Intel is launching Israeli Bana Labs chips – and will compete with Anvidia in the field of artificial intelligence
The value of the chip market for artificial intelligence is expected to reach $ 200 billion by 2030, and Hub Security wants to grow in an area where Anvidia controls 80% of it. Intel has introduced two dedicated new chips, but alongside the intention to develop in the field, it remains due to a significant Achilles heel in the software world
Intel today (Tuesday) launched two new chips designed to improve its positioning in the AI field. The chips were developed by Habana Labs, which was acquired about three years ago by the chip giant.
The background to Intel’s current attack on Anvidia stems from the simple fact that while the first dominates the market of computer processors and data centers, Anvidia completely dominates the chip market for AI (artificial intelligence).
This is a huge market that is expected to grow to about $ 200 billion by 2030, according to data from the research company Allied Market Research and published in the Wall Street Journal about a month ago. The value of the market about two years ago was about $ 8 billion. Anvidia controls about 80% of this market and Intel wants to increase the company’s footprint in the field.
And yet it is a very difficult task for which Intel has mobilized all the talent it has in the field. The Bana Labs is the secret weapon she has acquired for herself to advance her abilities. The unit, currently headed by Eitan Medina, Chief Operating Officer at Bana Labs Laboratories has introduced two chips, the first Gaudi2 is actually an improvement on the first generation of artificial intelligence chip that the company developed even before the acquisition. This is a 7-nanometer chip compared to 16 nanometers in the first generation and Intel showed in a demo presented to journalists that the chip is capable of 2 times higher performance than the fastest Anvidia chip today – however it is not clear if its cost is equivalent or a much more expensive product. “Compared to the A100 graphics processor, which is applied in the same manufacturing process and about the same size of piece of silicon, Gaudi2 delivers leading training performance clearly as demonstrated in the same comparison of key workloads,” said Ethan Medina, chief operating officer at Bana Labs Labs. “This in-depth learning acceleration architecture is more efficient, and backed by a strong roadmap.”
The second chip is called Greco and is designed to draw conclusions for artificial intelligence and strengthen relationships with developers to help them “migrate” existing graphics-based models to Gaudi’s hardware platform to create a new SynapseAI software package. In other words, it is a processor that aims to allow Anvidia customers who want to switch to Intel to transfer their workloads to the latest products. “Developers and data scientists often tend to work with the software infrastructure of Anvidia processors. At Habana Labs they have identified this barrier, and in order to allow beyond the Habana Labs processors have taken care to create new software,” according to Intel.
Amazon as a strategic customer
To promote the launch of the processors, Intel even sent Israel’s vice president and CEO of the company’s data center and artificial intelligence group, Sandra Rivera, who provided the numbers behind the current announcement to the Israeli press. “Artificial intelligence is growing in all areas – The artificial intelligence industry will grow to tens of billions of dollars by the middle of the decade, with the largest growth occurring in an artificial intelligence-driven sector. “We anticipate that the size of the market that Intel will be able to address with artificial intelligence-based solutions will be more than $ 40 billion by 2026,” according to Rivera.
And yet it is a small part of the market which as mentioned is mostly dominated by Anvidia chips. Not for nothing has its market capitalization soared to about $ 442 billion compared to a market capitalization of only about $ 176 billion for Intel today. However, Intel explains that quite a few of the workloads in the field of AI are run on Xeon processors – which are used for servers in data centers. And yet, chips for general use cannot replace chips designed for artificial intelligence calculations over time. As the market progresses towards the use of artificial intelligence for the operation of various services such as language recognition – It is found in a variety of applications, especially in the voice assistants of big-tech companies – or image recognition and analysis of big data to generate business insights, so the need for chips that can provide computing capabilities dedicated to the field is rising. Intel understands this meaning well, not for nothing has the company re-entered the field of graphics chips that form the basis for AI calculations with the introduction of its first product about two months ago.
“When it comes to artificial intelligence, many choose to focus on deep learning (DL) training and graphics processor performance. Graphic processors get a lot of attention because training scenarios tend to be parallel and massive. But this is only part of the landscape of artificial intelligence. “Machine learning classics and in-depth learning models with a small to medium level of complexity,” Rivera explains Intel’s approach. According to her, these are requirements that are in the capabilities of Xeon processors.
It is difficult to say that the task that Intel has taken on will be easy or simple. Anvidia is already a kind of standard in the world of artificial intelligence thanks to the many years of investment in the field compared to Intel, which somewhat abandoned the race until the acquisition of the Bana Labs. The latter was acquired with 180 employees and today, according to Intel, it stands at about 1,000 employees and continues to recruit all the time. But from a business standpoint and even though it is trying to show a rapid growth in its business these are still probably not very impressive. So far Intel has managed to get Amazon as a strategic customer – But mainly due to the fact that it was a customer of the Bana Labs before the purchase – a result of the business capabilities of Avigdor Wilenz from its founders. Beyond Amazon, Habana Labs has only Mobilai, which is also a company of Intel, although it is expected to go public soon. Other companies that appear on Intel’s customer list mainly include hardware manufacturers that offer the Bana processors in their products.
The last point where Intel will need to improve is the world of software. This is her old Achilles heel and so far she has not been able to leverage her software platforms, which is a barrier to the transition of developers from the worlds of Anvidia to Hub Security products. To promote the field that is currently worth only about $ 100 million for it, it is promoting open source solutions such as oneAPI. “Hardware is getting better and better, but it’s just one part of the puzzle. Software is a critical part of our artificial intelligence strategy,” says Rivera. “We’ve received feedback from analysts that our AI portfolio is ‘just a bag of parts’, but in reality, we’ve focused heavily on mapping to ensure developer compliance and a clear understanding of how software connects the dots. Intel continues to invest in open source software solutions like oneAPI, and industry frameworks like TensorFlow and “PyTorch to strengthen collaboration between developers, academia and the wider ecosystem,” Rivera said.