The Blurred Boundary Between CPU and GPU

Intel Xeon E3-1200 V4 die (courtesy of Intel). The chip maker begins integrating Iris Pro graphics into the same chip.

With the launch of new processors with integrated Iris Pro graphics, Intel begins describing its products as more than CPUs. The Xeon E3-1200 V4 is "a CPU and GPU on the same chip," said Jim Blakley, general manager of Visual Cloud Computing for Intel’s Data Center Group.

Iris Pro is a major part of Intel's strategy to drive growth in traditional GPU market segments. (For more details on the GPU market, read Market Watch published by JPR.)  To speed up graphics-heavy workloads (like ray-traced rendering) and compute-intense workloads (like simulation), engineering and design software users tend to rely on the GPU as a coprocessor to augment the CPU. The GPU's highly parallel nature makes it an attractive device for such applications. However, Intel's CPU with integrated graphics can disrupt the market dynamics, especially for the low-end and mid-range segments.

Blakeley clarified, "Iris Pro graphics is not meant to compete in the high-end GPU market. The performance of Iris Pro is roughly equivalent of a mainstream graphics card." GPU maker NVIDIA, for instance, offers its NVIDIA Tesla cards to the high-end market, and Quadro cards to mainstream market, which encompasses CAD software users.

In the last four to five years, some researchers and scientists began using the GPU for much more than graphics. In their deployment of high performance computing (HPC) clusters, they harvest the GPUs' parallel processing horsepower to tackle large-scale computing tasks such as pattern recognition, data analysis, and weather prediction. To perform such tasks, however, the software has to support NVIDIA's CUDA parallel processing language. Some computer-aided engineering (CAE) software makers have refined their solvers to accelerate on the GPU by adding CUDA support.

Intel's Blakely suggests software vendors can tap into the additional horsepower in Iris Pro-based CPUs with little or no additional work. He explained, "If the application is written to use OpenGL or DirectX, then Iris Pro work without any additional programming effort. NVIDIA also supports OpenGL and DirectX. These are graphics programming APIs. If you want to do things that are more compute-oriented, then we support OpenCL programming on the GPU. This is the direct comparison with CUDA. Whereas CUDA is proprietary NVIDIA programming model, OpenCL is industry standard."

For highly parallel applications, computer users are expected to continue to rely on NVIDIA Tesla GPU or its equivalent. However, if software programs can take advantage of Iris Pro's additional horsepower without much effort, it's set to challenge the GPU's dominance in visualization-related tasks like image editing and 3D modeling.

Workstation users may also split their workload on the two GPUs in the same machine. Blakely pointed out, "You can use one of those GPUs for graphics, another for compute-intensive work."

Whereas Intel's Xeon E3-1200 V4 with integrated Iris Pro graphics represents a CPU market leader's penetration of the GPU's territory, NVIDIA's Tegra marked the GPU maker's encroachment of the CPU market. Used in many mobile devices in the market, the NVIDIA Tegra combines ARM architecture CPU and a GPU into a single system on a chip (SoC).

The boundary between CPU and GPU continues to blur as the two market leaders prepare to move into each other's territories. Intel's Blakely said, "Going forward, we'll add substantially better improvement to Iris Pro. We'll be developing more and more GPU features and continuing to increase its performance.