CPU+GPU+DSP Becomes Mainstream

“With the advent of visual computing era, users hope a richer graphics experience from equipment. Artwork generation is a highly parallel task; when resolution improves; content becomes increasingly complex, GPU and other non-contiguous units can help CPU handle continuous tasks,” says Dennis Laudick, market VP at ARM’s multimedia processor division.

 

As we can find CPU is moving the road of multi-cores; then question comes: will GPU rapidly upgrade the number of cores just like CPU? “From GPU perspective, the industry is easy to draw affirmative answer,” notes Dennis, “however, more ‘cores’ doesn’t mean higher performance; even with 2 times, 3 times or 10 times more cores than current GPU, the product would finally see lower performance and higher power consumption.”

 

“In the eyes of different people, good GPU has different standards. For graphics engineer, a good GPU needs good performance and the most advanced features, such as OpenGL ES3; for system-on-chip (SoC) engineers, low power consumption, small size, advanced energy management and high-quality design are important; product managers pose requirements for high scalability, maximized efficiency and minimized market risk; while for OEM vendors, a good GPU can help open up new application markets, just as ARM open up new application markets with GPU computing,” says Dennis.

 

Bao Shanquan, director of product marketing at Qualcomm says “we don’t place emphasis on the number of core, because first, GPU’s performance can’t be determined by simple number; then ‘Cooperative Work’ and ‘heterogeneous computing’ by GPU and other components is the direction of mobile processor.”

By contrast, David Harold, senior director of marketing communications at Imagination Technologies holds different point: adding core is significant for GPU. “Considering performance, power consumption and size, GPU and CPU exists fundamentally different. For GPU, adding more cores can get straight extensions in performance.”

 

Vice president of marketing at ICube, Felix agrees, “Comparing with CPU, GPU shows different. GPU naturally is a multi-core multithreaded architecture; current GPU software is highly parallel. Hardware parallelisms can immediately benefits software performance, so the auditing of GPU cores is more natural and practical than CPU. On the content side, games and 3D content enables continuing to increase complexity, which poses continuing requirements for GPU performance. For instance, 'Temple Run’ is not particularly brilliant; but it has drained a lot of GPU performance of many chips.”

 

Felix also stressed, GPU’s processing object is pixel; the improvement of screen resolution and fluency (refresh rate) directly raise requirements for GPU’s data processing bandwidth and computing. What users feel from GPU stagnation is dull picture; but multi-turn progress bar in CPU stagnation; apparently, users like the former.

 

Cooperative work is playing more important role. Felix believes using CPU、GPU、DSP and video processor in the calculation-level camera shows more advantages than CPU, which is not the most efficient processors for a variety of workloads. Via making other processors easier to use and more programmable, application developers can adopt GPU, DSP and other efficient programmable processor to reduce CPU’s burden.

 

It is estimated that among currently mobile phone processor, CPU processes task of 15%-20%; more are completed by GPU and DSP. Concept of ‘cooperative work’ is spreading in GPU makers.

 

Bao adds, the simple increase of CPU cores is not the direction of improve chip performance. Limited by heating and power ‘budget’, mobile processors must use ‘heterogeneous computing’ to generate new experience. That is to say, processor attributes performance to multiple components.

 

“Taking Google’s newly released Android 4.3 as an example, OpenGL ES3.0 API support is one of the biggest bright spots. The industry takes the norm as the new starting of mobile 3D graphics. Google is promoting Renderscript technology, which, from another point of view, is encouraging developers to make full use of GPU, DSP to develop content,” he says.

 

Bao believes GPU has similar importance to CPU. The reason for Google is that CPU is only suitable for simple logic operations; while user experience-related content should be handled by GPU and DSP. Qualcomm and Google know the importance of GPU and DSP in the phone.

 

As chips and operating systems companies are boosting GPU and DSP; CPU+GPU+DSP ‘cooperative work’ is becoming mainstream.