A premier, deeply granular performance analysis tool. Computer scientists use it to optimize thread execution, memory access, and vectorization to ensure their code is extracting maximum theoretical performance from the underlying CPU/GPU architecture.

HPCToolkit analyzes program performance to pinpoint exact bottlenecks (e.g., CPU idle time, cache misses, memory overhead) across thousands of nodes without heavily slowing down the application during testing.

A deep learning optimization library that makes distributed training of massive AI models (like Large Language Models) possible. It manages memory and pipeline parallelism across thousands of GPUs on an HPC, solving the problem of models being too large to fit in a single machine's VRAM.

A container solution specifically designed for HPC systems. Unlike Docker, it does not require root privileges (enhancing security) and integrates seamlessly with underlying hardware (GPUs, Infiniband).

A state-of-the-art package manager built specifically for HPC. It allows researchers to install multiple software versions with different compilers (e.g., GCC, Intel) without them conflicting with one another.

One of the most popular open-source resource managers and job schedulers. It is responsible for receiving user "jobs," queuing them, and allocating the appropriate compute nodes for their execution.