Message boards :
Number crunching :
Computing Preferences Settings
Message board moderation
Author | Message |
---|---|
Send message Joined: 12 Nov 21 Posts: 236 Credit: 575,038,236 RAC: 4,912 |
I have 3 computers running MW@H. Slug, Potato, and the Big Iron. For Slug and Potato, the computing preferences are set to store at least 2 days of work, and store up to an additional 0.5 days of work. This has allowed them to be at essentially 100% GPU/CPU throughout the server down times. I.e. no work lost. The Big Iron has the exact same settings, but this does not work. It does indeed store 2.5 days for the GPU, but only about 5 hours or so extra for the CPU(Ryzen 9 3900X, 12 core 32 threads). So it has run dry for CPU tasks on every outage. Opportunity lost! Can this be fixed by user preference settings? As an experiment, I disallowed all Big Iron new GPU work and allowed all GPU tasks to complete. The hope was that after the GPU tasks ran out, that BOINC/MW@H would honor the 2.5 day preference to kick in, and bring in 2.5 days worth of CPU work. Didn't happen. Any ideas? If not, I will add this to the make a wish list. |
Send message Joined: 8 May 09 Posts: 3319 Credit: 520,346,569 RAC: 21,896 |
I have 3 computers running MW@H. Slug, Potato, and the Big Iron. For Slug and Potato, the computing preferences are set to store at least 2 days of work, and store up to an additional 0.5 days of work. This has allowed them to be at essentially 100% GPU/CPU throughout the server down times. I.e. no work lost. The Big Iron has the exact same settings, but this does not work. It does indeed store 2.5 days for the GPU, but only about 5 hours or so extra for the CPU(Ryzen 9 3900X, 12 core 32 threads). So it has run dry for CPU tasks on every outage. Opportunity lost! I think the problem is that MW has a max number of tasks an individual pc can get at any one time, meaning you are running thru your daily limit and MW isn't allowing you to get more tasks for the day. You can maybe work around this a little bit by telling Boinc you have MORE cpu cores than you really do but if the limit is say '300 cpu tasks per person per day' then nothing will work except you setting up a 2nd cpu project with a zero resource share. That means that project only sends you tasks if MW isn't and you only get enough to fill the available cpu cores not the full cache size. This is a huge problem for people with the faster gpu's in that they run thru their tasks very fast and then quickly hit their limit for the day. |
©2024 Astroinformatics Group