China Academy of Building Research dramatically improves the performance of earthquake analysis software with AMD FirePro™ professional graphics
Ensuring the safety of tall towers
Dominated by tall buildings the Beijing skyline illustrates the construction boom that has resulted in numerous skyscrapers all over China. Although it has a long history of earthquake incidents, China also has very strict building codes. All developers must submit evidence that they are meeting the obligatory standards of safety and performance for any proposed designs. Such regulations enable high-rise buildings to survive so that people are protected and damage is minimized.
The standards and regulations that apply to the construction industry are created and enforced by China Academy of Building Research (CABR). This organization conducts quality tests on engineering and construction processes as well as products such as air conditioning equipment, elevators and building materials.
As a commercial entity that is half state-owned, CABR develops and manages PKPM software, which is used for every aspect of construction design and engineering, including water supply and drainage, heating, ventilation, electrical and facilities management. PKPM is used for day-to-day work by tens of thousands of institutes and companies, representing most of the construction CAD market in China.
One of the most widely used modules of PKPM is SAUSAGE, which can simulate how a structure will respond to an earthquake. Because it has to analyze a wide and dynamic range of parameters such as elasticity, plasticity, time and history, SAUSAGE is extremely hungry for computing power. The software was originally developed on a parallel programming and computing platform, and some analysis scenarios can easily take 60 hours, in some cases one scenario might take a week to run. Because many different scenarios have to be carried out for each building, the analysis process sometimes takes several months. The impact on nation-wide construction projects is unnecessary delay and costly unforeseen expenditure.
On a quest for socially responsible efficiency
Responding to this problem and to user demand for access to GPU acceleration such as that provided by AMD graphics cards, CABR decided to develop an OpenCL version of SAUSAGE. In 2014, CABR engineers who understand SAUSAGE algorithms worked closely with AMD OpenCL specialists and together they ported most of the computing for SAUSAGE to OpenCL, making it possible to use the power of a graphics processing unit to speed up analysis.
Under the lead of PKPM Project Manager, Dr Qi Nie, who is responsible for OpenCL optimization, CABR developers went on to test a range of professional graphics solutions. These included two AMD cards, one CPU solution and one card designed for the original version of SAUSAGE. Each was used for processing five different sets of data.
The SAUSAGE algorithms are particularly suitable for GPU parallel computing, partly because the software requires extensive double precision data processing. Testing demonstrated huge improvements in performance with the OpenCL version of SAUSAGE running on AMD FirePro™ W8100 and AMD FirePro™ W9100 professional graphics.
Peak performance now available
PKPM users now have the choice of running an OpenCL version of SAUSAGE on AMD Firepro™ professional graphics -- an option for accelerated performance that was not available to them before.
“Since we cooperated with AMD, we have been using its professional graphics cards and OpenCL programming technology,” said Dr.Qi Nie, China Academy of Building Research. “Having started in 2014, we are still working on the project in 2015. We have received excellent support from AMD and now we are using AMD FirePro™ W9100 professional graphics cards because they provide outstanding results.”
AMD FireProTM W8100 Fast Facts:
- Memory: 8GB GDDR5
- Compute performance: Up to 4.2 TFLOPs peak single precision floating-point performance
- AMD Eyefinity technology: support up to 4 displays1
- Supports: OpenCL™ 2.0
AMD FireProTM W9100 Fast Facts:
- Memory: 16GB GDDR5
- Compute performance: Up to 5.24 TFLOPs peak single precision floating-point performance
- AMD Eyefinity technology: support up to 6 displays1
- Supports: OpenCL™ 2.0
