Sungbum Hong research blog

My research would primarily focus on performance and security improvement in GPU.

Unlike CPU that has been core compartment of the computer for very long time, GPU recently became crucial processing unit for the computer.

However, popularity of GPU started increasing dramatically in recent years. Now, GPU not only processing graphic but also other computer instruction which requires high parallelism.

Despite this every increasing demand for GPU, the research conducted in order to increase performance and security of GPU is scarce compared to CPU.

Moreover, many optimization and security technique developed for CPU hasn’t been properly transcribed to GPU architecture due to inherent difference in design.

During my summer research, I plan to device some optimization scheme inspired from CPU and implement these methods to GPU.

Currently, I plan to focus on data vulnerability of GPU data and minimization of overhead resulted from encryption and decryption process.

I speculate that speculative execution of GPU utilizing its parallel structure can hide overhead arised from encryption process and decrease latency for each cycle to be finished.

In order to evaluate this scheme, i first plan to measure overhead of encryption process and its ratio with number of inputs. Then use that data to normalize optimization result.