Get in Touch
 Duration 21 hours

Course Outline

Overview of Biren GPU Architecture

  • Biren introduction and typical applications
  • Hardware configuration: cores, memory, and compute clusters
  • Benchmarking against NVIDIA and AMD GPUs

Configuring the Biren Programming Environment

  • Installation of Biren SDK and runtime components
  • Grasping the toolchain and compiler architecture
  • Foundational project setup and build procedures

GPU Development via the Biren Stack

  • Thread and block operational models
  • Memory handling and data transfer mechanisms
  • Kernel creation and execution patterns

Migrating Code from CUDA to Biren

  • Strategies for translating CUDA codebases
  • Mapping common APIs and necessary adaptations
  • Practical sessions for code conversion and practice

Debugging and Performance Analysis

  • Utilizing Biren’s integrated debugger and profiler
  • Pinpointing performance bottlenecks
  • Analyzing memory access patterns and optimization opportunities

Performance Optimization Strategies

  • Thread scheduling and instruction pipelining techniques
  • Loop unrolling and efficient shared memory utilization
  • Advanced kernel adjustments for maximizing throughput

Real-World Case Studies and Applications

  • Training machine learning models on Biren accelerators
  • Porting and analyzing vision or NLP models
  • Performance comparison against CUDA/NVIDIA standards

Course Wrap-up and Future Directions

Requirements

  • A solid grasp of GPU architecture and parallel processing principles.
  • Hands-on experience with CUDA, OpenCL, or comparable GPU programming frameworks.
  • Proficiency with deep learning platforms such as PyTorch or TensorFlow.

Intended Audience

  • HPC developers.
  • AI infrastructure engineers.
  • Specialists in performance optimization.

Testimonials (2)

Related Categories