HSA is supposed to be better than Nvidia’s CUDA and Intel’s Xeon Phi altogether

Jul 11, 2012 15:01 GMT  ·  By

Many of our readers heard about HSA and probably already found out about the AMD-funded HSA Foundation, but not that many really know what it is all about and what the main differences from Nvidia or Intel’s approach are.

The news unveiled at AMD’s AFDS 2012 event was that it, along with Imagination Technologies, ARM, MediaTek and Texas Instruments have founded the Heterogeneous System Architecture Foundation with initial funding from AMD.

The main goal of the foundation is to work on, develop and standardize all future HSA technologies.

Currently, there are two main directions for a software developer to head on when attempting to achieve the highest performance for its program.

The first direction is x86 or generally CPU programming and optimization, may that be x86, RISC, ARM, MIPS or any other CPU architecture.

There are millions of programmers that know how to work and optimize for general purpose processing architectures like x86 or ARM.

One other direction for the software developer is GPU compute programming and parallelization, but only specific tasks can take advantage and achieve greatly increased performance if they are specially optimized for GPU compute.

On the other hand, the software developer must keep in mind that he must invest in GPU compute programmers or to specialize his own team on the concept.

Then, he must weigh the performance increase predictions against the cost of the GPU compute extra programming work and the investments in hiring programmers that know how to do it.

AMD’s HSA vision consists in a special programing tool library that will effectively create a layer between the programmer and the hardware.

Practically the software programmer will write and code programs just like he used to or use very small modifications and simple optimizations, but the HSA layer will know how to best execute the code to take advantage of the hardware.

They are not saying that there won’t be any GPU compute work to do, but AMD is emphasizing on the fact that there is a much smaller effort involved in programming using the HSA method for HSA hardware than doing device specific coding like CUDA.

There is much sense in AMD’s initiative, as Intel practically agrees with them by also emphasizing that coding and porting for its Xeon Phi GPU compute architecture requires much less effort and investment than optimizing for CUDA.

Having huge and influential companies like Texas Instruments and ARM backing the initiative can only reassure AMD that HSA is going to make a serious impact on the industry.

Photo Gallery (6 Images)

Heterogeneous System Architecture Foundation
Heterogeneous System Architecture FoundationHeterogeneous System Architecture Foundation
+3more