2018 48th Annual IEEE/IFIP International Conference on Dependable Systems and Networks (DSN) 2018
DOI: 10.1109/dsn.2018.00043
|View full text |Cite
|
Sign up to set email alerts
|

A Framework for Evaluating Software on Reduced Margins Hardware

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
12
0

Year Published

2020
2020
2022
2022

Publication Types

Select...
6
1
1

Relationship

0
8

Authors

Journals

citations
Cited by 14 publications
(12 citation statements)
references
References 16 publications
0
12
0
Order By: Relevance
“…There are two approaches to undervolting studies: i) simulation-based studies [89,127,132,108], or ii) direct implementation or testing on real hardware fabrics, mainly performed on CPUs, GPUs, ASICs, and DRAMs [138,9,78,18,81,50]. The simulation-based approach requires less engineering effort.…”
Section: Undervolting: Supply Voltage Underscaling Below the Nominal Voltage Levelmentioning
confidence: 99%
See 1 more Smart Citation
“…There are two approaches to undervolting studies: i) simulation-based studies [89,127,132,108], or ii) direct implementation or testing on real hardware fabrics, mainly performed on CPUs, GPUs, ASICs, and DRAMs [138,9,78,18,81,50]. The simulation-based approach requires less engineering effort.…”
Section: Undervolting: Supply Voltage Underscaling Below the Nominal Voltage Levelmentioning
confidence: 99%
“…However, this approach lacks the information of real hardware, and thus, validation of results is the main concern. Most of the existing simulation-based studies are for CPUs [89,127,108,81] and specifically for CPU components such as caches [2,118,119,126,23] and branch predictors [20]. There are also studies for ASIC CNN accelerators [86,132,5].…”
Section: Undervoltingmentioning
confidence: 99%
“…Undervolting is also studied for other memory types like SRAM [67,68] and flash [4,5,6,7,8,9,10,36]. In addition to real chips, undervolting is studied at the simulationlevel, e.g., for CPUs [44,61], FPGAs [32], ASICs [49,69], and SRAMs [1,63,64].…”
Section: Related Workmentioning
confidence: 99%
“…Since we investigate the interplay between reducing CPU voltage margins and power capping mechanisms, it is necessary to quantify the voltage margins of the target platforms through a characterization process. We employ the (XM) 2 framework (discussed in Chapter 6) and apply a methodology similar to that in prior work [20,21,22,23,24], for the CPUs of Table 2.1, using SPEC CPU2006 [6], Parsec [16] and micro-virus [24] benchmarks. More specifically, for each CPU frequency point we run each benchmark 20 times at each voltage step, in the entire range from the nominal voltage point down to the lowest sub-nominal voltage, referred as V min , at which any workload executes successfully without any hardware reported error, silent data corruption (SDC), or system crash.…”
Section: Characterization Of Voltage Marginsmentioning
confidence: 99%