LOGIN | REGISTER NOW | SUBSCRIBE
 
 
News on Monday
more>>
SharePoint Tech Report
more>>


   

 
 
Download Current Issue
ISSUE 7/1/2009 PDF

Need Back Issues?
DOWNLOAD HERE

Receive the print Edition?


 
A knockout blow for Borland?
MicroFocus has upped its offer for Borland Software to $1.50, hoping to chase off a mystery suitor also pursuing the ALM vendor.
07/06/2009 12:26 PM EST

Is the mystery Borland suitor Serena?
Borland software is considering an offer from another company after a preliminary deal with MicroFocus. Is Serena the new company?
06/30/2009 01:55 PM EST

Windows 7 - An eBayer's dream product?
Windows 7 pre-orders can make people money on eBay.
06/29/2009 03:48 PM EST

 

Microsoft Worldwide Partner Conf.
7/13/2009 to 7/16/2009
New Orleans
Microsoft

OSCON (Open Source Convention)
7/20/2009 to 7/24/2009
San Jose
O'Reilly Media

XBRL Technology Workshop & Summit
7/28/2009 to 7/30/2009
Santa Clara
XBRL US

ACM SIGGRAPH
8/3/2009 to 8/7/2009
New Orleans
ACM SIGGRAPH

OpenSource World (formerly LinuxWorld)
8/12/2009 to 8/13/2009
San Francisco
IDG World Expo


 
print Printable version 
Most Read Latest News Blog Resources
Kapow automates website data retrieval
Kapow Technology's Web Data Server 7.0 can be used to create modules that navigate through...

Telerik looks to boost 3D for Silverlight
In a new release of its RadControls component suite, the company introduced a 3D charting ...

Oracle's Fusion 11g lays groundwork for ERP platform
Fusion's components are standards-based and integrated, but some abilities are lost if non...

Digg!  Digg
Reddit  Reddit
Slashdot  Slashdot
Share on facebook  Facebook
Share on Twitter  Twitter

PeakStream Is Extraordinarily Serious Runtime for HPC


By Edward J. Correia
October 1, 2006 — 
PeakStream is not about playing games. But the new company has its eye on IBM’s Cell processor and high-speed graphics processor units, even though those circuits are being used today mainly to run entertainment software.

Former executives of Sun Microsystems, VMware and Nvidia, the graphics accelerator chip maker, have joined to form PeakStream Inc. Flush with US$17 million in Series B funding, the company in September unveiled the PeakStream Platform, a runtime solution for high-performance computing that it claims abstracts the complexities and pitfalls of programming for multiple processors or cores while improving application performance by more than twentyfold.

The all-software solution includes the PeakStream Server, which installs on top of an Intel-based machine with an optional board containing one or more graphics processor units (GPUs). “This is substantial technology that allows you to run an application on top of a graphics processor,” said Matt Papakipos, PeakStream’s CTO and former architecture lead with Nvidia.

An evaluation version for Linux was set to be released on Sept 18; general availability is scheduled before the end of this year. Pricing is set at $2,000 per server.

Papakipos said that to take advantage of the PeakStream APIs, a developer need only include the appropriate PeakStream libraries in new or existing C/C++ apps. “A virtual machine provides an abstraction layer on multiple types of CPUs and GPUs so code can execute anywhere,” he said. Initial VMs will support Cell, multicore 32- and 64-bit Xeon and ATI GPUs (now owned by AMD). “This is the magic of the technology; apps can run anywhere with no rewrite,” Papakipos claimed, adding that the VM handles matrix math, core math, signal processing and user-developed intrinsics.

After reviewing a PeakStream white paper, Steven Wallach, a supercomputer designer and consultant to Los Alamos National Laboratory in New Mexico, said he believes the technology has potential. “For a certain limited set of applications, they could probably see significant increases,” in performance, he said, provided they’re 32-bit apps performing lots of floating-point operations. “GPUs are 32-bit, single-precision processors,” he said.

There is a degree of work involved, but far less than would be without the VM, Papakipos said, and with greatly reduced risk. “The process of taking an array of data and [manually] assigning it to multiple cores is program-directed,” he said, and by design creates the need to lock one processor while another finishes its work. “PeakStream is different. We steer clear of the issues that cause multicore programs to hang.” Arrays are treated as first-class objects and are completely atomic, he explained. “Each operation is completed before the next one begins. The programmer is not assigning work to processors, so it’s impossible to get those race and deadlock conditions.”

ALL APPS MUST CHANGE
PeakStream introduces a technique it calls stream programming, in which program parallelism is expressed in arrays. “Arrays match the character of high-performance computing because [HPC apps] tend to be data-intensive and have many mathematical operations,” said Papakipos. The runtime handles work scheduling and read/write operations with non-uniform graphics and system memory. Latencies associated with movement of data from system to graphics memory are improved by a shared cache.

“We’ve also included the ability to do explicit reads and writes to graphics memory” or to main memory, said Michael Mullany, PeakStream’s vice president of marketing. “That’s an advantage because it gives the programmer more control of what is being moved to memory and when,” he said, adding that specific computations can be earmarked for processing where they will be handled most quickly and efficiently. “We recommend that arrays [contain] at least 10,000 elements for optimal performance,” he added. Unmodified applications will run, but will not be accelerated.

Based on technology developed at Stanford, stream and array-oriented languages have been in use in HPC and academia for decades. “But we’re the first commercial stream programming platform, and the model we are introducing is for doing high-performance computing.”


Share this link: http://www.sdtimes.com/link/29632
 

Add comment


Name*
Email*  
Country     


  • Comment
  • Preview
Loading