VSCSE Many-core Processors 2010 Google Group

Fernanda Foertter wants to chat

ffoert...@gmail.com (Fernanda Foertter) — Sat, 16 Jul 2011 03:17:36 UT

------------------------------ ------------------------------ -----------
Fernanda Foertter wants to stay in better touch using some of Google's
coolest new
products.
If you already have Gmail or Google Talk, visit:
[link]
You'll need to click this link to be able to chat with Fernanda Foertter.

Fernanda Foertter wants to chat

ffoert...@gmail.com (Fernanda Foertter) — Sat, 16 Jul 2011 03:17:16 UT

Re: [Many-core Processors] Large dense matrix inverse

nerang...@gmail.com (Suranga Edirisinghe) — Tue, 08 Mar 2011 02:24:06 UT

Thanks everyone for very useful comments.
I really appreciate your help.
Suranga

Re: [Many-core Processors] Large dense matrix inverse

joaan...@umich.edu (Joshua A. Anderson) — Mon, 07 Mar 2011 23:21:38 UT

Try MAGMA - [link] . It offers lapack type operations accelerated on the GPU.
--------
Joshua A. Anderson, Ph.D.
Chemical Engineering Department, University of Michigan

Large dense matrix inverse

nerang...@gmail.com (Suranga Edirisinghe) — Mon, 07 Mar 2011 22:45:48 UT

Hi,
I'm not sure whether this group is still active, but just giving a try.
I need to find inverse of large dense matrix (about 3000*3000 complex or
more). I have tried to use ATLAS (lapack and blas). But it is exceptionally
slow.
(I assume these codes are not cash optimized, and degrade performances, or

Re: [Many-core Processors] cuda code takes forever to compile and consume all system memory

john.a.strat...@gmail.com (John Stratton) — Tue, 04 Jan 2011 20:44:45 UT

Ah, yes, that makes sense.
When I say "declare a different kernel for each", what I mean is that your
application might just be too big to accelerate all in one kernel launch.
If you already had scratch space in global memory, it might be better to
break it up into many kernel launches. The reason most people don't do this

Re: [Many-core Processors] cuda code takes forever to compile and consume all system memory

songkai...@gmail.com (Kai Song) — Tue, 04 Jan 2011 19:53:14 UT

Hi John,
Thanks for the replay. Great information! By declaring "__local__" will
probably help the compiler a little bit, but as you pointed out the
potential issue with large kernel, I indeed put all the functions into one
single kernel. The application involves many 5x5 matrix/vector operations,
so there are a big number of functions are matrix/vector functions. Given

Re: [Many-core Processors] cuda code takes forever to compile and consume all system memory

john.a.strat...@gmail.com (John Stratton) — Tue, 04 Jan 2011 19:08:42 UT

Interesting situation. I haven't seen too many applications that try to
work with a data structure that large for every element. So yes, you'll
probably run into some less common issues.
My first comment is that registers are used for more than local variables.
In fact, some local variables may not take up an "extra" register at all.

cuda code takes forever to compile and consume all system memory

songkai...@gmail.com (Kai Song) — Tue, 04 Jan 2011 18:48:50 UT

Hi,
I am not sure if this mailing list is still active, but I have been stuck on
this problem for a long time, and maybe some of you have run into this
problem and can give me a clue.
My application has a big data structure, namely, 63692 Bytes. My intended
implementation is that each thread has a copy of this data structure with

Want more training? Apply for funding to attend PASI

lorena.ba...@gmail.com (Lorena Barba) — Wed, 29 Sep 2010 00:37:04 UT

Announcement:
==============

Categories: Training and Conferences, Education

Pan-American Advanced Studies Institute Scientific Computing in the
Americas: The Challenge of Massive Parallelism
(3-14 Jan. 2011)

[link]

Thanks to NSF and DOE funding, there is travel support for up to 30

Re: [Many-core Processors] View this page "Course Presentations"

trsc...@gmail.com (Tom Scavo) — Tue, 24 Aug 2010 21:26:21 UT

I think that's fine, Andy.
Tom

RE: [Many-core Processors] View this page "Course Presentations"

asc...@illinois.edu (Schuh, Andrew) — Tue, 24 Aug 2010 21:23:04 UT

Can will you have a problem if I create public links to the presentations and videos on gpucomputing.net?
___________________
Andrew E. Schuh
University of Illinois
Cell: 217-898-8348
-----Original Message-----
Sent: Tuesday, August 24, 2010 4:10 PM
To: VSCSE Many-core Processors 2010
The presentation video recordings for Many-core Processors have been

View this page "Course Presentations"

trsc...@gmail.com (Tom Scavo) — Tue, 24 Aug 2010 21:09:32 UT

The presentation video recordings for Many-core Processors have been
uploaded and linked.

Click on [link]
- or copy & paste it into your browser's address bar if that doesn't
work.

Re: Are our accounts on ac.ncsa.uiuc.edu suspended?

zxue.pur...@gmail.com (Jerry Z Xue) — Sun, 15 Aug 2010 22:46:06 UT

Great. Thanks a lot for the info.

Jerry

Re: [Many-core Processors] Are our accounts on ac.ncsa.uiuc.edu suspended?

trsc...@gmail.com (Tom Scavo) — Sun, 15 Aug 2010 21:40:36 UT

I just found out what's going with the AC compute system. There is a
cooling outage in the building that houses AC, which has forced a
shutdown of the system. The sysadmins are hoping to have AC restored
by noon on Monday. This should give you enough time to grab your files
before the accounts are deactivated at the end of the day.