Message boards :
Number crunching :
cannot identify GPU task from CPU task
Message board moderation
Author | Message |
---|---|
![]() ![]() Send message Joined: 18 Nov 08 Posts: 291 Credit: 2,461,705,302 RAC: 1,123 ![]() ![]() |
There is no identifying information in the returned task id that allows one to determine whether a GPU was used or not. I did notice that the application version was 20 for gpu tasks and 19 for cpu tasks. I asssume those versions are subject to change when your code changes and should not be used to determine if a GPU was used or not. How does one determine whether a GPU was used and for that matter, which GPU? I keep statistical infomation http://swri.info/images/bs.txt about my various projects and like to be able to compare the performance of one system to another and this project does not provide key information. Obviously, there is a huge difference in processing time as shown below ![]() The problem is that the this project show nothing useful here ![]() Whereas other gpu accelerated projects provide meaningful results like these ![]() At a minimum, please include the phrase "CUDA" in the results where a gpu (CUDA) coprocessor was used. thanks! |
Send message Joined: 26 Jun 09 Posts: 47 Credit: 276,827,695 RAC: 0 ![]() ![]() |
Isn't the CPU runtime a pretty good clue? Bryan ![]() |
Send message Joined: 16 Jun 09 Posts: 85 Credit: 172,476 RAC: 0 ![]() ![]() |
I have a new version ready, after the weekend I'll put it up on BOINC. |
![]() ![]() Send message Joined: 18 Nov 08 Posts: 291 Credit: 2,461,705,302 RAC: 1,123 ![]() ![]() |
I have a new version ready, after the weekend I'll put it up on BOINC. Thanks Anthony! BTW, are there any functions in the gpu that can provide the time the gpu actually spends processing its client? I have seen elapsed time being displayed, but I think that includes time suspended. Observation: I also see that results are being deleted out of the database very quickly. Yesterday I ran a statistical comparision http://swri.info/images/bs.txt and got 28 cuda results with many more pending from milkyway, today I see very few results and all of them are pending. If a space problem, the results can be deleted from the server, but there should be some statistics left in the database for the user to review. thanks for looking |
![]() ![]() Send message Joined: 18 Nov 08 Posts: 291 Credit: 2,461,705,302 RAC: 1,123 ![]() ![]() |
I have a new version ready, after the weekend I'll put it up on BOINC. Anthony: The latest cuda results do not include any phrase such as "cuda" or identify the type of coprocessor that was being used. Can you add some identifier to the CDATA field of the taskid results? For example <core_client_version>6.10.6</core_client_version> <![CDATA[ <stderr_txt> using cuda device gtx 280 (or whatever) called boinc_finish </stderr_txt> ]]> I am working on a performance analysis program here and milkyway is not being properly identified as to whether a cpu or a gpu is being used. thanks! |
©2025 Astroinformatics Group