For twenty years Magro was a user of supercomputers, a physicist who made models and ran them on some of the few supercomputers that existed at the time, then he moved on to develop the HPC at Intel and made the concept of powerful parallel computers more accessible and now, as mentioned, he promotes this technology further at Google where he Assists with artificial intelligence
Bill Magro, CTO of high-performance computing at Google: The cloud can be used as a solution to the shortage of chips. Magro was for 20 years a physicist in the field of quantum physics who used supercomputers to run complex models, then for about twenty years at Intel where he promoted the development of the field of high-performance computing as a technological leader and for the last two years he is the CTO of the field of high-performance computing and the cloud at Google, which despite its position as a user it developed Chips that will accelerate the training of artificial intelligence systems and optimize YouTube video coding.
Q: Please tell us a little about yourself, your career and how the transition from academia to industry took place
“While I have worked in high-performance computing my entire career, my formal training is as a quantum physicist. In graduate school, my research involved conducting large-scale parallel simulations of concentrated matter systems, and I was fortunate to have access to four National Science Foundation-funded supercomputer centers in the United States."
"I did my master's research at NCSA at the University of Illinois and my postdoctoral work at the Cornell Theory Center (two of the four centers of the US National Science Foundation - NSF). These experiences gave me exposure to a wide variety of HPC system architectures and application programming models, and I decided to pursue my interest in HPC instead of continuing as a physics researcher.”
"I joined a small company that provides software tools for HPC, which was acquired by Intel. At Intel, I spent about 20 years in a technical leadership role, developing and executing strategies to build and grow the HPC business. My main interests were HPC and cloud and I decided in 2020 to join Google Cloud to continue this full time position.”
Q: What made it possible for not only advanced users like you, but almost everyone to use high performance computing?
“In the early days of HPC, supercomputers were very expensive and required extensive expertise to operate them. There was also a high financial barrier to purchasing and operating these systems. Gaining access was difficult, and those who had access often had to be experts not only in the use but also in the simulation and modeling applications.”
“Many things have changed and made HPC more accessible. First, the advent of sophisticated and well-supported commercial HPC software reduced technical complexity for many users, allowing them to focus on the problems they wanted to solve rather than on implementing the tools themselves. Another important factor was the increasing computational power of microprocessor-based systems, which led to the industry's transition to distributed or clustered HPC architectures. The shift to industry-standard x86 servers has strengthened and Linux has matured, the hardware costs of HPC systems have been significantly reduced."
"However, several barriers remain, including the need for expertise in system design and operation, the need for data center facilities and the high cost of unused infrastructure. The emergence of cloud-based HPC helps address the remaining challenges, making HPC more accessible than ever to a growing number of users.”
Q: What is unique about Google's HPC solution?
Like other cloud providers, Google Cloud can run most common workloads in HPC environments. What is different about Google's cloud is its commitment to an open, hybrid and multi-cloud future. We feel that hybrid HPC environments that combine on-premise and cloud resources are often the best choice for customers, and we adopt open technology standards to enable workload mobility – between on-premise environments, Google Cloud and even competing cloud provider environments.
We also feel our AI and analytics capabilities are best-in-class, and we help HPC users leverage these capabilities to accelerate their time to insights.
Finally, we like to see ourselves more as a partner than a supplier, bringing the best of Google to our engagements and not just products.
Q: Why does Google as an internet/software company engage in chip development?
We have multiple services, each with billions of users spread across the globe. To provide services on such a scale, we have become one of the world's largest consumers of commercial silicon.
However, we sometimes see opportunities to optimize the chips for very large scale use. Two good examples would be The video encoding chips ours, which provide very high-performance and power-efficient hardware acceleration for video encoding, and the Tensor processing units ours, which provide best-in-class performance and carrier efficiency for machine learning training.
That. How does Google see the importance of chips for the high-tech industry and how will the lack of chips be solved?
It is known that the demand for chips continues to rise, but disruptions in the supply chain are affecting production, leading to severe chip shortages. While it is clear that access to computing resources is critical for both industry and research institutions, it is not clear how and when the shortage will be resolved.
One thing we're seeing is that customers who normally buy on-premise systems are finding that they can get their task done months and sometimes a year earlier by accessing the Google Cloud infrastructure. The sheer scale of the cloud, combined with the shared access model, makes it an ideal way to mitigate, if not solve, some of these supply chain issues.
Q: What is the role of artificial intelligence in your field?
Artificial intelligence is an extremely important capability for Google, and we use it in a number of innovative ways, from improving our services to lowering the environmental footprint of our data centers.
In HPC, in particular, artificial intelligence is interesting in a number of ways. First, some AI workloads, such as deep learning training, benefit from or even require the same kinds of high-performance infrastructure as those used for engineering and scientific computing. In other words, some see deep learning training as an HPC workload, and deploy it in HPC environments.
Artificial intelligence is also changing a wide range of disciplines, bringing new capabilities and improved productivity to workers in a wide range of industries. Scientists and engineers who have been primary HPC users are also using AI as a way to increase their productivity and speed time to insight.
Finally, there is active interest around the world in ways to use artificial intelligence to directly speed up the simulation process, thus making a further contribution to shortening the time to reach conclusions. Google Research Team published Recently promising results on accelerating computational fluid dynamics simulations, which are common in areas such as product design, data center designs and weather forecasting.
To register for the ChipEx2022 conference
More of the topic in Hayadan:
2 תגובות
High Performance Computing - high performance computing. There is no point in calling it a supercomputer - that is reserved for the most powerful computers in the world of this type because there are a lot of them.
I didn't count how many times the phrase "HPC" appears - but what does appear is 0 explanation of what the initials mean. You taught me already in the first year of engineering studies to write what each letter and initials mean. For idiots like me who are experts in other fields but want to understand developments in the world, this article did not provide me with anything.