Computing Models for FPGA-Based Accelerators.

Area of Science:

• Computer Science
• Computational Chemistry
• Hardware Acceleration

Background:

• Field-programmable gate arrays (FPGAs) are increasingly utilized as accelerators for demanding computational tasks.
• Effective application development on FPGAs hinges on selecting appropriate computing models.
• FPGA computing offers unique advantages in parallelism and associative operations.

Purpose of the Study:

• To explore the critical phase of mapping applications to suitable computing models on FPGAs.
• To demonstrate the utility of FPGA computing models for accelerating molecular modeling.
• To highlight the benefits of fine-grained parallelism and associative operations in FPGA development.

Main Methods:

• Investigated FPGA computing models, focusing on fine-grained parallelism and associative operations (e.g., broadcast, collective response).
• Developed and evaluated FPGA applications using these models.
• Conducted case studies in the domain of molecular modeling.

Main Results:

• FPGA computing models provide highly flexible fine-grained parallelism.
• Associative operations like broadcast and collective response are effectively supported.
• Case studies confirmed the efficacy of these models for molecular modeling acceleration.

Conclusions:

• The selection of appropriate computing models is crucial for successful FPGA application development.
• FPGA-based computing models are effective for accelerating compute-intensive applications like molecular modeling.
• Leveraging fine-grained parallelism and associative operations enhances FPGA application performance.