It was mentioned that the GPU has thousands of cores, in constrast to current CPUs which have a much smaller number of cores (i.e. 4 or 6).
The reason for this, as Nvidia puts it, is that the cores in the GPU are smaller and more efficient, and designed to handle multiple tasks at the same time. On the other hand, the fewer CPU cores are optimized for more sequential processing.
This comment was marked helpful 0 times.
kayvonf
@RICEric22. I'll address this in class on Wed.
This comment was marked helpful 0 times.
ffeng
As @RICEric22 has mentioned, GPU's and CPU's are optimized for different kinds of tasks. Today, GPU and CPU are used together (general-purpose computing on graphics processing units also known as GPGPU) to take advantage of the two different optimizations they provide. While the GPU handles parallel computation, the CPU handles sequential computations.
This comment was marked helpful 0 times.
yihuaf
@ffeng To expand your idea into a more general idea, putting GPUs and CPUs together into one computer system is a form of heterogeneous computing. The idea is to utilize different computing components/units in one system and allow each units to be specialize to a particular task and therefore can get the most efficiency out of the whole system. This idea is a result of the need for high performance computing when we hit the power-wall.
There will be at least one lecture dedicated to heterogeneous computing later in the course.
This comment was marked helpful 0 times.
DunkMaster
Wondering if it possible for the future computer to have a simple CPU and carry out most of the computations on the GPUs.
This comment was marked helpful 0 times.
dtaveras
@DunkMaster thats a good question but I would be a bit skeptical about that since the cores in the GPU are already simple cores. My thought is that somewhere in the system you must have complex cores capable of executing sequential code very fast since at times you have no choice but to execute code serially.
It was mentioned that the GPU has thousands of cores, in constrast to current CPUs which have a much smaller number of cores (i.e. 4 or 6).
The reason for this, as Nvidia puts it, is that the cores in the GPU are smaller and more efficient, and designed to handle multiple tasks at the same time. On the other hand, the fewer CPU cores are optimized for more sequential processing.
This comment was marked helpful 0 times.
@RICEric22. I'll address this in class on Wed.
This comment was marked helpful 0 times.
As @RICEric22 has mentioned, GPU's and CPU's are optimized for different kinds of tasks. Today, GPU and CPU are used together (general-purpose computing on graphics processing units also known as GPGPU) to take advantage of the two different optimizations they provide. While the GPU handles parallel computation, the CPU handles sequential computations.
This comment was marked helpful 0 times.
@ffeng To expand your idea into a more general idea, putting GPUs and CPUs together into one computer system is a form of heterogeneous computing. The idea is to utilize different computing components/units in one system and allow each units to be specialize to a particular task and therefore can get the most efficiency out of the whole system. This idea is a result of the need for high performance computing when we hit the power-wall.
There will be at least one lecture dedicated to heterogeneous computing later in the course.
This comment was marked helpful 0 times.
Wondering if it possible for the future computer to have a simple CPU and carry out most of the computations on the GPUs.
This comment was marked helpful 0 times.
@DunkMaster thats a good question but I would be a bit skeptical about that since the cores in the GPU are already simple cores. My thought is that somewhere in the system you must have complex cores capable of executing sequential code very fast since at times you have no choice but to execute code serially.
This comment was marked helpful 1 times.