As a symbolic AI scientist I am really interested in making graph read/writes faster for fighting the complexity curve. The graal seems to be graph processors

As a symbolic AI scientist I am really interested in making graph read/writes faster for fighting the complexity curve. The graal seems to be graph processors (ASICS) e.g this Darpa project that promise a x1000 vs current x86. https://www.tomshardware.com/news/darpa-intel-qualcomm-graph-analytics,34665.html https://www.tomshardware.com/news/darpa-intel-qualcomm-graph... But I never thought of using à DSL (excluding openCL that I consider far less domain specific) and that is an excellent Idea, the benchmarks are promising ! But I Wonder How does this compare to those algorithms run on a modern GPU (or even with HSA)