Hacker News new | past | comments | ask | show | jobs | submit login




It looks difficult to code for (although the paper mentions they have their own parallelizing compiler & have successfully compiled several real applications).


3.0624 Tflop/s per node - that is impressive. And then you can fix 1024 nodes per cabinet for a peak rating of >3 Pflop/s per cabinet.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: