Only Buffer When You Need To: Reducing On-chip GPU Traffic with Reconfigurable Local Atomic Buffers.
Preyesh DalmiaRohan MahapatraMatthew D. SinclairPublished in: HPCA (2022)
Keyphrases
- low cost
- buffer size
- real time
- power reduction
- heterogeneous computing
- high speed
- power saving
- high density
- general purpose
- low power
- power consumption
- network on chip
- network traffic
- loss probability
- parallel processing
- multithreading
- production system
- buffer management
- gpu implementation
- reconfigurable hardware
- circuit design
- physical design
- data transfer
- floating point
- traffic flow
- parallel implementation
- hardware implementation
- graphics processing units
- reconfigurable architecture
- systolic array
- memory bandwidth
- road network
- blocking probability
- parallel computing
- data center
- traffic congestion
- graphics hardware