| Project/Area Number |
23650012
|
|---|
| Research Category |
Grant-in-Aid for Challenging Exploratory Research
|
| Allocation Type | Multi-year Fund |
|---|
| Research Field |
Software
|
|---|
| Research Institution | Tokyo Institute of Technology |
|---|
Principal Investigator |
NUKADA AKIRA 東京工業大学, 学術国際情報センター, 特任准教授 (40545688)
|
|---|
| Project Period (FY) |
2011 – 2012
|
| Project Status |
Completed (Fiscal Year 2013)
|
| Budget Amount *help |
¥3,640,000 (Direct Cost: ¥2,800,000、Indirect Cost: ¥840,000)
Fiscal Year 2012: ¥2,210,000 (Direct Cost: ¥1,700,000、Indirect Cost: ¥510,000)
Fiscal Year 2011: ¥1,430,000 (Direct Cost: ¥1,100,000、Indirect Cost: ¥330,000)
|
|---|
| Keywords | 並列処理・分散処理 / ＧＰＧＰＵ / ＣＵＤＡ / ＦＦＴ / ＩｎｆｉｎｉＢａｎｄ / ＴＳＵＢＡＭＥ / GPGPU / CUDA / 自動チューニング / GPU間転送 |
|---|
| Research Abstract |
In 2006, NVIDIA presented CUDA as GPU computing environment for genericl computations, and after that many time-consuming applications are ported to GPU and achieved extreme speed-ups. For large-scale computation using supercomputers, efficient data transfer between GPUs is the most important. FFT is used in many scientific simulations and requires all-to-all communications between GPUs. We presented a novel scheduling techniques of PCI-Express data transfer between host and GPU, and InfiniBand data transfers between nodes. As a result, we could achieve scalable performance even if many other users' jobs are running on the shared network.
|
