With more than two decades of high-performance data mining, artificial intelligence, and machine learning expertise, it did not take long for Antonella Rubicco (CEO) and Emilio Billi (CTO) to set up A3Cube Inc., a leading-edge enterprise in the Artificial Intelligence (AI) industry. As prominent members of several globally-recognized associations, the co-founders have led several of their clients to international fame. “We give our customers the capabilities to generate the maximum value from their data at incredible scale, making everything extremely simple to deploy, maintain and manage,” says Rubicco. The industry veterans continue to work for the enterprise’s ascent and have taken A3Cube to the next level of computing. AI promises to profoundly reduce human errors, provide endless virtual assistance and therefore, leave more time for creativity. In fact, 72% of business leaders termed AI as a business advantage and believe it is going to be fundamental in the future. About 47% of digitally-mature organizations have a defined AI strategy.
There is no doubt that AI is ubiquitous and here to stay. Most aspects of our lives from booking flights, to managing stocks, to personal health assistants, and even some great entertainment, have been touched and shaped by AI in one way or another. By replicating human intellect and behavior, AI automatically emulates and envisages patterns in real-time data, objects, and numbers using unique methodologies. AI is primarily determined by massive multi-variable linear algebra and intricate software systems.
AI has not only made enormous strides in our personal lives but has an immense potential to change businesses forever. The process of predicting patterns opens up new and exciting opportunities for companies to discover value from their data and thereby, take appropriate actions to accelerate business growth and excellence.
Unfortunately, only 61% of enterprises who have an innovation strategy are using AI to identify opportunities in data that would otherwise go unnoticed. Without the ability to understand and visualize hidden paths in data, determining the next steps or creating personalized customer experiences pose a significant challenge. While AI can be promising, existing solutions in the field are single boxes with a specified number of computational devices that cannot be scaled by the addition of nodes or other devices without having to rewrite the entire application code. Thus, these solutions are not only time consuming and costly but also result in resource inefficiency.
California-based company – A3Cube, is markedly different from current solution providers with their uniquely optimized and game-changing systems designed to address today’s AI challenges. “We have effectively evolved from an era of high-performance computing (HPC) to an era of high-performance data (HPD), and the key is to migrate from traditional storage expertise to supercomputing expertise centered on data,” says Billi. A3Cube helps overcome the limitations of current AI solutions with well-designed and easy-to-scale systems that fit various application needs.
Users are no longer constrained to single boxes, but a host of computational devices and nodes can be integrated easily without changing the application code. A3Cube’s systems are designed to scale elastically to match the required performance level without having to worry about the system hardware. Smaller computing elements amalgamate together to deliver a magnitude of computing power, at a fraction of the costs and time respect to other solutions on the market. "Critical data-driven applications and the amount of information these applications have to manage require a new approach to super-computing. A3Cube supercomputers enable greater precision, higher revenue, lesser costs and faster results," adds Rubicco.
Understanding the Ground Reality
The probability of AI to transform the world is becoming increasingly apparent. Gartner predicts that by 2020, AI technologies will be virtually present with every new product, service, and software. Despite this pervasiveness, AI still faces significant stumbling blocks, that need to be addressed on priority, so that AI’s utmost potential can be unlocked completely. There are four key challenges in today’s AI solutions:
Limitations of Hardware: Many of today’s AI solutions face severe restraints regarding hardware. Despite the increasing popularity of multi-GPU usage, even the most advanced and specialized servers are not designed to host more than 8-16 GPUs. Additionally, traditional legacy methods find it challenging to fit the timing required to maximize the potential of the new application. This inadequacy in current designs limits the ability to scale elastically without profoundly altering applications. Not to mention, the overall clustering architecture will critically affect performance. A study by Deloitte revealed that 47% of enterprises find it challenging to integrate AI projects with current systems and processes. A3Cube’s systems scale elastically, and users can easily include devices or nodes that fuse into the system without the need for any software modification.
Inefficient Use of Computing Resources: The inability to allocate resources for other computational tasks that need GPU acceleration limits the complete utilization of the GPUs. For example, if a function requires GPU acceleration but uses only 20% of the available GPU resources, it still cannot share the remaining 80% with another task because the whole GPU is locked by that single function. What this leads to is the complete underutilization of otherwise, very valuable assets. A3Cube creates platforms that permit the dynamic and concurrent use of all GPU idle cores for other workloads.
Data Access Latency: One of the most misjudged factors of incompetent performance in modern day applications is data access latency, especially when time-to-result is a critical factor. Latency not only limits scalability but also drastically affects exploitation of resources and efficacy of applications. The latency figures of A3Cube’s systems are the lowest in the market owing to A3Cube’s holistic approach to optimization from the OS to the hardware. “We solved the problem of latency by optimizing the storage/data layer and implementing a direct communication between different processing units,” reveals Rubicco.
Code Modification: To expand and amplify computational resources, many applications today no longer fit the hardware as a result of the execution time required. Developers, therefore, spend a lot of time and effort to reorganize, requalify and even rewrite the code that they have tested on a small platform so that it can run on a large scale. This challenge costs a lot of money and time for organizations. A3Cube has helped addressed this pain point by developing solutions that enable companies to seamlessly run their applications without any further software modification.
Empowering Businesses to be Future Ready
A3Cube’s robust, radical and purpose-built supercomputing architecture aims at maximizing performance while capitalizing on resource efficiency for computing-intensive data applications. Besides, A3Cube’s plug and play integrated computing platform aims to change the strategy of data analysis by ensuring the removal of every single bottleneck in the computation process. The company firmly believes in the importance of providing improved and better equipment (a synergetic and collaborative approach to software and hardware) to clients so they can generate the most valuable information from their data. A3Cube uniquely addresses the most complex challenges with a unique and easier solution widely different from what the current market offers.
The genesis of A3Cube’s ideas come from one of the most spectacular computer architectures of all times – The Connection Machine II. Created by Daniel W. Hillis, the Connection Machine was not just the first of its kind, but also represents a very realistic and modern approach to AI. “We were inspired by the Connection Machine to consider the needs for AI scalable hardware tools that can be used to solve any problem in the most efficient way. This approach provides our scientists and engineers with real concrete and ready to use systems that are easy to implement and one that can meet their actual needs,” expresses Rubicco. As a result, the company decided to explicitly develop products and solutions for high-performance machine learning, artificial intelligence, and data mining.
However, today’s supercomputers are not optimized for data-oriented operations, but for massive crunching number, creating significant challenges in the industry. Since data centers cannot handle intensive computing workloads, they also fail to address these problems competently. The key to being efficient lies in implementing parallelism among the different processing units. A distributed computing approach is required in combination with dedicated and massive parallel vector processors like GPUs to resolve problems in a suitable timeframe. “What A3Cube does is delivering the right computational power needed to train these giant mathematical models to fit the timing required and achieve goals in a complete plug and play solution,” explains Billi.
Furthermore, A3Cube emphasizes on two important strategies to achieve maximum productivity in its appliances and distributed computing.
The first strategy is to optimize all critical elements in the distributed computing chain that can pose as potential blocks for performance scalability. The key to a seamless communication between nodes is not in the bandwidth, but the actual problem lies in latency. Latency here represents the amount of time that each processor takes to transfer data. Considering that most standard commercial distributed computing systems today use Ethernet for communication, the issue of high latency is massive even with an entry level processor capable of more than 800 operations every 100 nanoseconds. On the other hand, A3Cube’s network can move about 4KB of data in 900 nanoseconds at 100 Gbit/s of speed. To transfer the same amount of data, the Ethernet takes 43,000 ns. During this time, the receiving processor is simply waiting, wasting precious time doing nothing. As a result, processors with the standard Ethernet communication lose about 340,000 potential operations in comparison to 7,200 with A3Cube’s technology, implying that A3Cube was able to successfully execute computing operations more than 47 times. The crucial difference here lies in the minute details.
The second strategy is to enhance memory in remote servers, to improve the possibility of storing massive neural networks and computing them in the fastest way possible. In fact, A3Cube’s systems can proficiently aggregate together 10s of terabytes of memory. Additionally, A3Cube has developed a process called RONNIEE Express. This process involves the usage of several coprocessors and other remote computing resources that belong to multiple systems as a single pool of assets for easy sharing and consistent in-memory communication.
“A3Cube’s two strategies build two classes of applications – one is the aggregation idea, and the other, built on the concept of most efficient distributed computing access. With these two classes of machines, we can address an array of computing needs for any existing and future application,” affirms Billi.
These high-level optimizations are made possible by A3Cube’s robust and highly-valuable architecture. A3Cube’s technology is pervasive and supports all types of GPUs and processors. The solutions are designed not only to meet existing applications but also to pave the way for future AI applications by moving the limits on the algorithms and not just on the computational tools.
Driving Business Excellence with an Innovation Lab
By considering all aspects that a new idea can bring forth, from complexity to programming to price, A3Cube built an Innovation Lab to explore and develop each solution in detail. “A3Cube reflects a love for nature. Everything in nature is optimized - the consumption of energy is the minimum imaginable, and results obtained are with maximum efficiency possible. In nature, the concept of “good enough” does not exist, and this is the motto of A3Cube technology and product development,” elucidates Billi.
The Innovation Lab focuses on ultra-fast communication, parallel data access, product-building optimization and new algorithm investigation. With an aim to test the limits of technology, the labs assess the impact of latency in any application, as well as the key advantages of employing different methods for moving data across machines. “We strongly believe that everything can change, modify, turn and even be removed as there no standards to follow. This compels us to invent new paths every day, thereby opening new scenarios and possibilities,” states Rubicco. Billi goes on to say, “In our labs, we give much importance to the interdisciplinary knowledge rather than in ultra-specialized skills. Our engineers and scientists must be skilled in multiple fields, with multiple specializations in different areas, and even multiple degrees. For us, this is the actual key to real innovation.”
A3Cube does not make any compromises in their radical approach for creativity. They are ready to take and stand by the right decision to elicit maximum computing power for each of their products and solutions. Tests in the innovation lab go deeper and deeper with each aspect, and this helps A3Cube rebuild successfully, driving excellence for enterprises every step of the way.
Accelerating Business Growth with Differentiated Products and Services
A3Cube’s cohesive, ultra-computing portfolio solution comprises of two separate series: The “Personal Supercomputing” Family and the “Exa-Supercomputing” Family. While the former starts from one up to 32 processing nodes designed specifically for data access optimization, the latter represents the enterprise’s cutting-edge structures that start from a handful of nodes that can scale exponentially up to 1000. “Our supercomputing solutions are powerful, scalable, flexible and a perfect fit for small organizations that require high-performance personal equipment without any internal technology experts,” explains Rubicco.
A3Cube’s range of products (KIRA-CS, KIRA-CS Hybrid, KIRA-DATA, KIRA-EXTREME, KIRA-GT, and GRIFO) are entirely plug-and-play turnkey solutions, specifically tailored to the applications. They come with scalable, optimized and pre-configured operating systems as well as tools that address bottlenecks in data mining, machine learning, big data analytics and artificial intelligence. “With a deep optimization in both software organization and hardware architecture, our products can scale up to thousands of computing cores, terabytes of memory and hundreds of GPUs shared in a unique way,” affirms Billi.
Explicitly optimized for critical AI applications are A3Cube’s KIRA-GT and GRIFO Composable System. KIRA-GT is a GPU-solution adapted for scaling out AI challenges and to produce complex neural networks on a rather large scale. The solution can efficiently run in a decentralized method and ascend to 100s of GPU that can be effortlessly shared among single or multiple tasks. Alternatively, GRIFO’s composable architecture powerfully scales resources up to 128 GPUs in a single machine, achieving unmatched performance in any AI application.
Redefining Customer Experience
A3Cube’s radical approach to ultra-computing provides customers with the opportunity to generate maximum value for their businesses. Additionally, these incredibly scalable solutions are easy to install, handle and sustain. With such an array of exclusively optimized and personalized solutions, A3Cube has assisted many of their clients to deal with the complexities and intricacies of today’s AI industry.
One such client involved in in-depth research for a rare disease discovered an unusual connection between a blood examination value with a specific geographic area. Upon further analysis using the neural network, the researcher uncovered the existence of a particular gene in that specific geographic population, thereby, opening up new and unconsidered pathways for further exploration.
On the other hand, in another fascinating case in point, A3Cube positively helped a retail chain improve and transformed their logistics across the state. Using AI, the client profiled customers accordingly and therefore were able to pay close attention to customer preferences and demands, ensuring the right delivery of goods. Furthermore, the client avoided the delivery of goods to the wrong place, as well as investing in products that do not match customer interest for each geographic area.
“To achieve this, the client had to train the AI to fit the problem with the timing requirement. When they incorporated a standard server with eight big GPUs earlier, they took 12 days to train the network and discovered that it was too time-consuming to achieve the desired results,” states Billi. Therefore, the client mitigated to A3Cube’s GRIFO Composable AI System where they managed to train the neural network within four hours, enabling the usage of the application for its scope without any changes. This slight alteration in the process not only enhanced efficiency but led to the proper utilization of resources, cost and time savings, and tremendous customer satisfaction.
Paving the Way for A Competitive Edge
With a mix of in-house inventors, technology experts and scientists, A3Cube has a diverse team that is at the center of all their production processes. “Everyone in our team knows that his or her work represents an indispensable contribution to the common progress and success: "overall quality" is the result of each person's own internal quality,” affirms Antonella. From the very beginning, A3Cube has always focused on providing the best solutions for storing, managing, exchanging and analyzing huge quantum of data. “For this reason, the increasing number of critical data-driven applications and the increasing amount of data these applications have to manage are strongly supporting the growth of the company,” she adds. One of the key points that help A3Cube stay ahead of the competition is the implementation of specific products and configurations tailored to the requirement of the application, instead of a single solution for everything.
Besides the employees and A3Cube’s unique modus operandi to data-driven computation, the company’s core strength are the co-founders. With their expertise and vision, the company can design advanced solutions and products, predict market changes effortlessly and create disruptive technologies that have transformed the complexities of the market. As a fast-growing company, A3Cube’s proposition is so unique; it has no comparison in today’s marketplace.
For the road ahead, the R&D team at A3Cube has the vision and competence to overpower competitors and stay ahead of the game. The firm focuses on breaking new ground in fresh markets, and to broaden the existing portfolio by implementing innovative and proficient products that deliver top-notch market performance. “We are also keen on providing technology as a service to cover the part of the market that does not prefer an on-premise solution. In a few years, A3Cube has the potential to be recognized as a company with the best performance for modern applications. We always want to be a step ahead with the right solutions,” concludes Rubicco.