I briefly looked over a threaded list of the many topics that I haven't read, and none seem to address this.
Originally, the reason people bought graphics cards (or better graphics cards) was to improve graphics performance. Which to a dinosaur like me means X11 and X servers. I am slowly learning about GPUs and OpenCL. I have a HD5450 doing BOINC (PrimeGrid). I would like to get it doing SETI, but it seems that on Linux that means compiling source. Not a problem in that I have compiled lots of things from source, but there are a bunch of other things one is supposed to to do. And finding time to experiment is at a premium. The HD5450 is limited to single precision floating point, which limits what algorithms can be thrown at it in general. I am not looking to get into BOINC or BitCoin mining, I often find myself doing numeric intensive stuff, and maybe GPU is the way to do that. But, reading Google and DuckDuckGo, it would seem there are no solutions. At the moment, I do xhost, and then restart boinc. Which allows the GPU jobs to run. It's clunky. But, while there are lots of offered solutions, are any of them good solutions? Accessing the monitor is important, and is something the CPU should control. While GPUs are designed to help graphics, should GPU access be via X? It seems to me, that GPU (even multiple GPU, not necessarily of the same architecture) is a facility the OS should control. If a computer has multiple GPU, and X wants to do some graphics work, it should start with the most capable GPU and work down to find a GPU to do the work, and if none respond, the CPU does the work. But in terms of number crunching (for myself), I would like to see a Mersenne Twister (or other long-period) implemented to provide uniform 0-1 deviates, and then to provide deviates from other distributions. But, in an environment where X has to be in the process tree, and the active terminal has to be graphical, just to get the GPU working, is not optimal. Thank you for your time. I will go back to writing more up to date HTML and other stuff. :-) Gord _______________________________________________ Dng mailing list Dng@lists.dyne.org https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/dng
