You are asked to accelerate a CPU serial algorithm by off-loading some computation to GPU. Suppose that your kernel func
Posted: Sun May 15, 2022 12:57 pm
You are asked to accelerate a CPU serial algorithm by
off-loading some computation to GPU.
Suppose that your kernel function is able to parallelize the
computation across thousands of
streaming processors (SPs) on the GPU. However, your benchmarking
shows that your new
algorithm is actually slower than the original CPU serial
algorithm. Can you describe one of the
potential causes for the performance degradation?
off-loading some computation to GPU.
Suppose that your kernel function is able to parallelize the
computation across thousands of
streaming processors (SPs) on the GPU. However, your benchmarking
shows that your new
algorithm is actually slower than the original CPU serial
algorithm. Can you describe one of the
potential causes for the performance degradation?