hi Keith

Make sure you are setting the limit of "% of the CPUs" rather than "% of the CPU time".

D

---

Beware, if you change your WU runtime preference, the BOINC Manager work requests take a while to adjust. So, I typically recommend people ensure their network settings are set to only have low amounts of work around when changing the runtime preference. Otherwise, if you change the 8hr default time to 24 hours, the BOINC Manager ends up requesting three times more work than it really needs. If you combine that with a 3 days cache of work requested, it ends up taking 9 days to complete it all.

So, either small work cache in the network preferences, or change work unit runtime preference incrementally. Going a notch or two in the direction of your desired value each day.

Well based on how my 3950X runs other cpu project tasks, it is not running very well on Rosetta. I run a locked 42 multiplier on all cores and 3.6Ghz CL14 on the memory. I normally process a cpu task in 30-60 minutes.
What I did not like was that the very first download of work netted way too much, 260 tasks in fact. No way that is going to finish in time before deadline.

I thought you mis-typed that number, so I checked. That first download was crazy. I guess it assumes you have a big debt to the new project.
If it's any consolation, as your tasks run Boinc quickly works out how many you really need, your uptime, work for other projects etc and doesn't do it again. If you miss deadlines, they'll abort of their own accord.
I'd also say this was a Boinc scheduling issue on that first download, not a Rosetta one

---

Yes, probably BOINC doing it's foolish things as usual. Deadline of 4 April and all the tasks are running in EDF or "High Priority" mode. No way to finish 260 of them in time. So the majority will time out and get resent to others. I have a rather high RAC for Seti, GPUGrid and Einstein and BOINC was probably trying to match REC with. Just set NNT and will wait them out. I also put in place an app_config and other controls to reduce the number of tasks in cache eventually to 20.

I added another 16GB of memory to the machine by stealing from others. That will allow me to run a reasonable amount of cores now. Was a waste of 3/4 of my 32 cores available because I didn't have enough RAM.

Yes, probably BOINC doing it's foolish things as usual. Deadline of 4 April and all the tasks are running in EDF or "High Priority" mode. No way to finish 260 of them in time. So the majority will time out and get resent to others. I have a rather high RAC for Seti, GPUGrid and Einstein and BOINC was probably trying to match REC with. Just set NNT and will wait them out. I also put in place an app_config and other controls to reduce the number of tasks in cache eventually to 20.

I added another 16GB of memory to the machine by stealing from others. That will allow me to run a reasonable amount of cores now. Was a waste of 3/4 of my 32 cores available because I didn't have enough RAM.

Right, so you're saying your RAM limitation is making it impossible to run all 32 cores on these tasks even if you manipulated things round (because 3x8hr x32 cores = 96 per 24hrs). Makes sense then. I have no idea what tasks you already have for other projects on top. Still, I'd try to run as many Rosetta as you can if I were you (which I'm not)

---

I am still running my original projects. Seti.Milkyway.Einstein and GPUGrid. I was reduced to only running one of my gpus when I didn't have enough memory to run the 6 Rosetta tasks when I only had 16GB in the host.

Kept getting out of memory error messages in the log and only one gpu running.

Adding the extra 16GB allowed me to regain my gpu work and also get 12 Rosetta tasks running.

From Rosetta@Home twitter account, the answer of a question:

I know, you have a lot of things to do, @RosettaAtHome , but when the COVID-19 emergence is finished, please, please a Windows 64 bit native app, not wrapper.

In the works! Stay tuned

Even if the native app has 0,1% of benefit from wrapper, the firsts 10 hosts in the project will overclass the lost of all 32bit hosts still active.

---

I have a principal PC with a Ryzen 9 3950X and I'm building a second PC with a Ryzen 7 1700 CPU. My girlfriend has a notebook with another Ryzen 7 1700, so I have a lot of computational power to help Rosetta@Home, but please, a 64bit app. Please!!!

---

ASUS X570 E-Gaming
AMD Ryzen 9 3950X, 16 core / 32 thread 4.4 GHz
AMD Radeon Sapphire RX 480 4GB Nitro+
Nvidia GTX 1080 Ti Gaming X Trio
4x16 GB Corsair Vengeance RGB 3466 MHz

What's the fascination with a 64bit application? Whether or not an application is 64bit or 32bit generally doesn't affect it's performance, only it's ability to address more RAM.
Making a 64bit version of a working 32bit application can actually result in a performance penalty due to the larger 64bit registers that have to be addressed. If a 32bit programme were to be re-written to take advantage of the extra registers, it may provide a performance boost- noticeable in benchmarks but not often in actual applications- unless their register use is a significant portion of their processing time.

Better to develop an application that takes advantage of AVX, as that would provide a significant boost in performance.

---

Grant
Darwin NT