Question Any optimization on E@H or F@H? PG user.

[irax]

Came from PrimeGrid, where CPU affinity(core cache), 50% core count, don't store work, etc. are vital stuff.
But in E@H there seems not much posts on that at all.
- Should I set BOINC cpu usage to 50% of cores to prevent hyperthreading ruins stuff?
- Should I set affinity to match each single cores?
- Should I leave a thread or a core for each gpu tasks or its auto? (I'm running 0.5 util on my GTX1080 Dell Blower to get 2 concurrent task)
- Do I need to leave CPU for intel iGPU? (In PG, iGPU OpenCl is said to be useless as cpu cores are already contested) Is iGPU actually useful in E@H?
Meanwhile, I'm very unfamiliar with F@H (outside BOINC I'm purely novice), are there much optimization I can do?

 

[irax]

Also, is storing extra work beneficial to lower server scheduler load? I'm always connected to net with lots of storage.

 

[mmonnin03]

E@H is mainly a GPU project credit wise.
Most projects do more work with HT, I would assume E@H does too.
I set affinity to leave a CPU core free for GPUs in Windows. Linux handles it better so I do not there.
E@H is one of the most stable BOINC projects in terms of server status. I wouldn't worry about keeping extra work for down time.

FAH can use iGPU? I haven't ran it in years. FAH at least use to be just maximizing core freq while still being stable. There was some minimal gains setting the next unit percentage to 100% if your connection was fast as QRB was determined from download to return time.

 

[Skillz]

What's your BOINC username?

The only benefit to storing a lot of work at E@H is for validation purposes. If you download ~1000 tasks today, then you'll have a higher chance of them being validated in a few days. However, after a while it will eventually level out no matter what. Validation is really only important during competitions where your points only count between a set time period. On most projects, I do like to keep a nice cache of queued tasks just in case network connectivity goes awol, project server has a hiccup (as @mmonnin03 said, E@H is pretty stable), or something else goes wrong that prevents you from downloading work at the very moment. You can also increase your PPD slightly if you always have a task waiting to run vs having to download a task after you've completed and reported a task.

Affinity is important mostly on projects that are multi threaded. Projects that run 1 task per logical processor wont benefit from it much.

I wouldn't bother with the iGPU for any project. The amount of power it uses compared to the work done isn't all that great anyway.

Also welcome to the forums!

 

[irax]

E@H is mainly a GPU project credit wise.
Most projects do more work with HT, I would assume E@H does too.
I set affinity to leave a CPU core free for GPUs in Windows. Linux handles it better so I do not there.
E@H is one of the most stable BOINC projects in terms of server status. I wouldn't worry about keeping extra work for down time.

FAH can use iGPU? I haven't ran it in years. FAH at least use to be just maximizing core freq while still being stable. There was some minimal gains setting the next unit percentage to 100% if your connection was fast as QRB was determined from download to return time.

E@H uses iGPU. I asked before on PrimeGrid forum and people there say it's not worth it to use openCL on iGPU if all your cores are crunching, but E@H uses that.