Overall, my research is focused on analysing different computing and system architectures through the lense of sustainablity.
This includes production method, energy consumption, cost effectiveness, and accessibility.
High Performance Computing in Low Cost Sensor Networks
Despite being low power, existing embedded GPUs provide disproportionally less computational capacity than discrete GPUs and are still power hungry.
Instead we focus on cost-effective architectures like the Raspberry Pi and the Hammerblade Manycore. By devleoping optimized BLAS Libraries,
our goal is to provide high performance computing capabilities to low-cost, low-power environments.
Task-Dependent Hardware Scaling in GPU based Data Centers
We aim to efficiently utilize GPUs in data center environment.
We leverage AMD's Directed Acyclic Graph Execution Engine and CU scaling abilities to properly allocate compute resources for a single application.
This allows resources to be shared amongs a variaty of workloads; increasing throughput without affecting indiviual latencty
GPUs are present in all ranges of systems from Data Centers to Mobile Devices.
With each environment having different thermal capacites and operating temperatures. However, accurate thermal and energy simulations are inaccable to acadamia.
Using infrared imaging and power monitoring we are able to model and validate a GPUs floorplan,thermal, and energy characteristics.