On Sunday 12 April 2009 01:11:36 pm Bob Friesenhahn wrote: > On Mon, 13 Apr 2009, Guy K. Kloss wrote: > > Well, I'm much more afraid of another issue: GPUs are being used > > much more and more by other processes they more belong to. E. g. the > > composite manager, etc. And one thing they're not good at at all is > > task switching. So if more *applications* on a general desktop are > > competing for the resource that could lead to a massive slowdown. In > > scientific computing that's not an issue (as I see here) as they are > > performed on dedicated boxes where one expects not to have a smooth > > user feeling. > > This is certainly a valid concern and stems from the same problem that > the GPU is an adjunct to the computer system architecture which has > been refined over the past 30 years. Modern computer system > architectures provide resource management and security. Besides > multiple applications contending for the same resource from one user, > you may have applications for multiple users contending for the same > resource. This is something that the traditional operating system > handles very well, but which is not adequately handled once one > departs from the operating system. > > To put this in context, if you were using lcms as part of a batch > processing application to color correct thousands of images (perhaps > part of a printing subsystem), then there could be many tens of > simultaneous requests for the same resource, and not necessarily by > the user who currently owns the system display. > > If Firefox is using the GPU, the operating system is using the GPU, > and a random lcms-using application is using the GPU, then there will > surely be contention and perhaps previously unknown security issues as > well (e.g. ability to deadlock the display). > > Bob
I have read reports on the net of users having more than one GPU so that they could use the second unit as a GPGPU and still leave the main GPU free to handle the display. The report I remember reading said that this was to prevent X from hanging if the GPGPU software caused a problem that hung execution of the GPU. This implies that the user is somehow able to create an affinity between the GPGPU software and a specific GPU unit in the machine. But I don't think this would be possible for cases like firefox since users likely would not have a way to configure this. So another issue is that these other apps using the GPU have the potential to to put the GPU in a state that causes it to hang preventing X from working. In addition there is getting to be more software out there that runs on the GPU. For example those here who do panoramas may have used a program named enblend and enblend since version 3.0 has had support for using the GPU to do it's blending operations. On my machine this reduced blending times by about 85%. So it is clear that using the GPU can result in dramatic performance increases. Enblend has had this ability for about 2 years now. AMD is working on a new architecture that they have named "Fusion" that combines a GPU onto the same silicon as the CPU with the intention that it will be used for both display tasks as well as general computing. This change in architecture is the main driver behind AMD's purchase of ATI and it is clearly AMDs direction going forward. In fact it may give them a significant advantage over Intel since the ATI GPUs are significantly more powerful than those from Intel unless Intel steps things up to a new level before AMD introduces the new architecture. These "Fusion" CPU/GPUs should start appearing in late 2010 or early 2011 in AMD mobile CPUs. Hal ------------------------------------------------------------------------------ This SF.net email is sponsored by: High Quality Requirements in a Collaborative Environment. Download a free trial of Rational Requirements Composer Now! http://p.sf.net/sfu/www-ibm-com _______________________________________________ Lcms-user mailing list Lcms-user@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/lcms-user
