As enterprises and startups in the country are staring at long waiting period to get supplies of graphics processing units (GPUs) from companies like NVIDIA to test and run their AI models, Intel feels they can use existing infrastructure such as its Xeon processors and Gaudi chipsets at lesser costs and better efficiency. Santhosh Viswanathan, vice president and managing director, India, spoke to Jain Grover about Intel’s plan on India’s AI mission, go-to-market strategy for AI products, and offerings for startups. Excerpts:
Q. Do you think the longer waiting period of GPUs will have an impact on Indian companies move towards using AI models?
A. It is about figuring out what problems you are trying to solve. If I have an AI model that has 15-30 billion type parameters, existing infrastructure may be able to run it. There are benefits that a GPU brings in and there is an alternative as well.The consumers should not make this as one size fits all model. They should evaluate their datasets, parameters, and then evaluate if they need CPU to get started, then at some stage if they need a GPU. They might evaluate an accelerator like Gaudi that gives similar performance at a lower cost or a lower power consumption.
Q. For the India market, what is your go-to-market strategy as there is a race for procuring NVIDIA GPUs?
A. First, consumers need not wait for 16 weeks (to procure GPUs), we can give Intel’s products much faster. Our go-to-market is to make customers realise that your existing infrastructure can run the models, but we need to understand what are you trying to run, how many parameters, what are the use cases look like, so that we can support. We have Intel Developer Cloud where customers can come and try out, run their models before going ahead.
Q.How much cost can be lowered with Intel’s solutions compared to NVIDIA GPU?
A. If you look at the Xeon cost, it starts at hundreds of dollars versus tens of thousands of dollars that a GPU is available for today. The cost is exponentially lower for using accelerators. Xeon is obviously there everywhere. So it’s going to be the lowest cost point for a company to start off. We recently announced Gaudi 3. It is about 50% better on inference performance compared to whatever is available in the market today from a GPU point of view. It’s also 40% lower in terms of power consumption.
Q. As part of IndiaAI mission, the government is looking to procure GPUs for startups. Do you plan to pitch your solutions?
A. The government’s plan on the AI mission is a great step in not just building infrastructure but also datasets, which are the building blocks. I hope it’s not just one single architecture conversation (which the government is thinking about). We absolutely want to go back and participate in it. Alternatives should be there from a Xeon, a Gaudi to a GPU. Whatever the breadth of infrastructure, it should be available. Our intent is to work closely with the government as well as the ecosystem partners so that there are enough options that are available.
