Super Micro Computer, Inc., one of the largest world manufacturers of solutions for AI/ml, HPC, Cloud, Storage and 5G/EDGE, has recently held in Madrid the Supermicro Innovate event! where he has presented his latest news. This room brings together the partners and customers of Super Micro de Europa, the Middle East and Africa with representatives of the different areas of the manufacturer to be able to know the possibilities they offer developed together with AMD, Intel or NVIDIA. Especially highlight the Data Center Building Block Solutions, with which Supermicro facilitates the implementation of modular IT infrastructure for new generation data centers with the level of sophistication necessary in a simple and scalable way.
During the Supermicro Innovate event! We had the opportunity to talk with Michael McNnerney, SVP Marketing & Network Security of the company, who shared with us his vision of the current state of technology. The conversation focused on the server and data centers market but always under the Undergrounds of the AI that is transforming everything.
(MCPRO) What news have you presented in Supermicro Innovate!?
(Supermicro) We are a computer company. We build storage, we build networks, we build servers. We do all these things. And now one hundred percent of the conversation is about AI. So, the storage is over ia, the network is over ia, everything is about ia. We have presented in this event new AI training servers and the extension of the Portfolio of EDGE solutions, with optimized systems for distributed loads of artificial intelligence, equipped with Intel Xeon 6 SOC processors and NVIDIA Jetson Orin NX. These systems are designed to offer high performance and energy efficiency solutions in environments where energy, cost and space are key factors. In addition, Supermicro also launches a shallow 1U system with Nvidia Grace C1, designed to integrate into the radio access network (RAN).
The key is to promote ecosystems capable of offering real solutions. For this we fully trust our partners, both those who accompany us here and those who are part of our global network. It is not only about collaborating with Supermicro, but also generating synergies among the different actors: software partners, distributors and service providers. The question is how to get those ecosystems and networks to integrate effectively to respond to current challenges. I believe that this is the most inspiring aspect in order to boost innovation.
(MCPRO) Indeed, data storage is the basis of AI and companies are giving much more importance to their control and processing. Partly for the legal requirements on privacy and data protection but also as a business and competitiveness engine. What needs are seeing their clients about it?
(Supermicro) The fear of losing control over a company’s data assets is a very real concern. And the way in which a key difference between some companies and others is really made, especially in data management and migration to the cloud. Many organizations only consider the financial aspect or speed that this transition can contribute, but that if it is not done well it can imply risks in processes accelerated. Today, one of the main problems is the sovereignty of data, especially in Europe, where different countries have different regulations.
In addition, to be effective with artificial intelligence, it is essential to understand that the closer the data are and the better organized, the greater value can be extracted from them. Supermicro has been a company traditionally focused on the calculation, but now this area is closely linked to storage within the network, an area that has always been present, but now charges even more relevance.
Artificial intelligence is based on data training, which puts several questions on the table: where are that data? How are they structured? In this event, we have numerous partners, and three quarters of them are specialists in storage because they are developing stepped storage architectures. For example, there are data stored in large deposits or data centers, but a high performance cache is also required to feed GPUS, since they cannot wait for prolonged times to access data stored in traditional tapes or discs.
Therefore, the data is the essential fuel for AI training, which represents a maximum priority. It should be noted that, within the scope of storage, one of the great achievements has been to maintain a neutral position, with our own hardware and relying on reference partners that provide the necessary storage software.
(MCPRO) How are storage technologies evolving to solve the challenges posed by AI?
(Supermicro) The question is: How to get everything to be loaded in the memory of the system as quickly as possible so that it can be efficiently prosecuted by the GPUS? Much of the current storage discussions revolve around how to build a complete channeling that covers from the historical data warehouse to the high -performance cacheo areas.
This implies structuring and preparing the data, and moving them to rapid storage as NVME devices. Although many of these technologies have existed in the field of high performance computing (HPC) for years, they are now experiencing accelerated growth and much greater demand.
In this context, the main storage approach is to optimize all data channeling, not only in having fast NVME devices. The really relevant is how information is transferred from storage to memory and how that tail transfer is organized. Many software providers are working on optimizing these processes and designing such data channeling efficiently.
(MCPRO) And, in the same line, how are servers evolving in data centers to process the huge amounts of data required in the training and inference processes associated with AI?
(Supermicro) We are attending a movement towards systems with greater density and greater computational power. And this is only possible if liquid cooling systems are used. We consider that for at least a couple of generations we will continue to use air cooling. We believe that during some generations, especially in areas such as Edge Computing, air cooling will remain present.
Air cooling will continue to be an option for a long time, because some technologies simply do not disappear; There is still the need for air cooling solutions when liquid cooling is not viable. We believe that this technology will remain and will be relevant.
The transition to liquid cooling is already happening. It’s like cars: before, Volkswagen used air cooling and was impressive. However, liquid cooling is a more efficient mechanism to dissipate heat, facilitates its containment because heat is transferred to the liquid, avoiding the need to cool all the air of the data center.
At the same time, many companies are already prepared for liquid cooling, which is a more efficient solution and, without a doubt, will be imposed progressively, although this transition will be carried out in a generational way. In this event we show our solutions to install equipment with liquid cooling in data centers that are not prepared for it. These are modules, air refrigerated, which contain all the infrastructure inside to support water refrigerated servers. In this way, the transition is facilitated since it allows to install state -generation and high performance processing equipment in places that are not yet adapted.
The great clusters, which are immense, will be refrigerated mainly by liquid. As for liquid cooling for large -scale training, these are high density solutions and more efficient materials to control heat. Last year we manufacture about fifty percent of the fluid refrigerated racks worldwide, specifically for large AI data centers.
(MCPRO) Supermicro has evolved from being a manufacturer of motherboards, which later entered the market of barking to be a reference today in the development of data centers in collaboration with the great leaders in processing manufacturing. How was that transformation and how would the current position of Supermicro define?
(Supermicro) Indeed, I consider that it is an important change. In the storage area we have managed to position ourselves as industry referents, consolidating us as platform suppliers. Storage solution manufacturers face a complex challenge: they need to meet multiple demands, from file systems with large volumes of discs to cache systems and other fundamental elements. The question that arises is: where can you get all these solutions? We offer that range of possibilities, and so we are becoming true market catalysts. Our goal is to make available to the sector and customers an open platform. More than half of the partners present here belong to the storage field. We celebrated a meeting this summer in which twenty -two partners participated; We even surprised the response, because we received requests from many more interested who wanted to join. In short, demand and growth are evident.
(MCPRO) Thank you very much for your time. It has been a pleasure.
