-
Notifications
You must be signed in to change notification settings - Fork 0
/
dstedc.html
87 lines (87 loc) · 3.73 KB
/
dstedc.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
<!DOCTYPE html>
<html>
<head>
<meta name="viewport" content="width=device-width" />
<link rel="shortcut icon" href="favicon.ico" />
<script type="text/javascript" async src="fonts-min.js"></script>
<link rel="stylesheet" href="style.css" />
<meta charset="utf-8">
<meta name="description" content="We present a memory-efficient
implementation of divide-and-conquer eigenvalue algorithm. It features
automatic profiling that dynamicallly partitions work according to machine
configuration." />
<title>
A memory-efficient algorithm for large-scale symmetric tridiagonal
eigenvalue problem on multi-GPU systems
</title>
</head>
<body>
<div id="container" itemscope
itemtype="http://www.schema.org/ScholarlyArticle">
<p>[<a href=".">← Go back to profile</a>]</p>
<h1 itemprop="name">
A memory-efficient algorithm for large-scale symmetric tridiagonal
eigenvalue problem on multi-GPU systems
</h1>
<p>
with
<a href="http://turing.cs.trincoll.edu/~pyoon/">Peter Yoon</a>
</p>
<p>
Paper presented at
<a href="http://worldacademyofscience.org/worldcomp14/ws/conferences/pdpta14">
WorldComp 2014: PDPTA</a> (2014)<br>
Poster presented at the
<a href="http://www.gputechconf.com/page/home.html">
GPU Technology Conference</a> (2014)
</p>
<h2>Download</h2>
<ul>
<li>Code: <a href="https://github.com/hcho3/dstedc_mgpu">
https://github.com/hcho3/dstedc_mgpu</a></li>
<li>GTC poster: [<a href="posters/dstedc_gtc.pdf">PDF</a>]</li>
<li>WorldComp PDPTA talk:
[<a href="talks/dstedc_talk.pptx">PPTX</a>]
[<a href="talks/dstedc_talk.pdf">PDF</a>]
</li>
<li>WorldComp PDPTA paper:
[<a href="preprints/dstedc_worldcomp.pdf">Paper</a>]</li>
</ul>
<h2>Synopsis</h2>
<p itemprop="description">
<strong>Divide-and-conquer algorithm</strong> is a numerically stable
and efficient algorithm that computes the eigenvalues and eigenvectors
of a symmetric tridiagonal matrix. We often face the situation where
the input matrix fits into the main memory but not into the on-chip
memory of a GPU device. We present an <strong>out-of-core
implementation</strong> where only part of the input matrix is
resident in GPU memory at any point in time. It works independently of
the physical size of GPU memory, handling any size of input as long as
it fits into the main memory. Work is dynamically allocated to
multiple GPUs and CPU cores, taking account of available workspaces
and progress of the algorithm. In addition, it delivers a performance
comparable to that of conventional multi-GPU implementations for cases
where workspaces fit into the GPU memory.
</p>
<h2>Publication Details</h2>
<ul>
<li>
Conference Paper:<br>
Hyunsu Cho and Peter Yoon. “A Memory-Efficient Algorithm for
Large-Scale Symmetric Tridiagonal Eigenvalue Problem on Multi-GPU
Systems,” <em>Proceedings of the 2014 International
Conference on Parallel and Distributed Processing Techniques
and Applications</em>, pp. 568-573, Las Vegas, NV, July 24, 2014.
</li>
<li>
Poster:<br>
Hyunsu Cho and Peter Yoon. “Symmetric Tridiagonal Eigenvalue
Problem on Multi-GPU Systems,”
<em>The GPU Technology Conference 2014</em>, San Jose, CA,
March 24, 2014.
</li>
</ul>
<p>[<a href=".">← Go back to profile</a>]</p>
</div>
</body>
</html>