CHANNELS
 
 
 
 
 
 
 
 
ON THE WEB
 
 
 
 
PRINT EDITION
 
 
 
 
BZ MEDIA
 
 
 
 
ADVERTISER LINKS
 
 
 
 
 
LOADING...
LOADING...
 
 
AS OF 8/21/2008 8:01PM EST
RapidMind Advances Multicore Support
Offers platform abstraction to speed adoption
By P. J. Connolly

June 15, 2007 — For years, processor vendors have improved performance by increasing the clock speed; that well having run dry, they are now cranking up processors by increasing the number of cores in a single package. But programming for multiple cores raises the question of how to allocate a workload across such a hardware platform without tying oneself so closely to the silicon that portability is impossible.

One company that thinks it has an answer is RapidMind, which announced in May the general availability of a new version of its namesake development platform for multicore and stream processors. The update adds support for IBM’s Cell BE processor.

RapidMind allows developers to write in standard C++ with existing tools. The software distributes the workload across the various cores in a way that wrings the most performance out of the hardware. This is especially important in high-performance computing applications, such as financial modeling, seismic analysis, and image and signal processing.

Ray DePaul, RapidMind president and CEO, explained that before multicore, software developers didn’t have to do much to take advantage of hardware improvements. “The challenge with multicore is that the software developers and ISVs actually have to do something to take advantage of it. The options today are quite difficult and error-prone.”

DePaul noted that the chipmakers are experiencing Moore’s Law, but on a new front: “They’re doubling cores at the rate they used to double clock speed, and there’s this gap in the software ecosystem’s ability to take advantage of that. That’s the gap we fill.”

The RapidMind platform incorporates a code optimizer that attempts to reduce overhead by analyzing computations for instructions that don’t need to be performed one step after another. A load manager schedules the work from one core to the next, while a data manager sorts out the flow of information and logging diagnostic tools report performance bottlenecks and other mishaps. Processor-specific support modules allow the RapidMind platform to communicate with the underlying hardware, while at the top of the stack top, the RapidMind API supports Linux and Windows applications.

DePaul argued that adding an abstraction layer makes developers’ lives simpler in a multicore world: “If you are fortunate enough to get [multicore applications] working, the next time the hardware revs, you have to go back to the drawing board. Our platform scales arbitrarily, to the number of cores.”

Today, the RapidMind platform works with the ATI x1X00 line of graphics cards, as well as NVIDIA’s Quadro card and the GeForce 6000, 7000 and 8000 series of cards. It also supports IBM’s 9-core Cell BE, in the company’s QS20/30 blade and in the Sony PlayStation 3. PlayStation users can run Yellow Dog Linux 5 on their PS3s; users on other platforms will have to be satisfied with Fedora Core 4, Red Hat Enterprise Linux 4, Ubuntu 6.10, Windows 2000 and later versions of those operating systems. RapidMind expects to announce support for AMD and Intel processors at a later date.
 


 
 
 
 
 
 
 
 
 
 
SUBSCRIBE TODAY!
 E-Newsletters:
  News on Mon/Thurs.
  Test & QA Report
  EclipseSource
 
 
JOB BOARD
 
PDF & PRINT EDITION
* Requires Resource Account!  LOGIN or SIGN UP

Download Current Issue!
ISSUE 8/15/2008 PDF

Need Back Issues?
DOWNLOAD HERE

Receive The Print Edition?
SUBSCRIBE HERE
 
REGISTER
 
GET NOTIFIED!
About all of the latest Resources
 
 
SD TIMES 100
It's time once again to
recognize the organizations
or individuals that have
demonstrated leadership in
their markets.