Message boards :
Number crunching :
Does 3D screen saver affect CUDA activity?
Message board moderation
Author | Message |
---|---|
Xanado Send message Joined: 25 Mar 00 Posts: 45 Credit: 7,086,573 RAC: 6 |
My BOINC starts the cool looking 3D screen saver after 5 minutes. Does this affect (or even interrupt) the CUDA activity? |
rob smith Send message Joined: 7 Mar 03 Posts: 22189 Credit: 416,307,556 RAC: 380 |
Any process that uses the graphics will have an impact on the performance of an application running on the GPU under CUDA (or OpenCL for that matter). The next question is "By how much"? There is no simple answer to that, it might be a couple of percent slower, or it might be bumped off the GPU totally depending on the process, the GPU and the settings for running the GPU CUDA/OpenCL application...... Bob Smith Member of Seti PIPPS (Pluto is a Planet Protest Society) Somewhere in the (un)known Universe? |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
The big questions are, "how much power do I want to pay for?" and "how much can/do I want to stress my hardware?". :-) Cheers |
Xanado Send message Joined: 25 Mar 00 Posts: 45 Credit: 7,086,573 RAC: 6 |
Well, the big question is "Did the CUDA-SETI@Home authors consider this and flip a switch, which may or may not exist, to hide the 3D screensaver, perhaps notifying the user of why, and giving them, insanely, the option to keep it on for some reason even if it did affect CUDA use non-trivially because we programmers are anal retentive and do not design products and want to include things like a 'do you want to switch off your artificial heart switch' within easy reach?" |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
At the time of nvidia's original app contribution we were talking 8xxx GPUs, and they came back on with a brief patch for Fermi release. Before Cuda 2.3 (and before Fermi) the FFTs and other libraries used were pretty 'clunky', and the screensaver was definitely a non-starter for general applications. that switch was in the code (hardwired). The current replacement stock Cuda multibeam app was originally 3rd-party optimised (Lunatics), as nVidia didn't come back for v7 updates and we had workable apps already anyway, and total disabling of the screensaver was always standard practice. With changing technologies, and some ways to make the graphics app possibly more lightweight and useful, I have been looking at possibilities for resurrecting it in some form, though no it's not a quick switch anymore. If and when I do so it would probably use a leaner modern graphics api, and not have 'bubbly wobbly bits', but more of a telemntry/monitoring tactical feel, meant to be useful in future heterogeneous situations (as well as impressing people at parties) "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
... and total disabling of the screensaver was always standard practice. He is talking about the "3D screen saver" of the stock CPU app - and wants somehow the CUDA app to stop the screensaver of the CPU app :D Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
... and total disabling of the screensaver was always standard practice. Well that's more complicated indeed :D. Not going to try that myself, thought should totally be possible through killing processes [apparently killing orocesses is normal in boincland. Why would you start a sreensaver you didn't want ?] "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Xanado Send message Joined: 25 Mar 00 Posts: 45 Credit: 7,086,573 RAC: 6 |
Well, I do want it, as it is indeed cool looking. However, I assume it uses the GPU, however lightly, rather than being a pure software operation, and so does that affect the CUDA, slowing it down, or even disabling it if you have "Disable CUDA when GPU in use" checked? Currently I have, via Windows, turned off the screen saver (though the other screen saver that shuts off the monitor is still working.) For that matter (and maybe this should be in a second thread) I have a hexacore machine, so it runs 6 normal calculations and 1 GPU CUDA application, which requires "0.227" of a normal CPU to keep it fed with stuff to do. So, guess the question: Does the CPU part of the CUDA calculation have higher priority than the normal CPU analysis, which perforce one of the six must be sharing a processor with? Now the CUDA is blazing fast, doing in 15 minutes what the processor takes 5 hours for, so it isn't that much of a slowdown at worst...or is it? If it adds 3 minutes, that's a 33% slowdown if the CPU part of the CUDA is the bottleneck for CUDA. Has this interaction been analyzed, either? ETA: Ok checked it out on the process monitor. The non-CUDA CPU apps are all bottom priority, and the CUDA-feeder CPU app is the next step up. I am happy. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Historically the penalty was around 10%-20% GPU utilisation and some substantial CPU (system dependant) cost for shuffling data to the graphics application. Back then, that was enough to break the processing sooner, by simply extending the long parts of processing by too much, and the drivers would reset. That was a long time ago now though, on older generations of GPUs, and since no graphics enabled GPU app existed here since those times, the current situation is not known. I suspect the GPU cost would be lower on modern cards, but the CPU and resulting time cost much the same (with current standard code). That's something I'll probably analyse in greater detail when I eventually look at adding the mentioned functionality (Actually before I do). By that time, the lower latency Vulkan and DX12 apis may be available, along with specialised hardware for moving that data around without as much CPU intervention. In other words it's possible at least some basic graphics app would then use parts of the hardware not already in use by the Cuda or OpenCL processing, so be more or less free by comparison to the old situation. Time will tell though. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
Traditionally the feeling always has been, that if you optimise (and the CUDA apps started out as being the optimised 3rd party version, though they are now the only thing) then you'll want as much processing power as possible to go towards the science and you don't use the pretty screensaver anyway, since it would use up some processing power. Of course that's been in the dark ages (as far as computing power is concerned) and maybe now the little bit you'd sacrifice on a screen saver would be acceptable to a part of the community. @Xanado do you actually want the option of a screensaver or were you just curious? A person who won't read has no advantage over one who can't read. (Mark Twain) |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
There are times and places where displaying a screensaver can be a positive benefit to the project - if you can leave a computer running in a public place, say a school or college, a shop or exhibition hall, it might tickle the curiosity of passers-by, and from that initial contact with ET, a lasting interest or even participation can grow. That's a different motivation from maximum efficiency in the current search. The question is whether there's enough overlap between the two roles - searching vs. recruiting - to justify the development work. If not, the publicity role falls to those who are prepared to run the stock app. |
Jord Send message Joined: 9 Jun 99 Posts: 15184 Credit: 4,362,181 RAC: 3 |
Nvidia has CUDA screen savers. Both it and OpenCL were developed with doing calculations on the GPGPU and show intricate 3D graphics at the same time. See Fast N-Body Simulation with CUDA which does an N-body simulation both in calculations and graphics in real time. Or see OpenCL Tutorial: N-Body Simulation. Einstein@Home shows that BOINC can perfectly fine run work on the GPU and do a screen saver on that same GPU. They offer the graphics application together with the GPU apps (both CUDA and OpenCL), as well as the CPU apps. Here at Seti, and only on Windows the CPU apps have the graphics application, which can run perfectly fine together with work being done on the GPU. One after another, people running a Macintosh are asking where the Seti screen saver went. It's not available for their Seti v7 or Astropulse applications. Why not? Who knows, perhaps because someone decided that the users didn't need it, or that it slows down the search on their computers. There can be plenty of threads found in Q&A as well on people wanting to see the screen saver on their GPU tasks. That graphics slows down the calculations on other people's computers is not something that you have to worry about, that's an option the user has to decide upon. It's the user who gets to decide if they want to run a screen saver, or not. Not the application developers. The graphics application is available, it's separate from the science application, it just needs porting to v7 and be made available to those willing to run it, on whichever piece of hardware they want. Well, that's my opinion, at least. :-) As for a new screen saver development, I do agree that it is about time a new one came to Seti, possibly have a different one for Astropulse and GBT as well. |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
One after another, people running a Macintosh are asking where the Seti screen saver went. It's not available only for the newer CPU apps (avx ssse3 sse41) introduced on 18 Sep 2015 http://setiathome.berkeley.edu/apps.php The old app (29 May 2013) (setiathome_7.00_i686-apple-darwin + the Graphics part setigraphics_7.00_i686-apple-darwin) is still distributed and on the SETI@home servers: https://setiathome.berkeley.edu/forum_thread.php?id=78228&postid=1734104#1734104 Of course the server will send more often tasks for the faster CPU apps  - ALF - "Find out what you don't do well ..... then don't do it!" :)  |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.