The Future of Computing: Integrated GPU and CPUs

For those that might not be as well versed in the realm of computer lingo, a general processing unit (otherwise known as a GPU) is the heart most video processing devices. Its difference from a CPU is its ability to run several processes in parallel compared to a CPU that runs a single process at a time but very quickly.

The GPU differs from the CPU as a result of application, where the CPU is designed to run single program quickly versus a GPU that is used at performing many calculations simultaneously to render graphics on a computer screen. What we will be seeing in the future is a merging of these two chips into a single processing unit and the company that is probably the best suited for this is likely AMD should they work out the kinks in their merger with ATi.

Several years ago, the computer industry was obsessed with improving computer speed by ramping speeding up the CPU clock cycle, or the number of calculations a single set of computer circuitry can handle per second. Unfortunately, further advancements in this area provides diminishing returns as CPUs require more power the faster they operate, which is something undesirable considering that many CPUs are now being used in battery operated Laptops. In addition, it is also becoming increasingly challenging to continually push the clock speed faster. So what the computer industry has done instead is go through a simpler route; the integration of several processing units into a single silicon package to reap further performance gains. This is possible due to patterning technologies allowing makers such as AMD and Intel make smaller and smaller CPUs making it cheaper to design chips with several integrated processors on a single chip, resulting in multi-core CPUs being currently sold on the market. I believe that this trend will change in the coming years.

A new revolution in computing devices is occurring with a push to smaller and more portable computing systems. My previous post covered touched lightly on the growing trend of ultra-portable computers-- expect to see more of them in the future. With computers moving in this direction, there will be a need for tighter integration of computer parts to achieve further miniaturization. I have covered this before in a previous post but will reiterate some of the important points here.

Components for both desktops and laptop computers are made of discretized parts. In the desktop case, the computer parts consist of modular cards that can be swapped into a mainboard. If you want a new video card, simply buy a newer version and swap the parts. Laptops, on the other hand, do away with the modular design to focus on integrating the video, sound and networking chips onto a single mainboard for a compact design. This miniaturization will continue, but this time will occur on the silicon chip level as opposed to the circuit board level and I believe that simply adding more processor cores to a silicon die is the wrong solution.

I predict that the next leap in processing technology will be the integration of a CPU and a GPU core onto a single die. I believe that this will be a very big step in the development of smaller ultra-portable hand-held computing devices.

The CPU and the GPU are the most computationally intensive chips in a computer system, quite often with the CPU continually communicating with the GPU to render graphics on to a computer monitor. A communication lag exists between the CPU and the GPU as a result of current computer design architecture. The following graph illustrates this issue:


The northbride/southbridge configuration of a computer (source: Wikipedia)

Computer chips and peripheral devices are connected using memory controller chips, otherwise known as the northbridge and southbridge chips. Data between devices flows between data lines, otherwise known as a "bus." Video cards, as a result of the amount of data they continually process are generally given a fast bus to allow the video card to transfer data quickly between the CPU or on-board RAM (which is not often as most video cards come integrated with their own RAM). Unfortunately, the data transfer limitation of the bus between the CPU and GPU is a bottleneck, especially for high resolution 3D rendering as sufficient data cannot reach the GPU quickly enough, sometimes resulting in choppy video at higher resolutions. The obvious solution is the tighter integration of the CPU and the GPU together resulting in superior data transfer rates and processing capabilities. The combination of these chips would also remove the need for a northbidge chip, thus simplifying the computer architecture further.

There are also other advantages to having a GPU core integrated with a CPU. Primarily for the reason that a GPU is a parallel processing unit, though somewhat slower than a CPU. Most operating systems operate using several concurrent processes running at the "same time." In reality, this is because the CPU operates very quickly; it rotates between running these processes giving the illusion that it does these things at the same time. Unfortunately, this is not an ideal solution because the computer must either save it's memory of what it was doing and for one process while it switches to another process. This memory swapping business is what often slows down computers running many concurrent programs. However, this is fine for processes with some little time lag; so that it can run off and do something else while it waits for data or commands to arrive.

Unfortunately, in other cases, this might not be possible when multiple processes require continual CPU attention and since a CPU has the equivalent of a one track mind, it becomes advantageous to have other computing components to deal with these continual demands. This is originally why discrete chips were often used for specialized functions including video, sound and networking. Enter the GPU, the silver bullet.

Because of the parallel design of the GPU, integration of a chip like this in addition to the CPU will allow for significant performance benefits, allowing a computer to handle more processes simultaneously while having a main CPU for computationally intensive processes. The GPU in it's current state could replace the sound chips with all processing done directly by the GPU. This is advantageous since sound and video often go hand-in-hand in video applications.

This design mirrors the architecture of the human brain and body as it should be no surprise that our sense of sight, hearing and taste are kept close to the brain reducing the lag time in processing and responding to important sensory data quickly. The sense of touch is somewhat different since it encompasses the whole body, but it is interesting to note that the body has a spare processor in the spine for reflex responses from sudden dangerous touch responses.

The future of computing will go in this direction and the integration of multi-core CPUs are a crude step in this direction-- expect more eloquent GPU-CPU chips to come out in the next few years with solutions in the works from both Intel and AMD.

I believe AMD is in the best position to accomplish this after acquiring ATi and it's portfolio of design expertise in developing video card technology. Unfortunately, the merger is not going very well at the moment, causing AMD to stumble. AMD's current offering of CPUs are somewhat inferior compared to Intel's offering in terms of maximum computational power, but AMD offers a better price/performance compared to Intel chips. What I am certain of is that Intel has limited design experience with their motherboard integrated video chips as they provide the poorest performance when compared against ATi and Nvidia's solutions. As Nvidia currently does not have the same close working relationship with Intel as ATi has with AMD, I doubt that and Intel-Nvidia CPU-GPU combination will be produced in the near future unless AMD-ATi dominates the market with their offerings.

The target for AMD's new CPU-GPU chips will most likely be for the growing mobile computing market segment and I believe this to be the wisest move. The next few years in computing technology will be quite interesting.