6. Using Taito-GPU
This part of the guide gives a short introduction to compiling, running, profiling and debugging applications that can utilize NVIDIA Tesla K80 and P100 GPUs installed on Taito.
GPGPU stands for General-Purpose computation on Graphics Processing Units. GPUs can be used to accelerate computationally intensive code. This is done together with CPUs by offloading some compute-intensive parts of the application to the GPUs. GPGPU implementation is well suited for data parallel and throughput intensive parts of the application. Data parallelism in this case means that GPU can execute the selected operation on different data elements simultaneously. GPUs have thousands of cores for this task (NVIDIA Tesla P100 has 3584 cuda cores). Also it is necessary to have lots of data otherwise the job is not a throughput intensive.
|Previous chapter||One level up||Next chapter|