The team behind The platform is the first step towards the company’s ambitious goal: making AI-assisted coding applicable to performance-critical codes and large-scale HPC applications. Currently, coding tools such as GitHub’s CoPilot and ChatGPT lack the necessary performance data about an application in their reasoning context. Daisytuner’s platform fills the missing gap, as it supports benchmarking across diverse architectures, including X86, ARM, GPUs, and the collection of a variety of relevant information such as profiler data and hardware utilization. As such, it will allow engineers to port high-performance applications much faster to a wide range of chips, including those developed by European companies such as Axelera AI.
‘Tools liks GitHub Copilot and ChatGPT don’t understand what happens when code hits real hardware. That’s where things break — memory bottlenecks, CPU stalls, low GPU occupancy — all invisible from just reading code. To tune performance, you need profilers and hardware metrics, and you need to iterate based on the data,’ explains Lukas, a HiPEAC member. ‘Integrating profiler and compiler output into AI coding copilots is essential for their success.’
The launch of the beta site follows a successful €1 million pre-seed funding round led by LEA Partners, which was joined by Angel Invest, Jens Lapinski, Christian Stiebner, Sara Schneider and Elias Schneider, Hans-Juergen Schmidtke, Jessica Holzbach and Mario Götze.
