Tuesday, 27 September 2011

Intel to Banish Ninja programmers forever?

While detailing Intel's multi-core future, Rattner said "You think you'd have to be some kind of freak to program for multi core, but you don't have to be a ninja programmer" and then he also said, "our goal at Intel is to banish ninja programmers forever."

Ian Phillips of ARM, on Sep 5th, in our Multi-core Challenge workshop at Bristol said, "There are problems that need scientists to solve. And there are problems that don't" implying that multi-core needs sophisticated programmers. That's more pragmatic than Rattner's talk. Isn't it?

Read more here Link to the article

Has the industry already addressed some of the multi-core problems found in 2006?

Gordon Haff in his blog post What became of multi-core programming problems? says that some of the gaming software have managed to really tax the desktop hardware by utilising some of the modern architectures.
I can say that Gordon is one of the rare talent pool.
I will try to get him for a webinar.

NVIDIA white paper on variable SMP

Have you read about NVIDIA's Project Kal-El processor? It implements a novel variable Symmetric Multiprocessing technology.

Essentially, there are five processors in this architecture. Four are designed to work at higher frequencies and the fifth at a much lower frequency. The low frequency chip will work on rendering multimedia. Based on the workload, these chips are enabled and disabled. The fifth CPU is called a companion which is not OS sensitive.

My question would be who would decide the enabling and disabling strategy? And at what level? If multiple programs act on enable/disable settings, would there be a middleware that takes sensible, real time decision?

What's it all about? Would it save energy without compromising on performance?
Read the whitepaper below for more information
SMP: A Multicore CPU Architecture for Low Power and High Performance

A Conversation with Intel’s John Hengeveld at IDF 2011

Here is a link to Greg Pfister's website containing transcript of his conversation with John Hengeveld of Intel.

The discussion spans several themes ... Use of MIC for streaming, Intel attributing performance to its programming model, MIC versus SoC/SCC, would it support CUDA and OpenCL.

An interesting read