U.S. National Security Community Needs to Address Major Speed Bump Ahead
ASCI will achieve its objectives…
- 600 TeraOP/s by 2004
- Full system simulation code by 2005-6
- Energize industry to build systems beyond Moore’s Law growth curve for 10+ years
-
-
ASCI is at its limit for acceleration
- 4X non-ASCI sites is not sustainable
- If the hardware don’t break, the software complexity will kill you…
-
-
Beyond 4X, we have a problem
- Too expensive
- Too much power, space
- SW won’t scale
- HW limitations
- Vendor choices decreasing
Quote from Ken Kennedy’s PITAC summary for PetaFLOP/s
“Suppliers of high-end systems suffer from unusual market pressures...”