Big tech companies are racing to develop custom chips that will greatly increase the efficiency and reduce costs of artificial intelligence (AI).
The meta is Latest generation custom computer chips To enhance AI capabilities and minimize dependence on external suppliers such as Nvidia.This announcement is as follows What Intel revealed The development of an improved AI “accelerator” comes as competitors such as Google are also working on developing AI chips in-house. Experts said AI chips could facilitate commercial applications.
“From a business perspective, it lowers the bar for training per-customer, per-task models and moves us away from simply leveraging APIs from large language model providers for specialized, high-security use cases. .” Amrit JassarCo-founder and Chief Technology Officer egniteThe company, which develops AI-powered software for businesses, told PYMNTS.
Custom chips could reduce AI costs for companies.The current cost of integrating generative AI into your business can vary greatlySoftware developer Itrex says it can cost anywhere from a few hundred dollars a month to hundreds of thousands of dollars for a custom solution based on a fine-tuned open source model. Last year, his CEO at NVIDIA, Jensen Huang, Said We believe that custom AI chips have the potential to reduce costs associated with generative AI.
AI super chip
The AI chip market is highly competitive as companies vie for a share of the rapidly expanding industry. On Tuesday (April 9), Intel announced a new AI chip. gaudi 3chip makers are rushing to develop semiconductors that can train and deploy large-scale AI models like those that power OpenAI's ChatGPT.
Intel says the Gaudi 3 chip is more than twice as power efficient as Nvidia's H100 GPU and can process AI models 1.5 times faster. It is available in a variety of configurations, including a set of eight Gaudi 3 chips on a single motherboard, or as a standalone card that can be integrated into existing systems.
The chip was tested on models such as Meta's open source Llama. Intel highlighted that Gaudi 3 is excellent for training and deploying models such as: stable diffusion Image generation and Whisper of OpenAI For voice recognition.
Meta announced Wednesday (April 10) the latest version of its custom chip designed specifically for artificial intelligence tasks. blog post. Newer chip iterations have shown significant performance improvements over previous chips. The tech giant says it supports Meta's algorithms for ranking and recommending ads on platforms like Facebook and Instagram.
“For users, we expect this change to manifest in increased personalization, reduced response latency, and improved local processing for products such as Meta Glass,” Jasal said.
PC manufacturers are also starting to use dedicated AI chips in their products.According to Bloomberg report Apple on Thursday (April 11) is nearing the start of production of its M4 computer processor with AI processing capabilities. The company plans to equip all of its Mac models with these new processors.
Benefits of AI-dedicated processors
The main advantage of AI-specific chips is economical. Rodolfo RossiniCEO of veiltold PYMNTS that it is a company that uses reversible computing to create low-energy silicon chips. He said it costs Nvidia $500 in materials to make the chip, which sells for $30,000, and there are still long lines to get an allocation.
“Beyond a certain scale, it makes sense for vendors to manufacture their own chips,” he says. “What’s more, it can be customized for specific workloads and algorithms.”
Meta is in the unique position of having most of its products free, so it needs to reduce its AI costs in order to deploy powerful tools such as large-scale language models to its user base, Rosini said. said.
“This is why other AI tools today require around $25 a month to operate,” he added. “By having its own infrastructure, Meta could potentially deploy this for free.”
Custom AI chips offer significant benefits such as enhanced privacy and unique data management, allowing businesses to steer their own AI futures. Michal OgrodekChief Technology Officer and Co-Founder of a Generative AI Company ivy eyehe told PYMNTS.
“This space is becoming increasingly crowded, and we expect to see a proliferation of competitors as AI solutions and providers rapidly proliferate,” he added. “Slow and steady typically wins the race, but Nvidia's and soon Intel's results show there's little time to lose. Leading companies are tackling the AI frenzy from all sides. .”
