2018 Fiscal Year Final Research Report
Reduced-precision formats for high-performance and energy-efficient computations
| Project/Area Number |
16K16062
|
|---|
| Research Category |
Grant-in-Aid for Young Scientists (B)
|
| Allocation Type | Multi-year Fund |
|---|
| Research Field |
High performance computing
|
|---|
| Research Institution | Tokyo Woman's Christian University (2017-2018)
Institute of Physical and Chemical Research (2016) |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2016-04-01 – 2019-03-31
|
| Keywords | Reduced-precision / Mixed-precision / GPU |
|---|
| Outline of Final Research Achievements |
This study explored the possibility of reduced-precision formats which have shorter bit length against the IEEE 32/64 bit floating-point for enchance the performance of numerical computations in terms of both computation speed and energy efficiency. We proposed a light-weight implementation of reduced-precision formats on software and demonstrated the performance improvement, in terms of both speed and energy efficiency, on some data-intensive operations on basic linear algebra.
|
| Free Research Field |
高性能計算
|
| Academic Significance and Societal Importance of the Research Achievements |
浮動小数点表現を工夫することによる計算の高性能化は，さまざまな計算処理に応用できる汎用的なアプローチである．本研究の開始時と比べると，現在では本研究と類似のアプローチを試みた研究がいくつか見られるようになり，本研究の着想や議論には意義があったと考えられる．また，本研究期間内には達成できなかった混合精度計算手法への適用や，近年発展が著しいFPGAへの適用が今後期待される．
|
