Problem in parsing DRAM energy
AsicDyc opened this issue · 0 comments
AsicDyc commented
Hi,
Thank you for providing such a useful tool. I was confused about the output about DRAM energy in timeloop-mapper.stats.txt
when using timeloop to analyze a three-level storage hierarchy (DRAM, Global Buffer and register file).
The energy for DRAM updates, writes, and reads listed in the ERT is 512
, and the DRAM vector access energy
listed in the SPECS
section in timeloop-mapper.stats.txt
is also 512
. However, the per-scalar-access energy
for DRAM used in the calculations in the STATS
section of the output file is 64
. Why are these three energy values inconsistent? The type of DRAM is LPDDR4
.
Here is ERT for DRAM:
actions:
- name: update
arguments:
global_cycle_seconds: 1e-09
action_latency_cycles: 1
energy: 512.0
- name: write
arguments:
global_cycle_seconds: 1e-09
action_latency_cycles: 1
energy: 512.0
- name: leak
arguments:
global_cycle_seconds: 1e-09
action_latency_cycles: 1
energy: 0.0
- name: read
arguments:
global_cycle_seconds: 1e-09
action_latency_cycles: 1
energy: 512.0
Here is SPEC
part and STATS
part in timeloop-mapper.stats.txt
SPECS
-----
Technology : DRAM
Size : -
Word bits : 8
Block size : 8
Cluster size : 1
Instances : 1 (1*1)
Shared bandwidth : -
Read bandwidth : 8589935000.00
Write bandwidth : 8589935000.00
Multiple buffering : 1.00
Effective size : -
Min utilization : 0.00
Vector access energy : 512.00 pJ
Vector access energy source : ERT
Per-instance-cycle leakage : 0.00 pJ
Instances sharing power gating : 1.00
Non-power-gated utilization : 100.00%
Leakage energy (total) : 0.00 pJ
Area : 0.00 um^2
Weights:
Partition size : 1536
Utilized capacity : 1536
Utilized instances (max) : 1
Utilized clusters (max) : 1
Scalar reads (per-instance) : 98304
Scalar fills (per-instance) : 0
Scalar updates (per-instance) : 0
Temporal reductions (per-instance) : 0
Address generations (per-cluster) : 98304
Energy (per-scalar-access) : 64.00 pJ
Energy (per-instance) : 6291456.00 pJ
Energy (total) : 6291456.00 pJ
Temporal Reduction Energy (per-instance) : 0.00 pJ
Temporal Reduction Energy (total) : 0.00 pJ
Address Generation Energy (per-cluster) : 0.00 pJ
Address Generation Energy (total) : 0.00 pJ
Bandwidth Consumption Scale : 1.00
Shared Bandwidth (per-instance) : 1.00 words/cycle
Shared Bandwidth (total) : 1.00 words/cycle
Read Bandwidth (per-instance) : 1.00 words/cycle
Read Bandwidth (total) : 1.00 words/cycle
Write Bandwidth (per-instance) : 0.00 words/cycle
Write Bandwidth (total) : 0.00 words/cycle