- Oct 9, 1999
- 31,516
- 167
- 106
We've talked about this for years, but it looks like GPU programming has finally become a reality, thanks to a Stanford project called "BrookGPU". BrookGPU is an implementation of C-style programming for GPUs, which allows programmers to do a fair amount of C programming for programs that will run directly on their GPU.
Now if everyone will quit drooling for a second, and clean themselves up, there are some limitations to this, since it's neither a full implementation of C, and there are hardware limits to keep in mind. The main problem here is that GPUs are still specialized devices, tuned to number operations such as matricies, and will perform greatly at those taks, while things like logic are going to be far slower than a CPU. The benefit however is enormous, as a test shader program on a GeForce FX 5900 pulled 20GFLOPS/sec, about 3 times the speed of the fastest P4 CPU.
As far as implications for DC projects go, it can vary by project. SETI uses a lot of things GPUs like, such as FFTs, which will compute far faster on a GPU than a CPU, and other projects that use similar math also stand to gain a lot. No one has ported a DC project to a GPU yet, but as far as the open source projects go, it's entirely possible we may see some sort of test implementation soon.
Now if everyone will quit drooling for a second, and clean themselves up, there are some limitations to this, since it's neither a full implementation of C, and there are hardware limits to keep in mind. The main problem here is that GPUs are still specialized devices, tuned to number operations such as matricies, and will perform greatly at those taks, while things like logic are going to be far slower than a CPU. The benefit however is enormous, as a test shader program on a GeForce FX 5900 pulled 20GFLOPS/sec, about 3 times the speed of the fastest P4 CPU.
As far as implications for DC projects go, it can vary by project. SETI uses a lot of things GPUs like, such as FFTs, which will compute far faster on a GPU than a CPU, and other projects that use similar math also stand to gain a lot. No one has ported a DC project to a GPU yet, but as far as the open source projects go, it's entirely possible we may see some sort of test implementation soon.
