Message boards :
Number crunching :
How to configure client to use GPU-only for MilkyWay@Home while still using CPU for N-Body Simulation
Message board moderation
Author | Message |
---|---|
Send message Joined: 26 Jul 13 Posts: 4 Credit: 67,803,592 RAC: 0 |
I've created an app_config.xml file to run multiple GPU work units: <app_config> I want my client to get CPU work units for the N-Body Simulation application, but I'd like to disable CPU work units for MilkyWay@Home 1.46. I tried disabling the CPU by adding the config below to my app_config but it doesn't seem to have an effect. I thought maybe the empty plan_class tag would tell the CPU app to use 0 CPUs, but still allow GPU tasks. <app_version> Is it possible to use GPU-only for 1.46 and still use the CPU for N-Body? Thanks, ~KT |
Send message Joined: 8 May 09 Posts: 3321 Credit: 520,555,354 RAC: 28,380 |
I've created an app_config.xml file to run multiple GPU work units: On the same computer I've never seen that before, on different pc's you can thru the 'venues' home work and school, just turn off the cpu tasks on one and the gpu tasks off on the other one. |
Send message Joined: 26 Jul 13 Posts: 4 Credit: 67,803,592 RAC: 0 |
Thanks for the response. However, even if that was a proper solution to my problem I wouldn't be able to implement it. I'm currently in a gridcoin pool so I have no control over the project settings. I was hoping for a client-side solution. ~KT |
Send message Joined: 8 May 09 Posts: 3321 Credit: 520,555,354 RAC: 28,380 |
Thanks for the response. However, even if that was a proper solution to my problem I wouldn't be able to implement it. I'm currently in a gridcoin pool so I have no control over the project settings. I thought of another way....do two different installs of the Boinc software on your pc, making each is in it's own set of directories, and don't use the gpu on one of them and only use the gpu on the other one. |
Send message Joined: 2 Oct 16 Posts: 167 Credit: 1,006,093,146 RAC: 46,584 |
Thanks for the response. However, even if that was a proper solution to my problem I wouldn't be able to implement it. I'm currently in a gridcoin pool so I have no control over the project settings. BOINC only needs to be installed a single time. Several clients can be started simultaneously using their own data directory and communicate to a manager via a different gui_rpc_port. This would be my suggestion too. I wrote a guide on my team forum on how to setup multiple clients. http://www.overclock.net/t/1628924/guide-setting-up-multiple-boinc-instances/0_30 |
Send message Joined: 26 Jul 13 Posts: 4 Credit: 67,803,592 RAC: 0 |
Correct me if I'm wrong, but this would still require access to the project settings via the web account? And each client running would have to use one of the custom preferences? For example: Client 1 uses "home" preferences that specifies GPU-Only and the N-Body application disabled. Client 2 uses "school" preferences that specifies CPU-Only and only the N-Body application enabled. |
Send message Joined: 2 Oct 16 Posts: 167 Credit: 1,006,093,146 RAC: 46,584 |
Correct me if I'm wrong, but this would still require access to the project settings via the web account? And each client running would have to use one of the custom preferences? Yes, from what I've seen most projects will have a Computer show up twice if there are two clients. Assign a different location to each to send it different work. I usually have a client for CPU work and another for GPU work. It helps set different queues for each and much more configurable during competitions. E@H has another benefit as the some of the CPU and GPU apps are named the same. BM averages the run times for CPU and GPU but they could be 10min and 2 hours with the average runtime of 30min. Keep the processors separate and the ETAs stay more accurate. |
Send message Joined: 26 Jul 13 Posts: 4 Credit: 67,803,592 RAC: 0 |
Thanks for the help. Hopefully I can use this when I'm not part of a pool and have access to the project settings. |
©2024 Astroinformatics Group