Engineers at the North Carolina State University endeavored to improve the way both the CPU and the GPU perform by engineering a solution that sees the GPU execute computational functions, while the CPU cores pre-fetch the data the GPUs need from off-chip main memory. In the research team's model, the GPU and the CPU are integrated on the same die and share the on-chip L3 cache and off-chip memory, similar to the Intel's Sandy Bridge and AMD's APU platforms.
"Chip manufacturers are now creating processors that have a 'fused architecture,' meaning that they include CPUs and GPUs on a single chip," said Dr. Huiyang Zhou, an associate professor of electrical and computer engineering who co-authored a paper on the research.
"This approach decreases manufacturing costs and makes computers more energy efficient. However, the CPU cores and GPU cores still work almost exclusively on separate functions. They rarely collaborate to execute any given program, so they aren’t as efficient as they could be. That’s the issue we’re trying to resolve."
Zhou's solution was to have the CPU do the leg work by determining what data the GPU needs and then going and retrieving it from off-chip main memory. This in turn leaves the GPU free to focus on executing the functions in question. The result of this collaboration is that the process takes less time and simulations have found that the new approach yields an average improved fused processor performance of 21.4 percent.
The paper will be presented at the 18th International Symposium on High Performance Computer Architecture, in New Orleans, later this month. In the meantime, you can check out more details on the project here.