Einstein@Home
- Thread starter RONNIE
- Start date
@Vester : "Ronnie, I added Einstein to my mining rig and cannot understand why Einstein only uses five GPUs instead of 8. I wrote an app_config.xml file and a cc_config.xml file and it still uses only 5 GPUs. From the posts I've read at Einstein they must all be SETI people, e.g. humorless and unhelpful".
Originally posted here: https://www.boincusa.com/forum/index.php?threads/milkyway-home.1849/page-3#post-6273
Originally posted here: https://www.boincusa.com/forum/index.php?threads/milkyway-home.1849/page-3#post-6273
I've used it for a very short time and with only one GPU. Have you seen the threads below? Not exactly what you need but perhaps they can help a little bit.
GENERIC MULTIPLE GPU DISCUSSSION
HOW DO I USE BOTH MY COMPUTER'S GPU'S?
BOINC GPU TASKS - COMPUTATION ERRORS
RUNNING MULTIPLE TASKS CONCURRENTLY PER GPU - HOW TO REVERT TO JUST ONE TASKI'm late replying here but yes, Einstein requires 1 thread per task at minimum. I've read some of the work may take nearly 2 to properly support the GPU. I haven't run anything there for quite awhile as my rigs are old and pretty much all GPU work is right on the edge of being CPU bound, if not over it.
I've run Einstein off and on for a while now. Best to leave at least one CPU core free to deal with GPU work loads. The good news is a 2080ti will crunch thru each task in about 3 minutes and the 3080ti can do it in 2. Lots of points to put on the board...
I don't have a high performance GPU since I sold my Radeon HD 7990s, but I am running an old Radeon HD 7790. I noticed that the usage was below 50% most of the time. Here is the app_config.xml file in notepad format to run three tasks per GPU with one CPU core per task. The GPU utilization is now 99%. I will experiment with using one CPU core for three tasks, <cpu_usage>0.33</cpu_usage>. (Edit: One CPU core per GPU works fine when running three tasks per GPU.)
Code:
<app_config>
<app>
<name>einstein_O3AS</name>
<gpu_versions>
<gpu_usage>0.33</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
</app_config>
Last edited:
I'm sure you probably know this so forgive me if you do, but i bet a lot of our new members dont... If you want to put more points on the board then you need to de-select the Allsky GPU jobs, they take too long and pay too little. Here are the jobs I'm running;
Binary Radio Pulsar Search (Arecibo, GPU)
Gamma-ray pulsar binary search #1
Gamma-ray pulsar search #5
Gamma-ray pulsar binary search #1 (GPU)
Gravitational Wave search O2 Multi-Directional
Here is my App_Config.xml for the Einstein Project, using this I can give preference to the GPU jobs and still leave a little for the CPU jobs.
Einstein:
<app_config>
<project_max_concurrent>5</project_max_concurrent>
<app>
<name>hsgamma_FGRPB1G</name>
<max_concurrent>3</max_concurrent>
<gpu_versions>
<gpu_usage>.33</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>
</app_config>
I save most of my CPU capability for Rosetta on this machine. Here is the App_Config for it giving preference to the Python jobs;
Rosetta:
<app_config>
<project_max_concurrent>9</project_max_concurrent>
<app>
<name>rosetta_python_projects</name>
<max_concurrent>8</max_concurrent>
</app>
</app_config>
Hope your staying warm!!
I didn't know this, I've been running whatever jobs are thrown at my RX580. Time to research how to set this up on my Windows rig. Thansk for the tips.
Every app for every project can be configured. That said I've only seen hsgamma_FGRPB1G tasks for the GPU all year.
If you ever want to adjust the app configuration for work units you are getting and you don't know the name of it you can always open your client_state.xml and search for <user_friendly_name> to quickly find it. Alternatively you can search for <name> or <app> but you'll get more irrelevant stuff to sift through.
Wow, My Einstein stats have almost doubled since following this guide. I tweaked these to match my CPU preferences, and I'm running World Community Grid as a secondary but, yeah, thanks for this! 
Last edited:
When running on Pop!_OS, I have a failure rate of about 25% of GPU tasks, mostly MeerKAT. That offsets the 25% advantage of running Linux. This would not happen so often if there were few Windows OS computers compared to Linux computers. I am still undecided about which OS to use to get peak points per day.
At the E@H forum:
At the E@H forum:
Keith Myers said:
Gamma-ray pulsar binary search #1 on GPUs (hsgamma_FGRPB1G) tasks seem to be the most reliable with the best credit.
A little surprising the Einstein team would let this through. There should be a way to match task wingmen on equivalent hardware: Nvidia to Nvidia, AMD to AMD etc. Maybe they don't because they want the most stringent double-checking? Either way very annoying to get invalids on a stable system. I have seen more invalids with Einstein than most projects and I bet this is why.
I'm a little confused, what is MeerKat on Linux?
At the last BOINC Workshop Colin Clark was saying they don't need perfect accuracy and they do "flexible validation" so results don't have to perfectly match a wingman's. Regardless something is messed up with how things are now with the new Binary Radio Pulsar Search (MeerKAT) (BRP7) application.
Binary Radio Pulsar Search (MeerKAT) (BRP7) is MeerKAT. I had about 25% failures when running those on GNOME 42.5 (Pop!_OS). Thanks for your help, Jason Jung.
On another note, I had self-inflicted failures and lost about 12 hours of crunching on Saturday/Sunday. Then last night and early this morning, we had bad storms in my area with power outages. The computer had graceful shutdowns but I was asleep some of the time.
Our team is in 11th place, and can overtake Auburn for 10th again.
On another note, I had self-inflicted failures and lost about 12 hours of crunching on Saturday/Sunday. Then last night and early this morning, we had bad storms in my area with power outages. The computer had graceful shutdowns but I was asleep some of the time.
Our team is in 11th place, and can overtake Auburn for 10th again.