For a long time, recreational computer users all over the world have benefitted from improvements to computing systems that were invented in order to facilitate research in fundamental physics. The foremost example is, of course, the World Wide Web which you are using to read this.
Now the time has come for the gamers to give back to physics. Of course, nobody would buy that as a moral argument, but money talks louder than most ethicists, and the market for games consoles and graphics cards has become huge and strongly driven by increases in computational performance, leading to ever faster graphics processors being developed to please the gamers. If you have a moderately recent desktop computer, odds are that the graphics card has more computational power than the CPU.
How does this help scientists? Well, the kinds of operations that need to be performed efficiently to render 3D graphics are quite similar to the kinds of operations that are needed in scientific simulations (at least when looking at both at a sufficiently abstract level): matrix-vector multiplications, scalar products, and so on.
But GPUs (Graphics Processing Units, the computational core of your graphics card) have to be really fast at doing the same bunch of those operations over and over and over again in order to enable the fast high-detail rendering gamers require nowadays; this is achieved by having a very large number of processor cores running in parallel, a level of parallelism that is hard to achieve using more conventional CPUs.
Thus it can pay to outsource computationally intensive, but highly parallelisable, "kernels" (such as matrix multiplications) found in many scientific simulations to a GPU for a significant (tenfold or more not being unheard of) speedup.
In fact, the use of GPUs for "general purpose" (i.e. non-graphics) computing is now beginning to be so widespread that nVidia is now marketing its Tesla system (which is based on nVidia's CUDA architecture) as a desktop supercomputer, thus saving scientists from the embarrassment of having to put in a purchase request for a high-end gaming graphics card at the dean's office.
Another example of gamers helping science is IBM's Cell processor which introduced a novel kind of multicore architecture wherein one standard CPU drives a number of more GPU-like cores. It was developped for use in Sony's PlayStation 3, but now is used in the first Petaflop supercomputer, Roadrunner, and will be used in QPACE, a new supercomputer specialised for lattice QCD simulations.
So I guess we should say "thank you" to all those gamers out there who have helped create a big enough market for fast parallel computing to make it worthwhile for companies to develop cost-efficient massively parallel solutions that will help the advancement of science.
Now the time has come for the gamers to give back to physics. Of course, nobody would buy that as a moral argument, but money talks louder than most ethicists, and the market for games consoles and graphics cards has become huge and strongly driven by increases in computational performance, leading to ever faster graphics processors being developed to please the gamers. If you have a moderately recent desktop computer, odds are that the graphics card has more computational power than the CPU.
How does this help scientists? Well, the kinds of operations that need to be performed efficiently to render 3D graphics are quite similar to the kinds of operations that are needed in scientific simulations (at least when looking at both at a sufficiently abstract level): matrix-vector multiplications, scalar products, and so on.
But GPUs (Graphics Processing Units, the computational core of your graphics card) have to be really fast at doing the same bunch of those operations over and over and over again in order to enable the fast high-detail rendering gamers require nowadays; this is achieved by having a very large number of processor cores running in parallel, a level of parallelism that is hard to achieve using more conventional CPUs.
Thus it can pay to outsource computationally intensive, but highly parallelisable, "kernels" (such as matrix multiplications) found in many scientific simulations to a GPU for a significant (tenfold or more not being unheard of) speedup.
In fact, the use of GPUs for "general purpose" (i.e. non-graphics) computing is now beginning to be so widespread that nVidia is now marketing its Tesla system (which is based on nVidia's CUDA architecture) as a desktop supercomputer, thus saving scientists from the embarrassment of having to put in a purchase request for a high-end gaming graphics card at the dean's office.
Another example of gamers helping science is IBM's Cell processor which introduced a novel kind of multicore architecture wherein one standard CPU drives a number of more GPU-like cores. It was developped for use in Sony's PlayStation 3, but now is used in the first Petaflop supercomputer, Roadrunner, and will be used in QPACE, a new supercomputer specialised for lattice QCD simulations.
So I guess we should say "thank you" to all those gamers out there who have helped create a big enough market for fast parallel computing to make it worthwhile for companies to develop cost-efficient massively parallel solutions that will help the advancement of science.
And it takes place on Mars, which is doubly scientifically awesome.
So at least one game company is trying to give something back to science. Processor design is also package design, of course, and that is EM analysis using Maxwell's Equations. Our latest shirt: