| Presentation | 2008/7/29 ソフトウェアECCによるGPUメモリの耐故障性の実現と評価(信頼性とセキュリティ,SWoPP佐賀2008-2008年並列/分散/協調処理に関する『佐賀』サマー・ワークショップ) Naoya MARUYAMA, Satoshi MATSUOKA, Yasuhiko OGATA, Akira NUKADA, Toshio ENDO, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | General-Purpose Processing on GPUs (GPGPUs) has rapidly been recoginized as a promissing HPC technology because of GPUs' much higher peak floating-point processing power. However, GPUs have originally been developed for graphics applications, such as 3D games, where reliability is not considered as an important issue as in HPC communities. One notable example is the lack of ECC in graphics memory systems. To improve the reliability of GPUs for HPC applications, we propose a software-based technique to generate and check ECC for graphics memory. Our library-based approache allows for CUDA-based GPGPU applications to be easily extended with ECC-based error checking with little manual intervention. To evaluate the applicability of our approach, we extended two CUDA applications with our ECC libarary: a matrix multiplication and an N-body problem. Our performance studies showed that while matrix multiplication can take up to 300% overhead, the N-body application only incurrs 15% of overhead. These results suggest that software-based ECC would be a promissing approach for computation-intensive applications such as N-body problems. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | GPGPU / dependability / ECC |
| Paper # | DC2008-20 |
| Date of Issue |
| Conference Information | |
| Committee | DC |
|---|---|
| Conference Date | 2008/7/29(1days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | |
| Chair | |
| Vice Chair | |
| Secretary | |
| Assistant | |
| Paper Information | |
| Registration To | Dependable Computing (DC) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | |
| Sub Title (in English) | |
| Keyword(1) | GPGPU |
| Keyword(2) | dependability |
| Keyword(3) | ECC |
| 1st Author's Name | Naoya MARUYAMA |
| 1st Author's Affiliation | Tokyo Institute of Technology:Japan Science and Technology Agency, CREST() |
| 2nd Author's Name | Satoshi MATSUOKA |
| 2nd Author's Affiliation | Tokyo Institute of Technology:National Institute of Informatics:Japan Science and Technology Agency, CREST |
| 3rd Author's Name | Yasuhiko OGATA |
| 3rd Author's Affiliation | Tokyo Institute of Technology:Japan Science and Technology Agency, CREST |
| 4th Author's Name | Akira NUKADA |
| 4th Author's Affiliation | Tokyo Institute of Technology:Japan Science and Technology Agency, CREST |
| 5th Author's Name | Toshio ENDO |
| 5th Author's Affiliation | Tokyo Institute of Technology:Japan Science and Technology Agency, CREST |
| Date | 2008/7/29 |
| Paper # | DC2008-20 |
| Volume (vol) | vol.108 |
| Number (no) | 181 |
| Page | pp.pp.- |
| #Pages | 7 |
| Date of Issue | |