Future Hardware and Its Implication
Jon Peddie began this roundtable discussion by describing how GPU server farms work, which can time slice. For instance, CAD users tend not to be very efficient, spending much of the day not exercising the CAD program. While the GPU server is not serving you, it is kept busy with another user, then another...
Henri Jensen of Luxion responded that the desktop keeps makes advances that causes cloud-pushed raster always seem like it is lagging. He pointed out that just yesterday Mr Peddie boasted about his office setup has two 4K monitors and another one that's vertically oriented -- "I am bathed in pixels," as he put it. Mr Jensen pointed out that that kind of data cannot be pushed from the cloud. Mr Peddie responded that a new video compression standard just being ratified now will handle this kind of huge raster stream and -- furthermore -- that Mike Riddle is working with a firm that can send realtime images of solid models from Virginia to Arizona with 100 millisecond latency. He admitted this is too slow for gaming, but fast enough for CAD users.
Now an argument is emerging between the techno-optimists and the techno-pessimists over whether the general population -- or even software -- can make use of things like the 5,000+ GPUs in the latest highest-end graphics board from AMD (nee ATi). Mr Jensen points out that CUDA is specific to just NVidia boards, while AMD and Apple support OpenCL kind of, but it requires that source code be sent, which can be intercepted by the nefarious.
There may be huge advances in compute power possible, but the practical problem we face is that it is taking longer to design new chips (with ever smaller spacing between chip components) and there is even the bigger hurdle in the cost of producing new chips.
"In the future, we will discover new problems that we don't know about today."