• Duration: 2 days
  
• Prerequisites: knowledge in C or Fortran, Linux ; knowledge in CUDA is a plus
  
• Training limited to 14 attendees
  

Download datasheet

Objectives

In this 2 day HMPP training, participants will learn the HMPP programming model and tips to reach performance. The training includes hands-on practical labs where participants will be able to progressively program high performance computations.

Training content

Day 1

Morning - CUDA Basics

• Introduction to GPU computing
• CUDA architecture and programming model
• CUDA API
• CUDA tools: debugging
• Introduction to CUDA performance

• Introduction to parallel hybrid programming
• HMPP overview
• Basis of HMPP programming
• HMPP compilation model

• Managing data transfers
• Grouping GPU computations
• Data movement optimizations

• Advanced CUDA performance
• Driving gridification
• Applying loop transformations: unrolling, splitting, jamming

• Duration: 3 days
• Prerequisites: knowledge in C, CUDA basis ; knowledge in OpenGL is a plus
• Training limited to 14 attendees

Download datasheet

In this 3 day CUDA training, participants will learn how to visualize their physical unit. The training includes hands-on practical labs where participants will be able to progressively build a graphic view of their results.

Training content

Day 1

Morning - CUDA Basics

• Introduction to GPU computing
• CUDA architecture and programming model
  
• CUDA API
  
• CUDA debugging
  
• Lab session: getting used to CUDA device
  
• Lab session: programming a basic addition
  

Afternoon - CUDA Kernel performance (1/2)

• Using 2D CUDA grid for large computations
• Lab session: programming 2D CUDA grid
  computation
  
• CUDA warps
  
• Data alignment & coalescing
  
• Lab session: ensuring coalescing
  

Day 2

Morning – CUDA Kernel performance (2/2)

• Texture memory & constant memory
• Shared memory
  
• Lab session: use CUDA shared memory as a cache
  
• Maximizing occupancy
• Interpreting profiler counters
• CUDA performance tools: Visual Profiler, ...

Afternoon - HMPP Basics

• Introduction to parallel hybrid programming
  
• HMPP overview
  
• Lab session: HMPP Hello World
  
• Basis of HMPP programming
  
• Lab session: offloading a computation into a GPU
  
• HMPP compilation model
  
• Lab session: compiling an HMPP application
  

Morning – HMPP Transfers optimization

• Managing data transfers
• Lab session: programming data transfers
  
• Grouping GPU computations
  
• Optimizing data movement

Afternoon - Optimizing GPU code generation with HMPP

• Advanced kernel performance
  
• Driving the code generation and gridification
  
• Automatic loop transformations: unrolling, splitting, jamming, ...
  
• Lab session: optimizing Sgemm code generation

• Duration: 2 days
  
• Prerequisites: knowledge in C
• Training limited to 14 attendees

Download datasheet

In this 2 day training, participants will learn the fundamentals of parallel programming. An overview of OpenMP and MPI will be provided with hands-on practical labs where participants will be able to progressively understand the concepts of parallel programming.

• Parallel architectures and programming models (talk)
• OpenMP (talks & practicals)

• Introduction to MPI
• Point to point communication (+exercises)
• Non-blocking communication (+exercises)
• Collective communication (+exercises)
• Other MPI-1 and MPI-2 features (talk)
• Hybrid programming (talk)
• Tools & libraries: debugging, performance analysis, other libraries (short talk)

• Duration: 2 days
• Prerequisites: knowledge in C
• Training limited to 14 attendees

Download datasheet

In this 2 day CUDA training, participants will learn the CUDA programming model and tips to reach performance. The training includes hands-on practical labs where participants will be able to progressively program high performance computations.

Training content

Day 1

Morning - CUDA Basics

• Introduction to GPU computing
  
• CUDA architecture and programming model
• CUDA API
  
• CUDA debugging
  

Afternoon - CUDA kernel performance

• CUDA warps
• Data alignment & coalescing

Day 2

Morning - CUDA kernel performance

• Texture memory & constant memory
  
• Shared memory
  

Afternoon - CUDA grids & kernels optimization

• Maximizing occupancy
• Interpreting profiler counters
• CUDA performance tools: Viual Profiler, ...

• Duration: 3 days
• Prerequisites: knowledge in C, CUDA basis
• Training limited to 14 attendees

Download datasheet

In this 3 day CUDA training, participants will learn the CUDA programming & streaming model and how to handle multi-GPU applications. The training includes hands-on practical labs where participants will be able to progressively build streamed and multi-GPU applications.

Training content

Day 1

Morning - CUDA Basics

• Introduction to GPU computing
• CUDA architecture and programming model
• CUDA API
• CUDA debugging

Afternoon - CUDA Kernel performance

• CUDA warps
• Data alignment & coalescing

Day 2

Morning - CUDA Kernel performance

• Texture memory & constant memory
• Shared memory

Afternoon - CUDA Grids a kernels optimization

• Maximizing occupancy
• Interpreting profiler counters
• CUDA performance tools: Visual Profiler, ...

Day 3

Morning - CUDA streaming

• Asynchronism & overlapping
  
• CUDA streams
  

Afternoon - CUDA with MPI

• MPI introduction
  
• Mixing CUDA and MPI
  
• Multi-GPU computing with CUDA and MPI