Watch & Discuss: Can Standard C++ Replace CUDA for GPU Acceleration?

Join us for an evening of modern C++ learning and networking! This month, we’ll be watching Can Standard C++ Replace CUDA for GPU Acceleration? Attendees will be able to ask brief questions during the talk and longer ones in the discussion after.

The Talk:

• The Evolution of GPU Memory: A technical overview of how memory management has moved from tedious manual copies to seamless Heterogeneous Memory Management (HMM) [23:10].
• Standard Parallelism: How the `std::execution::par_unseq` policy allows standard library algorithms to run directly on GPUs with minimal code changes [24:54].
• Practical Pitfalls: A deep dive into the limitations of standard C++, such as the lack of support for exceptions or virtual functions in GPU code, and common performance traps like "first touch" memory issues [37:06].
• Portability vs. Performance: An analysis of current compiler support across NVIDIA and AMD ecosystems, concluding that while writing accelerated code is easier than ever, developers must still understand underlying hardware constraints to achieve peak performance [44:22].

Agenda:

• 6:30 PM: Doors Open & Networking
• 7:00 PM: Video Screening starts
• 8:00 PM: Group Discussion & Networking

Location & Instructions:
We are meeting at the Business Center of Lenox Burnet. Here is a map showing where to park and how to get to the entrance.. Please RSVP to ensure space is available.

• Entrance: Look for the glass door facing Burnet Road.
• Directions: The Business Center is located to the right of the leasing office.
• Access: If the door is locked or you don't see us, please post a comment on this Meetup page, and someone will come to let you in.

See you there!