Literally GPGPU means: "General Purpose computing on Graphics Processing Units". Or, using the tremendous computing power of the GPU to accelerate any computation or data intensive processing.
Following, a few examples of the benefits from adopting GPGPU. |
Flooding simulationThis company performs risk assessment for house owners, real estate investors and insurance companies. One of the aspects they investigate is risk of flooding. To determine that risk for a specific location they use a high detail simulation. Traditionally a typical simulation run takes 3 days to complete. The transition to GPGPU reduces that to one and a half hours. GPGPU advantages:
|
Artifact detection in materialsA research institute develops a way to inspect the integrity of solid materials by use of ultrasonic waves. The procedure is to compare a simulated run with actual measurements. Differences could point to artifacts in for instance welds. For this technology to work in a product, inspectors would have to be able to take testing equipment to locations like factories and assembly lines. GPGPU unlocks enough computing power in a laptop to enable this. GPGPU advantages:
|
 |
 |
Crowd simulationA research department specializes in safety and risk analysis for buildings, cities and festivities. Crowd simulation is an important part of their investigation. GPGPU advantages:
|
Ink-jet PCB printingA technology start-up developed a new and efficient way to print PCB designs using Ink-jet technology. To control application of ink, complex calculations involved incorporating various ink flow properties. To reach sufficient precision PCB designs come as bitmaps of over 60000 mega pixels. The time it takes to calculate printing instructions for one design was over 3 hours. Which was unacceptable. GPGPU on a single GPU reduces calculation time to 8 minutes. Further optimizations can improve on that still. GPGPU advantages:
|
 |
A fast GPU board costs a few hundred euros and can be used to upgrade most contemporary computer systems. So a small investment can extent the lifetime of existing hardware considerably. In many cases the need for expensive dedicated hardware is eliminated.
A single modern GPU is on average 30 times faster then a typical multi-core CPU. In many cases acceleration of 50 times is possible. Sometimes, speed increase of up to 100 times can be achieved using GPGPU.
Why would a GPU be so much faster. Simply put, because it was optimized to do thousands of jobs at the same time. Originally the GPU was designed for complex calculations to determine the color of millions of pixels many times per second.
The speed gain can be used to deliver faster or to reduce cost. But it can also be used to do more precise calculations, increase the complexity of models, upgrade algorithms or use better simulations. An other way to increase quality can be to spend more time in quality control by performing extra checks, verifying results or to increase the number of iterations.
Calculations or quality levels that demanded large machines or off-line calculation could now be done on a desktop, possibly in real-time. This opens up new possibilities for products that until recent were limited to a select audience.
Since it is possible to equip a laptop with a GPU, GPGPU can also make your product or service mobile.
If a application requires more processing power then a single GPU can deliver, GPGPU can scale easily over multiple GPUs. Adding processing power until the target is met.
OpenCL is a GPGPU solution that allows for transparent distribution of calculations over many different processing hardware. This can unlock all available power of a system. For example processing can be distributed over cores of one or several CPUs, GPUs, or other supported hardware. The number of hardware vendors supporting OpenCL is growing. IBM has recently released OpenCL support for their Cell Broadband Engine.
