New Opportunities for Computer Architecture Research
Download
Report
Transcript New Opportunities for Computer Architecture Research
Challenges and Opportunities
for System SW in the CMP Era
Chuck Moore
AMD Senior Fellow
June 2007
WIOSCA Workshop Panel
Unexpected lessons learned during
the past decade
Strength of the compatibility requirement
Change occurs much more slowly than people want to believe
x86 & IBM S/390 longevity have lapped most RISC architectures
Our history shows:
Viable products don’t bet on wildly incompatible solutions
Actual results weakest link in the HW/SW symbiotic relationship
The economic realities of our industry can be painful
The best technology rarely wins
Simplicity usually trumps complexity Time-to-Market
Power issues emerged as the one of the most limiting
constraints of modern chip design
06/09/07
WIOSCA Panel - ISCA 2007
2
Some Specific Challenges
Dealing w/ the implicit serial “structural
compatibility” requirement of most of today’s
applications
Optimization balance between single-thread
performance and throughput performance
Building reliable platforms from increasingly
unreliable technology
Control mechanisms to appropriately manage
non-traditional resources
06/09/07
Power, Heat, Maintenance, Errors, and more coming …
WIOSCA Panel - ISCA 2007
3
Some Specific Opportunities
System-level ISA
Coprocessor enablement
“It’s the synchronization, stupid ”
Abstraction and compatibility
Think page ops, table lookups, transforms, etc
Envision the “7-layer ISO model for computing ”
Scarcity versus abundance
06/09/07
Abundant:
Scarce:
transistors, designers, raw FLOPs
pwr-eff, bw-eff, TTM, flexibility,
programmer productivity
WIOSCA Panel - ISCA 2007
4