By Edward Lim, Managing Consultant, CIZA Concept
Founded in 2006, Twistcode does research and development in high performance computing technology, especially massive-parallel computing and applications using central processing units (CPUs) and graphics processing units (GPUs). It thrives in accelerating any repetitive algorithm and/or process in most environments.
With its focus on high performance computing, the young and dynamic company hopes to let its clients – enterprises and government organisations – concentrate on what’s important to them. It provides services such as rendering and modelling, and tweaks technologies and twist codes to adapt to customers’ needs.
Frustration over cell architecture
It was frustration over Sony’s cell architecture that prompted Twistcode to consider other options for its work.
“We were testing the Sony cell infrastructure for gaming and wanted to learn more about it. However, it was difficult to code and we had problems getting licensing from Sony, said Nurazam Malim bin Sidik Malim, Chief Executive Officer of Twistcode.
In 2007, the company began looking at CUDA, a parallel computing platform and programming model that enables dramatic increases in computing performance by harnessing the power of the GPU. Twistcode was impressed with what CUDA could do.
“We chose CUDA because it is well documented and expanding rapidly,” noted Nurazam.
Since then, Twistcode has been using CUDA and NVIDIA GPUs for a multitude of projects, the latest being in high performance computing on big data framework in partnership with Telekom Malaysia (TM) R&D.
High performance computing on big data framework
According to Nurazam, they are combining big data and GPU together because they see a convergence of high velocity big data with high computational problems.
“We work with Twistcode because we wanted to find a better way to traditional Grid computing, which is distributive computation, whereas big data is about distributive instances. Each worker will have a GPU to work on sub-instances,” said Mohammad Shazri bin Shahrir, Researcher of TM R&D
Under the project’s high performance computing on big data framework, rich input will result in rich output through the process of acquisition, marshaling, analytics, and action.
In the acquisition stage, GPUs add value by decreasing computational time of compression and encryption and ensuring secure communication with high throughput. When marshaling, there is a reduction in computational time of compression, improvement in storage security with encrypted storage, and secure storage with fast access.
Moving on to the analytics stage, GPUs cuts computational time for analytics, migrates more algorithms towards ad-hoc analytics, and achieves high performance analytics with rich and ad-hoc output.
In the final stage, action, GPUs are able to deliver richer presentation on an ad-hoc basis from a bigger array of algorithms for reports and dashboard.
The GPU difference
To highlight the difference GPU computing can make, Nurazam shared that the Malaysia Genome Institute used 20 Xeon dual-socket processors and took 14 days to generate one DNA while Twistcode used one NVIDIA GTX 690 and took just 5 days to do the same thing.
“To create a model of a 47-storey building using Esteem 8, it took 300 minutes on a very fast computer but just 18 minutes with an NVIDIA GTX 690,” he added.