This will be a short “trip report” on the talks I watched over the week of the virtual GTC 2023 (March 20 - 24). All talks are freely available online. You can read my GTC trip report from 2022 here.
Summary
| Meaning | |
|---|---|
| 🌟 | Keynote |
| ⭐ | Best talks |
 |
Available on YouTube |
GTC Keynote
Speaker: Jensen Huang, CEO of NVIDIA
How to Write a CUDA Program
Speaker: Stephen Jones
This was a great talk, especially if you are new to CUDA. It walks you through a building a simple CUDA program in steps:
- Downloading CUDA and the CUDA toolkit
- Writing a “hello world” GPU kernel
- Determining the number of blocks and threads for your kernel (image below)
- When in doubt, use a block size of 256
- Corollary: avoid block sizes smaller than 128 if you can
- Using NVIDIA Nsight Systems for profiling and analysis
⭐ CUDA: New Features and Beyond (2023)
Speaker: Stephen Jones
This is a summary of the latest and greatest things in CUDA and the CUDA ecosystem. Highlights include:
- Overview of cuNumeric and Legate
- Overview of CUTLASS 3.0 and CuTe
- Overview of NVIDIA GPU Computing Tools
and much more!
C++ Standard Parallelism
Speaker: Bryce Adelstein Lelbach
This talk is largely the same as Bryce’s GTC 2022 with the same name. However, it has been updated, as shown by the slide below. ⬇️
Hope to See You Next Year!
This was my fourth virtual GTC I have “attended” since joining NVIDIA back in 2019. I’ve got my fingers crossed that in 2024 we will be able to attend in person and I can meet some of you there!