Contrary to popular belief, high-performance computing (HPC) is not reserved for big-ticket government projects or scientific undertakings. Although it has enabled major strides in the field of genomics, for example, HPC is a practical enterprise solution than many realise.

A common concern raised by enterprises regarding HPC is cost, and its upkeep entails a lot of money. However, new developments in the field prove that maintaining and operating an HPC infrastructure is much more cost-efficient now. 

But how exactly can enterprises wield the power of HPC, and what preparations are needed in terms of infrastructure and human resources?

To address these concerns, Sinisa Nikolic, Director Asia Pacific, High Performance Computing and AI at Lenovo, discussed the current HPC landscape and where it is headed during a fireside chat entitled “The Mainstreaming of High Performance Computing,” organised by Jicara Media and sponsored by Lenovo.

To start off, Sinisa shared how artificial intelligence (AI) workloads in manufacturing leverages HPC.

“You’re looking at medical devices or chemicals, whatever it is, but having these types of cameras, as an example, on every manufacturing line looking at quality control, you can actually take the human eye out. Not that you want to remove people from the workforce, but to augment what they do allows you to be more efficient,” he said.

Meanwhile for the retail industry, HPC can also help businesses predict trends at a faster pace.

“We’re working on a project at the moment with a large retailer. During the night, they have a robot, or a number of robots at shutdown, that go up and down the aisles, scan the shelves and look for missing items and other things. But as the robots are doing that, they are picking up a lot of other data, which are the areas that are mostly affected with this stock shift, etc. So it’s a three-dimensional effective model of data that goes in a matrix of data that goes in to help them understand ‘Where should we actually place stock? What do we have to restock? What do we get rid of?,” Sinisa said.

The question of cost

Among the areas of hesitancy when it comes to HPC adoption is how much it will cost business organisations, as compared to using cloud, for example. According to Sinisa, the cost issue runs beyond financial expenditure, although the OPEX versus CAPEX debate is a valid concern among decision-makers.

“One of the major criteria for clients moving to cloud is the differential from CAPEX. I don’t want to have a capital asset on the floor because I don’t want to sweat the asset down for three years. I don’t want the depreciation,” Sinisa relayed.

“There are apps or services that will run in the cloud just fine. But you have potential data locality issues which will hamper that. A cloud services provider that is not located in Singapore, as an example, may not be able to run your workload anywhere in the world, because of this data locality requirement. In fact, in every country, when it comes to sensitive data, there are locality laws,” he noted.

In defence of cloud, Sinisa said that because it’s subscription-based, it has the scalability that most enterprise customers demand. However, there are several setbacks that accompany cloud subscriptions.

“It doesn’t answer the question on data locality. It doesn’t answer the question on availability, and it doesn’t answer the question on how quickly you can actually move services to it, nor can you control ‘mission/expense creep’,” he said.

Earlier this year, Lenovo unveiled its TruScale High Performance Computing as a Service (HPCaaS), which offers cloud-like computing experience done on-prem. 

Among the features of Lenovo TruScale HPCaaS are:

• Pay for what you consume with no hidden costs.
• Easily scale HPC clusters by removing resource limitations.
• Run applications at scale without architectural bottlenecks.
• Translate capital investments into operational expenses, avoiding long and intensive budget cycles and procurement delays.
• Manage budgets and visualise consumption and billing through Lenovo’s TruScale Portal.
• Access to specialised hardware.

“When you’re running your business, when you’re running your workloads, you can actually do that ‘as a service’. And that’s a financial mechanism that allows you to actually have on-premises infrastructure, whilst treating it as OPEX,” Sinisa explained.

Beginning the journey

Apart from cost concerns, most companies are at a loss where to start when it comes to AI and the technologies associated with it. As an example, will it be compatible with the company’s existing systems? What skill level of in-house talent is required to operate it? What skills does one need to build data models? What are the KPIs? 

According to Sinisa, when one starts their AI journey, the biggest thing to consider is their ultimate goal. “What is a problem that we are trying to solve? Companies should work with their data scientists. A data scientist uses analytical, statistical, and programming skills to collect large data sets. They develop data-driven solutions tailored toward the needs of an organisation, i.e. the KPIs that I mentioned earlier,” he said.

“Another factor to consider when choosing a vendor for the AI journey should be whether or not your vendor has data scientists available to advise on technical concerns and strategies,” Sinisa continued. “This way, there are clear links and collaboration between vendor and client.”

“We have five facilities around the world that are called Centres of Competence for AI and Data Sciences. They’re actually available for our software partners and clients to sandbox and run proofs of concepts (POC). We don’t actually charge for that. I would suggest our clients use this data sciences team as consultants to work through both a POC and use cases,” he said.

Sinisa also emphasised the importance of extracting value from data, which is one of the benefits of HPC for enterprises, regardless of industry.

“Data is just data. A number of years ago, you wanted to turn data into information. Now with AI and HPC, that has turned data into action.” Sinisa noted.

“Data intelligence based on your business, what you do, your supply chain-related issues, how you get to market— I would say that that is the best place for a business to look at, these areas would derive the best use cases and ROI using AI,” he added.

In the near future, Sinisa predicted that how companies leverage data will eventually decide whether they survive in the market or not.

“Looking at your manufacturing line for example: How will design changes in the ‘widget’ I am manufacturing affect the process? Will I need different materials, cost changes, supply chain needs, and environmental effects? One then uses these data points in an AI mathematical model, which will give the business a predictive position of what may actually happen with the changes to that ‘widget’,” he explained.

This output, said Sinisa, will allow business to take action on that concern.

“I think as we go further forward in the world over time. Those organisations that don’t look towards AI and HPC over the next four or five years will struggle,” he remarked.