The term “performance portability” is often used loosely, but the “Processor Trends and What They Mean for Software” panel at HPSFCon 2025 in Chicago offered a grounded perspective. The discussion explored what it truly takes to develop software that performs consistently across platforms without overburdening engineering teams.
Spoiler: it’s complicated, expensive, and a major reason HPSF exists—to turn performance portability from a buzzword into a practical engineering goal.
Big thanks to our moderator: Christian Trott, Sandia National Laboratories; and panelists: Barton Fiske, NVIDIA; Brian Cornille, AMD; Michael James, Cerebras; Patrick Fasano, NextSilicon; Kristi Belcher, Lawrence Livermore National Laboratory.
Watch the full video here.
1. Performance portability
Writing portable and fast software takes work. HPSF is helping reduce that load through open collaboration between researchers, developers, and vendors, all under one roof.
Takeaway: Portability is worth it, but only if there’s support. HPSF is building that support.
2. Models matter
Choosing the right programming model makes or breaks your performance. HPSF’s Programming Models working group helps vet and guide development of models that don’t just promise performance, but deliver it across systems.
Takeaway: Pick a model that fits your team, your app, and your performance goals.
3. Collaboration across teams is critical
This was a recurring theme: performance portability isn’t one team’s job. HPSF was built on that principle through bridging software, hardware, tools, and standards communities.
Takeaway: Real performance takes teamwork. HPSF exists to make that coordination easier.
4. Toolchains are catching up (but slowly)
You’ll spend more time than you want wrangling tools. Profilers, debuggers, and compilers still don’t always play nice with multi-platform code.
Several of HPSF’s projects, like E4S and its compiler toolchain support efforts, are aimed at making this less painful for everyone.
Takeaway: Bake in extra time for tooling headaches, and advocate for better support early.
Conclusion
This panel echoed what HPSF already knows: software and hardware co-design needs serious coordination. By creating a shared space for that work, HPSF is helping the ecosystem move from duct tape to deployment-ready.
Want to get involved with HPSF? Whether you’re building software or shaping standards, there’s a place for you in the High Performance Software Foundation community. Check out working groups, contribute to open projects, or join us at future events to help push the ecosystem forward.