Intel Takes Different Approach to Large Parallel Computations

By Peter Varhol

Thanks to the splash caused by NVIDIA with the introduction of its Tesla high-performance graphics and floating point processor and its massively parallel CUDA architecture, there is growing interest in using multiple processor cores to accelerate the computation of parallel operations. NVIDIA used the Tesla processor to build an inexpensive 960-core computer specializing in rapidly solving highly parallel computing problems. The Tesla has a parallel “many-core” architecture, each core capable of running thousands of threads simultaneously. If an application is suited to this kind of architecture, the processor can offer large performance benefits.

Because the X86-standard processors are general purpose in nature, the theory goes that it cannot be a specialized compute processor for parallel and graphical operations. These processors, produced in large numbers by Intel and competitor AMD, power most of the desktop and laptop computers used today.

Despite this almost universal use, the X86 has never been a speed burner. In the 1990s,  it was outclassed by virtually all of the reduced instruction set computing (RISC) processors. Intel even has its own replacement for the X86, the Itanium,  which is anywhere from marginally to blazingly faster, depending on the computing problem. To use the X86 as a massively parallel processor in the same way as the Tesla seems like a losing proposition.

But NVIDIA’s strategy is a difficult one, too. Despite having a kick-ass processor and an architecture that supports many cores for parallel operations, it doesn’t run existing software. Design-tool vendors will have to port their existing products to run on this architecture, a process that is often easy enough for 90 percent of the code, but devilishly difficult for the last 10 percent. And then developers have to work at making the code more parallel—that is, better able to use all of those cores.

Now Intel is about to put the theory that you can’t teach an old dog new tricks to the test once again. This week at SIGGRAPH, the company is scheduled to reveal details of its new Larrabee architecture and processor family. Larrabee, which is Intel’s first multicore architecture designed for high throughput applications and features a programmable graphics pipeline, uses multiple X66 processor cores rather than some exotic processor that may be more suited for the task.

The first Larrabee discrete graphics products are due in the late 2009-early 2010 timeframe. The same logic that made the less-than-perfect X86 the primary choice for a general-purpose processor might also work in making it a specialized graphics engine.

The compelling thing about Larrabee is that far fewer code changes will be necessary in order to use existing applications. In fact, all existing applications should be able to use it, albeit not nearly to its potential. Rewriting code to make it more parallel where appropriate will offer sometimes-dramatic performance gains, but that requires additional development effort. Still, it will not take the effort required to port the code to an entirely new processor.

For rewriting code to be more parallel, Intel has introduced the Intel Parallel Studio, a set of plug-ins for Microsoft Visual Studio designed to more easily develop code that can run on multiple cores. Intel Parallel Composer consists of a parallel debugger plug-in that simplifies the debugging of parallel code and helps to ensure thread accuracy. Parallel Inspector detects threading and memory errors and provides guidance to help ensure application reliability. Parallel Amplifier is a performance profiler that makes it straightforward to quickly find multicore performance bottlenecks without needing to know the processor architecture or assembly code.

Intel knows about producing the best code for its own processors, and developers can use these tools as well as the Intel C and C++ compiler from within Microsoft Visual Studio, the development environment of choice for many. My bet is that Intel knows how to make its processors appealing, and will do so yet again with Larrabee.