First Article of the first year of the year!

I happened to look at the concept of parallel programming recently. Along with cuda, the latest hardware trend has been compromised. I can't help but lament that there are many times out of date... Then we can see that Intel's Larrabee cannot help but get angry. Isn't that making 3D grating programmers fail to live ....
8800gt can be used with 128 processors to support thousands of threads. In addition, shared memory cache breaks the limitation that the memory of the video card can only be read in one way. This is a step closer to general computing. After reading the parameters, I felt I was lost. 8600gt is only 16. You can buy one at twice the cost.
128 processor 8800gtx... Lost
Larrabee is said to be 32 x86 mini core. One core has 4 threads. There are also 128 threads in total. It is mainly used for parallel computing.
I am furious at Intel's attempt to use 32 microkernel segments, transplant ray-tracing to replace the grating display hardware, and re-break the GPU into a reverse retrograde system with only the 2D bitmap copy function. Grace ..

However, it is tested that crysis is a single-threaded software. Unreal3 only uses two cores in a 4-core PC. Parallel Computing is indeed a trend in the future. Games are software in the natural parallel task mode. It is very suitable for the Design Based on Multithread from the root. However, intensive Parallel Computing Based on Data Stream parallel is not widely used except for the N-body problem in physics. In my actual project, CPU and GPU are both low-end and Middle-end platforms. Floating Point computing is not only the maximum overhead of pure rendering, but also less than 20% of the total engine overhead. Optimization has not improved significantly. The final bottleneck lies in GPU rendering, 3D API calling, and driver time. The ideal multi task should use both the CPU cores and GPU as parallel task processors for unified coordination.