High-performance computing (HPC) is the ability to carry out large-scale and complex calculations quickly. Many computational problems require the ability to analyze very large data sets data quickly, including weather forecast, climate modeling, crash test simulations for cars, and in general entire classes of applications called computational fluid dynamics. In fact, these computational challenges are so large that they cannot be solved on a single computer even with many processor cores. Those problems require thousands to hundreds of thousands of processor cores to solve in a reasonable time frame. As a result, it is not possible to simply let a single computer run for a very long time to solve a problem – after all, what good does the forecast for tomorrow’s weather do if it is ready next week or month? HPC makes these societal, industrial, and technological advancements possible. Axel Kloth, President and CEO of Abacus Semiconductor Corporation, long ago recognized this tendency and is striving to address it with improved performance scaling, which would improve HPC.
Axel Kloth has a background in both physics and computer science. Additionally, he is a successful serial entrepreneur who is at the same time knowledgeable about technology and its business applications. He established SSRLabs (Scalable Systems Research Labs), an HPC-focused business, and Parimics, a manufacturer of vision processors and systems. These businesses paved the way for real-time object detection, motion analysis, and object tracking in hardware. With Parimics, the number of trackable objects was limited only by the number of objects in the frame set. SSRLabs was focused on accelerators for HPC. At SSRLabs Axel also developed security processor technology that was spun off into another startup named Axiado. While Axel was a founder and CTO of Axiado, he returned to HPC with Abacus Semiconductor Corporation.
Idea Behind Abacus Semiconductor Corporation
It was sheer frustration over the lack of linear scaling and all of the waste associated with it. As a physicist, Axel uses supercomputers. In his opinion that is now mirrored by experts in the field, a supercomputer does not scale as expected. The performance of a supercomputer with 100,000 servers does not always equal 100,000 times that of one server; in some cases, it is only 5000 times as fast. That leads to a waste of resources, including time, effort, space, management costs, and acquisition costs. Being a computer scientist, Axel tried to come up with ways to reduce this waste. Abacus Semiconductor Corporation was created as a result.
Revolutionizing the Market
With a novel system partitioning, Abacus Semiconductor Corporation has a processor, a smart memory, and two accelerators for special-purpose applications. Axel and his team succeeded in achieving two objectives: more linear performance scaling across thousands of cores across processors, and including all accelerator cores while maintaining the advantages of common memory and an open Instruction Set Architecture (ISA) with an established and expanding ecosystem. This processor is made to operate with well-known programming languages and standard compilers so that users can quickly deploy it in their operations.
These applications range from weather forecast to climate modeling, from crash test simulations of cars to aerodynamic research on aircraft and turbines to improve fuel efficiency, and they include research on how to make anodes, cathodes, and entire batteries last longer and provide better quick-charge capabilities. Large Memory Models such as GPT-3 and ChatGPT also benefit from processors with many cores and large-scale coherent memories.
As the world is moving on the phase of digitalization, artificial Intelligence (Al) and machine Learning (ML) becomes frontline of the game. “This has stepped out of being a niche product and is now being deployed on a broad scale. Both AI and ML are these days referred to as HPC (High Performance Compute) applications in the same way that traditional supercomputing was and still is” says Axel.
As a President & CEO
Axel is responsible to build a world-class team, to ensure the satisfaction of end users, and that the company thrives. “We are different from other companies in that we look at the potential of any new hire over just pedigree.” specifies by Axel.
Employees at Abacus Semiconductor Corporation are trained, and Axel is committed to expanding institutional knowledge. While in the short run it may be cheaper to outsource and buy IP, Axel believes that in the long run, internal IP designed by internal resources is vastly more useful and valuable. He says, “We only deviate from that if there simply is no internal talent available and if we cannot find talent from the outside that we can bring in.”
Skeptical of the Industry
The semiconductor industry has been the most important driving force behind the accumulation of knowledge in human history. Axel seeks to ensure that everyone benefits from this technology. According to him, it must be avoided that certain groups of people are excluded from business advancements. He argues that no approach should be unfamiliar or difficult for the average person to understand. If the cutting-edge implications of computers, the Internet, AI, and ML remain inaccessible to the average person, it is defined as a failure to communicate the benefits to a high degree. Axel suggests, “We, unfortunately, have to prepare for people to be skeptical of science and for the fact that they embrace anti-scientific sentiments. As an industry, we have to counter this sentiment and must work towards a more general understanding and acceptance of scientific principles.“
Constructive Criticism is Necessary
Axel is of the strong opinion that good managers and supervisors hire to solve a specific problem within their company. In order for that to happen, every supervisor and manager must hire those that are best suited for the task. It is also entirely necessary to build a company culture that honors, institutes, and lives constructive criticism. After all, supervisors and managers are there to provide guidance, and not solutions. As such, it is necessary that every employee feels like their input and feedback is valued. That feedback includes constructive criticism. That means that anyone can be wrong, and everyone has the chance and the right to prove his supervisor or manager, or even team mates wrong. A good leader will always value constructive criticism even if they are proven wrong. He didn’t realize this for a while in the beginning of his career.
As a CEO today, he encourages his staff to speak up for the greater good, insisting that even if a suggestion contradicts his own, if it proves to be a better solution than his, credit should go to the one who made it. As a CEO, his biggest risk is having excellent employees who don’t think they can question and disagree with him, leading to frustration and turnover.
Corporate Responsibility Strategy
Resources are a problem for any startup in the semiconductor industry. Some businesses are very well-funded, and as a result, they have more accountability. Simply put, Axel and the company have an obligation to their shareholders when and if they raise funds. This obligation is to ensure that the business continues to exist and grow to the point when an exit (sale or merger and acquisition) provides the founder, shareholders, and employees with enough funds to make them want to collaborate once more.
Towards the Customer
Axel mentioned that HPC is a very conservative industry that requires long-term building of trust. Developing trust in the customer’s views is crucial. “We show every customer what we have, where we stand, and what we can do,” confirms Axel. When and if his company is unable to optimize a customer’s application runtime, it notifies them. If his teams believe and have data to back up the assumption that they can improve runtime by a factor of 10, customers must be able to trust them, as supercomputers are pretty expensive. He works immensely hard to gain customer loyalty and plans for future development. Axel hopes that by then they will see the first supercomputers with their processors, accelerators, and smart memories.
Written by Steve Sanchez.
