“Our core competency is accelerating code for heterogeneous architectures”: MulticoreWare

Multicore ware was outset in 2009 by A. G. Karunakaran (President and CEO); Headquartered in Saratoga, California with offices in St.Louis, Missouri, Urbana Champaign, Illinois, Beijing, China and Chennai, India. MulticoreWare offers programmer productivity tools for OpenCL, CUDA and other multi-core programming models. MulticoreWare, Inc is a leading provider of high performance video, computer vision and imaging software libraries, and a software solutions company, providing developer tools and professional services focusing on accelerating compute-intensive applications.

MulticoreWare; in a flash

MulticoreWare was founded by a team of researchers that wanted a better way to program for heterogeneous architectures. With the advent of GPGPUs and the increasing prevalence of multi-core, multi-architecture platforms, the company’s clients were struggling with the difficulties of using these platforms efficiently.

‘We started as a boot-strapped services company and have since expanded our portfolio to span products and services related to compilers, machine learning, video codecs, image processing and augmented/virtual reality. Our hardware expertise has also expanded with our team; we now employ experts on HPC and Cloud Computing, GPGPUs, DSPs, FPGAs, and mobile and embedded platforms. We specialize in accelerating software and algorithms, so if your code targets a multi-core, heterogeneous platform, we can help’.

Today, the company has expanded to over 250 employees located in the US, China, India, and Taiwan. Our global reach provides us unique advantages; our engineers are able work with client teams for high-bandwidth development, offices in China and India allow us to target rapidly growing Asian markets, and we can provide localized support in these regions.

Technology at its best!

H.265/HEVC Tools & Libraries

MulticoreWare founded and maintain the leading open-source H.265/HEVC encoder through the x265 project. While it took x264 a number of years to mature to the point of being commercially competitive, we accelerated the pace of development by bringing commercial funding to the x265 project. HEVC is extremely compute-intensive and our expertise in performance optimization gave our x265 development team a critical advantage over traditional video codec vendors.

Machine Learning & Neural Networks

Deep learning techniques, specifically convolutional neural networks (CNNs) have led to break-throughs in advanced driver assistance (ADAS), security, surveillance, robotics, manufacturing and other industries. This wave of technologies will enable autonomous vehicles and an unprecedented level of automation across all industries. MulticoreWare has a team of researchers developing advanced CNN solutions- we are applying our expertise in heterogeneous computing to enable these applications on low-power embedded, FPGA, DSP, and GPU hardware.

Compliers and code Analysis

State-of-the-art compilers are necessary to create efficient, high-performance code for heterogeneous platforms. MulticoreWare develops compiler technologies that allow the same code to target multiple architectures and maintain performance and efficiency. Our core compiler technology is designed with LLVM by our research team located at the University of Illinois at Urbana-Champaign. We also develop tools to reduce code duplication and reveal performance bottlenecks in existing code bases.

Vision for Augmented & Virtual Reality

Augmented reality (AR) and virtual reality (VR) applications have had exciting new advancements in recent years. AR and VR headsets are now equipped with complex heterogeneous architectures that require computer vision software that provides both high-performance and low power usage. MulticoreWare has experience porting vision algorithms to virtually all the unique architectures found in these headsets. We are well-versed in popular vision libraries such as OpenCV and OpenVX, but often customize functions for improved performance.

Products and Services

Full-stack Neural Network Services Data Labeling

MulticoreWare has a dedicated team of in-house data analysts to perform any type of labeling task. Tagging can be performed confidentially and the team can be trained for complex labeling tasks that cannot typically be done by untrained eyes. We use our own custom tools to perform labeling quickly and accurately and also provide these tools to our customers.

Design & Training

Choosing neural network architecture for your target hardware and applications and training it is no trivial task. The team of 50+ experts can leverage their research knowledge to design and train a network for your task and provide a comprehensive solution using our own proprietary technology or using open-source technology such as Torch7, Caffe, CUDA Convnet, Theano and others. The company can also provide consulting services based around these technologies.

Platform Optimization

The platform for running the prediction path of neural networks can include server-class hardware with accelerators, mobile device SoCs, FPGAs, and custom embedded platforms. Based on your target platform, compute resources, memory footprint, and mathematical precision must be chosen very carefully. We’re experienced in optimizing a neural network path for a number of architectures and platforms to improve performance and power efficiency while retaining its accuracy and predictive capability.

Incremental Updates and Maintenance

Updating and maintaining a neural network solution is essential to its continued viability. We can update and incrementally train a network as more data becomes available to continuously improve its accuracy. As the target hardware for the network advances, we can also upgrade the network’s predictive capabilities by using additional computer power, expanded memory, and changing power profiles. We can also deliver solutions that allow the end user to easily add data and incrementally train and improve the system.

Meet the Master brain

G. Karunakaran, President and CEO: AGK has 25 years experience in General Management, Engineering and Marketing for the computing, semi-conductor and embedded markets.

AGK was the founding President and CEO of GDA Technologies Inc., a leading Intellectual Property licensing and electronics design services Company. GDA was purchased by L & T Infotech, India in March 2007. At GDA, he was responsible for leadership development, growth strategy, prudent cash management and worked with leading Semiconductor companies to commercialize the silicon intellectual property blocks. Under his leadership GDA made it to the INC 500 list, Si100 lists, grew to 400 employees, delivered on significant product development engagements with leading systems and semiconductor companies and became a leading supplier of High speed Serial I/O Semiconductor Intellectual blocks. He has deep experience in bootstrapping companies and in strategic M&A transactions, having consummated deals for GDA with L&T Infotech and Rambus.

“We’ve applied our expertise to accelerate video, machine learning, neural networks, computer vision, compiler technologies, and more.”