The chip industry will need more software to catch up with Nvidia’s leading position in AI | Instant News

Anil Mankar, head of product development for AI chip startup BrainChip, introduced details of the company’s technology at the famous Linley Fall Processor conference on Tuesday. Linley Gwennap, the organizer of the conference, presented a case in which the entire industry needs more software functions to catch up with Nvidia’s huge leading position in the AI ​​field.

Brain chip.

The semiconductor industry is in the midst of a renaissance in chip design and performance improvement, but to catch up with graphics chip giant Nvidia, more software is needed to catch up.

of Linley Fall Processor ConferenceThis is a virtual event this week and next week, and is one of the main annual meeting events for promising young chip companies.

To kick off the curtain, the conference host Linley Gwennap (Linley Gwennap) has been a semiconductor analyst for 20 years. He delivered a keynote speech on Tuesday morning. He said that software is still all that has to challenge Nvidia. The company’s leading edge in manual processing is a stumbling block. intelligence.

“Software is the hardest word,” Gwennapp quipped, referring to the struggles of competitors.

Gwennap said: “The weak software stack hinders cloud vendors and start-up competitors.” He pointed out how the company does not support certain aspects of popular AI frameworks (such as TensorFlow), or certain AI applications for competing chips. It may not even compile correctly.

Nvidia dominates the training operations of building neural networks, leading more than a decade in a software system called CUDA. The luminous body in the AI ​​field that cooperates with Nvidia chips to build neural networks has repeatedly stated that the AI ​​field needs competition to break Nvidia’s control in the scientific field.


Hiren Majmudar, general manager of Global Foundries, introduced in detail the various innovations in equipment physics the company has brought to AI startups to give them performance advantages.

Global foundries

The use of AI has expanded from traditionally developed cloud computing data centers to embedded devices in cars and infrastructure. A division of chip equipment giant Applied Materials, such as suppliers such as Imagination and Think Silicon in the UK, is pushing the boundaries of low-power designs that can be used in power-constrained devices, such as battery-powered microcontroller products.

Since Nvidia announced its intention to acquire Arm Plc for $40 billion last month, the deal seems to have suddenly become more and more dangerous. Arm regards intellectual property as the core of all chips produced by all challengers in the chip industry.Therefore, Nvidia’s software is Ready to gain greater influence.

and also: Nvidia completely cleared the MLPerf artificial intelligence prediction benchmark

Companies hoping to seize market share from Nvidia must build their own software to replicate certain features of Nvidia, but have yet to provide products to bridge this gap.

He said: “There are some opening up efforts in progress, but they have not gained much traction.” “It depends on most companies to develop their own alternatives, which is why it takes so long.”


Despite the surge in competition from BrainChip and other companies, Nvidia’s chips (such as the A100) continue to lead in market share and overall performance.


There are many newly established companies at the conference, and each company has some innovations that can make Nvidia outperform its competitors in terms of raw performance for certain AI tasks.They include things like Tenstorrent, Brainchip and SiFive. Well-known companies including Intel and Google also participated.

The contract chip manufacturer Global Foundries is a contract chip manufacturer that produces the processor of countless chip companies, it illustrates the breadth and depth of innovation. The company described how to improve the physical performance of the underlying chip to create better transistors, which are the basic building blocks of all chips.

Despite these advancements, the latest AI benchmark results prove that Nvidia’s new and old components can still maintain impressive performance. MLPerf industry organization released on Wednesday.

Gwennap said that software is still the crux of the gap.

Gwennap said: “Intel and Qualcomm have a good software stack. They have invested heavily in software and have a lot of resources. We have seen some progress made by well-funded startups, and good progress has been made. But yes, this requires Time.” When asked what measures can make Nvidia substantively compete.


Linley Gwennap, the organizer of the conference, said that software is still the biggest weakness of companies competing with Nvidia.

Linley Group

Companies that sell their own computing systems, including Brain system with Picture core, A brand-new software program has been established to optimize the way the neural network chip handles neural networks. However, the efforts of these individuals may balance the use of new chips.

In contrast, Nvidia’s CUDA provides a consistent platform for artificial intelligence designers so that they can concentrate on their work.

Gwennap suggested that a combination of different efforts may be needed to challenge Nvidia reliably.

Gwennap said: “I think there will be integration at some point to help some of these companies increase software investment.”

Gwennap told the conference that the conference has been fifteen years old and there are more than 1,000 attendees this year. ZDNetThe event was held in the hotel banquet hall in the Silicon Valley area, more than three times the number in previous years.

The conference will continue on Wednesday and Thursday, and there will be a second speech on Tuesday, Wednesday and Thursday, including the Google keynote on Tuesday.

image source

to request modification Contact us at Here or [email protected]