by Anton Shilov
07/22/2010 | 06:04 PM
Graphics processing units (GPUs) will continue to evolve in the same direction that they have been doing so far: towards more flexibility and greater performance, said Nvidia Corp.'s fellow David Kirk on Thursday, which is likely to open new markets. In spite of recent criticism towards personal computers by his own colleagues, Mr. Kirk believes that personal computers in general and desktops in particular will continue to exist and offer best performance possible.
The general evolution of graphics chips has always been towards and flexibility; meanwhile, the evolution of already flexible central processing units (CPUs) has always been towards higher parallelism. No surprising that at some point Intel Corp. decided at a point in the past to wed flexibility of x86 microprocessors with massive parallelism of GPUs; but without any success. According to David Kirk, the trend will still continue to exist and eventually it will be much easier to program for GPUs, a crucial feature in order to compete against CPUs when it comes to serious supercomputer deployments.
"I think that we will continue to see GPUs becoming more pervasive and useful for PC platforms as well as other computing devices and consumer electronics. I am pretty sure that GPUs will become increasingly parallel and flexible and will also become more computationally efficient. We are also working to make GPUs easier to program and this involves continuing to evolve CUDA to make it more powerful," said Nvidia's fellow during a public discussion of the present and future of GPUs.
The high-ranking executive of Nvidia did not elaborate about possible "other" computing products that GPUs may address. At present graphics processors are useful for rendering of video games, assisting certain consumer-oriented programs or using in high-performance computing (HPC) area. One thing that GPUs cannot do is running Microsoft Windows operating systems (OSs). But Mr. Kirk reminds that not everything depends on x86 and Windows and eventually it may be possible to run certain OS on a special chip with a highly-parallel graphics core. For example, Nvidia's Tegra-series of system-on-chip (SoC) devices can run OSs that are designed for ARM processors; but those SoCs are aimed at low-power devices and do not feature advanced graphics cores.
"Certainly anything is possible! We have already demonstrated an integrated SoC in the form of Tegra, that runs an OS as well as graphics tasks. You could probably almost run an OS on a GPU now, but it would be a pretty inefficient use of the massively parallel hardware. In our current opinion, it is better to pair the GPU with a CPU to run the OS, and handle keyboard, mouse, etc.," Mr. Kirk said.
Recently Neil Trevett, the head of Khronos Group that defines computer standards like OpenGL or OpenCL, and who is also the president of mobile content development at Nvidia, said that personal computers in their current form(s) would seem archaic. But, according to Mr. Kirk desktop personal computers will continue to push the boundaries as well as lead in terms of performance and capabilities.
"I think that the desktop computer certainly has a future, because some users will always want absolutely as much performance as they can get. There is no way a laptop or handheld can get to the same level. So, I believe that desktops will continue to push the boundary of performance and features, including things like 3D vision, surround gaming, and other powerful technologies," said David Kirk.