At the moment under project preferences, I can choose if I want to do Nbody and/or seperation, and if I want to do GPU or CPU work. But there's no way to get Nbody on CPU and Seperation on GPU on the same host.

Milkyway's prefs: https://milkyway.cs.rpi.edu/milkyway/prefs_edit.php?subset=project
Einstein's more detailed prefs: https://einsteinathome.org/account/prefs/project

Rather large operating manual here to refer to: https://boinc.berkeley.edu/trac/wiki/ProjectMain

You could even ask the big guy - davea [at] berkeley.edu

Without changing the server's source code the project admin would have to split the separation app into
- an app for GPUs only
- an app for CPUs only

Then volunteers could select the apps they want to run per venue at the prefs page:
https://milkyway.cs.rpi.edu/milkyway/prefs.php?subset=project

Without changing the server's source code the project admin would have to split the separation app into
- an app for GPUs only
- an app for CPUs only

Then volunteers could select the apps they want to run per venue at the prefs page:
https://milkyway.cs.rpi.edu/milkyway/prefs.php?subset=project

I think that's what Peter is asking for the Server side code so the Admin over there can split the current selections into more choices

Would the preferences look something like this?

Run the selected applications:

( ) Milkyway@home N-body simulation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU


( ) Milkyway@home Separation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU

I'm being a little bit facetious about the n body GPU selections, but a fella's gotta have some goals....

PS, forgot to ask, can this really be implemented without any source code changes(not counting n body GPU selections)?

If so, this would scratch a lot of itches I have with MW@H.

That’s the selections I would like to see. The integrated Intel GPU I and thousands of others have ,we can’t use on this project. It would be nice to make a contribution with them, especially on Separation. They do a reasonable job on Einstein and Collatz so I am sure would make a worthwhile contribution here.

Would the preferences look something like this?

Run the selected applications:

( ) Milkyway@home N-body simulation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU


( ) Milkyway@home Separation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU

This is what I like:

https://www.dropbox.com/s/1j181qkr8uvxovd/options.jpg?dl=0

Pick the app you want, then tick CPU and or GPU. (Note in the example there isn't actually an option for both at the moment but you get the idea).

I'm being a little bit facetious about the n body GPU selections, but a fella's gotta have some goals....

It was tried, but the GPUs didn't go much faster than the CPUs.

That’s the selections I would like to see. The integrated Intel GPU I and thousands of others have ,we can’t use on this project. It would be nice to make a contribution with them, especially on Separation. They do a reasonable job on Einstein and Collatz so I am sure would make a worthwhile contribution here.

I think it depends on the code. Primegrid say it isn't any more effective than just using the CPU as a CPU. I too have had success with them on Einstein and Collatz though. I think they're good with double precision though, so should do well here.

Would the preferences look something like this?

Run the selected applications:

( ) Milkyway@home N-body simulation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU


( ) Milkyway@home Separation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU

This is what I like:

https://www.dropbox.com/s/1j181qkr8uvxovd/options.jpg?dl=0

Pick the app you want, then tick CPU and or GPU. (Note in the example there isn't actually an option for both at the moment but you get the idea).

I'm being a little bit facetious about the n body GPU selections, but a fella's gotta have some goals....

It was tried, but the GPUs didn't go much faster than the CPUs.

The link is pretty much what I have described, except I added specificity for the GPU type.

Would the preferences look something like this?

Run the selected applications:

( ) Milkyway@home N-body simulation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU


( ) Milkyway@home Separation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU

This is what I like:

https://www.dropbox.com/s/1j181qkr8uvxovd/options.jpg?dl=0

Pick the app you want, then tick CPU and or GPU. (Note in the example there isn't actually an option for both at the moment but you get the idea).

I'm being a little bit facetious about the n body GPU selections, but a fella's gotta have some goals....

It was tried, but the GPUs didn't go much faster than the CPUs.

The link is pretty much what I have described, except I added specificity for the GPU type.

You don't need the selection to 'use both' because if you check both the cpu and gpu box you would be doing the same thing.

As for using the Intel gpu that could involve a whole new bunch of coding that if it isn't already in there, I don't use so don't know, that could be beyond what they can do right now. It's not like you can take the AMD or Nvidia or even cpu coding and modify it to Intel Gpu coding, it's different so needs its own application. The current application page doesn't even show that they can use AMD and Nvidia gpu's at all it just mentions cpu's https://milkyway.cs.rpi.edu/milkyway/apps.php

This has morphed into a mostly GPU project anyway. Any ideas what % of total scientific output is GPU vs CPU these days?

You don't need the selection to 'use both' because if you check both the CPU and GPU box you would be doing the same thing.

Maybe. Maybe not. It seems contradictory to check CPU only, and check GPU only at the same time. Kinda makes you wonder what happened to the word 'only'. The way it's written removes that pesky English language ambiguity. At any rate, it's an implementers' choice. Besides, I'm not sure the program would be interested in this change. Although it sure would be really nice to have.

As for using the Intel gpu that could involve a whole new bunch of coding that if it isn't already in there, I don't use so don't know, that could be beyond what they can do right now. It's not like you can take the AMD or Nvidia or even cpu coding and modify it to Intel Gpu coding, it's different so needs its own application. The current application page doesn't even show that they can use AMD and Nvidia gpu's at all it just mentions cpu's https://milkyway.cs.rpi.edu/milkyway/apps.php

Agreed. People talk about using Intel GPUs in other programs, and Septimus indicates there are bunches and bunches out there that could help here. It might just be a big endian little endian thing, but probably more than that. I myself just use AMD and NVIDIA. It could be worth it to leave in there, if even only for a place holder.

This has morphed into a mostly GPU project anyway. Any ideas what % of total scientific output is GPU vs CPU these days?

I'm sure Boinc Stats has it but I don't know how to find it. As far as cpu versus gpu tasks I currently have 600 cpu tasks 'in progress' and will finish at least 99% of them before I reach my next badge at 25k. The Boinc Project wuprop https://wuprop.boinc-af.org/ measures the hours you put in for each application at each Boinc Project and you can see below I already have over 50K hours in the Separation app but only 19k in the cpu app.

MilkyWay@Home Separation 54,106.67
MilkyWay@Home N-Body Simulation 19,322.25

Would the preferences look something like this?

Run the selected applications:

( ) Milkyway@home N-body simulation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU


( ) Milkyway@home Separation
( ) CPU only ( ) GPU only ( ) both CPU and GPU
( ) Use ATI GPU ( ) Use NVIDIA GPU ( ) Use Intel GPU

This is what I like:

https://www.dropbox.com/s/1j181qkr8uvxovd/options.jpg?dl=0

Pick the app you want, then tick CPU and or GPU. (Note in the example there isn't actually an option for both at the moment but you get the idea).

I'm being a little bit facetious about the n body GPU selections, but a fella's gotta have some goals....

It was tried, but the GPUs didn't go much faster than the CPUs.

The link is pretty much what I have described, except I added specificity for the GPU type.

Probably don't need that complicating the options, since it would appear the server has to have seperate queues for each. Most projects have the CPU/Nvidia/AMD as a seperate choice, which tells your client which to ask for. The server then uses the other settings to decide what app to send out.

You don't need the selection to 'use both' because if you check both the CPU and GPU box you would be doing the same thing.

Maybe. Maybe not. It seems contradictory to check CPU only, and check GPU only at the same time. Kinda makes you wonder what happened to the word 'only'. The way it's written removes that pesky English language ambiguity. At any rate, it's an implementers' choice. Besides, I'm not sure the program would be interested in this change. Although it sure would be really nice to have.

Seems easier to have a tickbox for each thing you want. Just like two lights in a room with one switch for each. You want both on, you press both switches. You don't need a third switch for "both".

This has morphed into a mostly GPU project anyway.

Incorrect. The Nbody app does completely different research.

This has morphed into a mostly GPU project anyway.

Incorrect. The Nbody app does completely different research.

The statement wasn't about research differences, rather the project wide payload done by gpu vs cpu. You stand corrected.

This has morphed into a mostly GPU project anyway.

Incorrect. The Nbody app does completely different research.

The statement wasn't about research differences, rather the project wide payload done by gpu vs cpu. You stand corrected.

Assuming this is showing the difference between cpu's and gpu's, cpu's are NOT dead yet:

Average floating point operations per second 1,330,142.2 GigaFLOPS / 1,330.142 TeraFLOPS

yes this is just for MilkyWay at BoincStats

This has morphed into a mostly GPU project anyway.

Incorrect. The Nbody app does completely different research.

The statement wasn't about research differences, rather the project wide payload done by gpu vs cpu. You stand corrected.

Your statement wasn't clear. You stand scorned.

This has morphed into a mostly GPU project anyway.

Incorrect. The Nbody app does completely different research.

The statement wasn't about research differences, rather the project wide payload done by gpu vs cpu. You stand corrected.

Assuming this is showing the difference between cpu's and gpu's, cpu's are NOT dead yet:

Average floating point operations per second 1,330,142.2 GigaFLOPS / 1,330.142 TeraFLOPS

yes this is just for MilkyWay at BoincStats

I was surprised to see my Ryzen 9 3900XT isn't that far behind my GPUs.