India powers Intel’s GPU rollout plan

Intel’s Xe generation of GPU (Graphics Processing Unit), which targets both data centres and PC gaming, has been the talk of the town in the geek world for a while now. Speculation on product release dates, likely performance and reported stagnation, mimics the anticipation and excitement ahead of a movie release.

Raja M Koduri, Senior Vice-President, Chief Architect and GM of Architecture, Graphics, and Software of Intel Corporation, who was in Hyderabad recently, refuses to give an exact timeline for the GPU product roll-out, but indicates that it would be released sometime in 2020.

Asked whether it is slated for a summer release, Raja, the chief architect behind Intel’s GPU push smiles disapprovingly. He refuses to provide any hints on the release date.

But Raj’s recent tweet with a team of engineers from Bengaluru, who are working on the XE HP, has kept interest alive on the new product line.

“It’s all Xe HP - the team here in @intel Bangalore celebrated crossing a significant milestone, on a journey to what would easily be the largest silicon designed in India and amongst the largest anywhere,” he said in the tweet.

Raj is confident about the product. “The team calls it the baap of all (the father of all),” he said in his early December tweet.

The chip major is going to extend its GPU research work to its recently opened Hyderabad centre, too. “We're going to do GPU designs, new GPU designs. We are also going to do software for open source,” he says.

“As we grow the engineering team here, we will work on what we call exascale computing. Besides the GPU, there are a lot of other chips and components required to put an exascale computer together. But we will start off with GPUs and also some very important software,” he says.

(Exascale computing does at least one quintillion (billion billion) calculations per second.)

GPU is beyond games

Regarded as a thought leader in the GPU computing space, Raj says the scope for GPUs goes much beyond the gaming space.

“GPUs started off with video games some 20 years ago. We will see more immersive computing. Some video games look almost real. In the next 20 years, we will see more immersive experiences. What we have now in virtual reality and augmented reality, is in the nascent stages. The whole notion of immersive computing is going to change completely,” he says.

He says GPUs have enormous computational capacity, which is required by AI. The next 10 years will see huge growth in immersive computing and AI, and GPUs will continue to play a major role.

“Our ability to analyse the data in real-time is very poor. On a scale of 1-10, we are at point-zero-zero-one at this point,” says Raj.

Challenges

“We need to analyse data in real-time to reap the benefits of the enormous data that we generate every minute. Your data needs to sit near you and not travel too far to get processed. You need faster networks to transfer the data. Technologies such as 5G will help in this regard,” he points out.

“Though everyone is talking about data and AI, we are at the initial stages and the problems that need to be solved begin right from where the data is collected, transferred and analysed,” he says.

“For one, you have a surveillance camera at some place and there are not enough machines to process such huge amounts of data. There is no use in analysing data after a week or a month, if you want actionable insights into events.

“You can respond in real-time (if you are able to analyse data in real-time) and can hopefully save some lives,” he says.

The challenge, he says, is the availability of processing power at the local level. “At present, not all countries and big companies have it. It needs to be available to everybody, to governments so that the computing needs of your region are met by computers in your region,” he observes.

“If the data needs to be transferred across oceans for processing, that's not real time. Your data needs to sit close to your where it is generated,” he says.

The second challenge is latency, how fast the data moves from one point to another. “With regard to AI, we are at the very beginning stages of AI. What to do with the data, how to understand it, label it and how to remove noise from the data,” he explains.

Market size

He says the addressable market for Intel changes significantly if one looks at the unfolding opportunities.

“Intel used to account for almost 90 per cent market share of the PC and data centre (business) segments. But if you look at the competitive landscape of the future, we are only 25 per cent of the market opportunity,” he contends.

“That means that when you are 90 per cent of a market, you don't have competition. When we are only 25 per cent of the whole growth opportunity, everybody is competition. We have competition everywhere. So we take both start-ups and big companies very seriously,” he says, when asked how Intel views competition from start-ups.

COMMENT NOW