Table 8.
Instruction | Description | Energy of | Energy of | Energy of | Total Energy |
---|---|---|---|---|---|
computation | each NPE | loop buffer | per instruction | ||
ADD/SUB/MUL | FP16 Arithmetic ops. | 0.5 | 1.3 | 0.9 | 1.4 |
2xINT8b Arithmetic ops. | 0.3 | 1.1 | 0.9 | 1.2 | |
GTH/MAX/MIN | FP16 Compare ops. | 0.3 | 1.1 | 0.9 | 1.2 |
EQL/ABS | 0.2 | 1.0 | 0.9 | 1.1 | |
AND/ORR | 16b Bit-wise ops. | 0.2 | 1.0 | 0.9 | 1.1 |
SHL/SHR | 0.3 | 1.1 | 0.9 | 1.2 | |
I2F | Data type cnv. | 0.3 | 1.0 | 0.9 | 1.1 |
RND | 0.6 | 1.3 | 0.9 | 1.4 | |
Instruction | Description | Energy of | Energy of | Energy of | Total energy |
Data-Mem-16b | each NPE | loop buffer | per instruction | ||
MLD | 16b Data Mem load | 2.9 | 0.6 | 1.6 | 3.7 |
MST | 16b Data Mem store | 3.5 | 0.2 | 1.6 | 3.9 |
Instruction | Description | Energy of | Energy of | Energy of | Total energy |
event generator | each NPE | loop buffer | per instruction | ||
EVC | Event capture | 0.6 | 0.4 | 0.5 | 0.5 |
+ per generated event | +1.1 | +0 | +0 | +1.1 | |
Instruction | Description | Energy of | Energy of | Energy of | Total Energy |
RV+Peripheries | Inst-Mem-32b | Data-Mem-32b | per instruction | ||
RISC-V Ops | Averaged per instruction | 5.9 | 5.7 | 0 | 11.6 |
+ Data Mem access | +10 | +10 | |||
NOC | Per 32b event transmission | – | – | – | 2 |
The energy of the computation is part of the NPE's energy, consumed by the involved compute logic inside the NPE. All the numbers are in pJ.