Industry News

Microsoft Developed AI Chip to Reduce the Cost of Running Generative Artificial Intelligence

Update time : 2023-05-03 14:55:41

Dylan-Patel, principal analyst at global semiconductor research firm SemiAnalysis, recently said that OpenAI could spend up to $700,000 per day to run ChatGPT because it runs on expensive computing infrastructure.

Whether it's writing a cover letter, generating a lesson plan, helping users optimize their profiles, or analyzing things based on facts or assumptions, ChatGPT requires a lot of computing power to provide feedback based on user input, which comes from expensive servers, Dylan-Patel said.
Both Dylan-Patel and his colleague Afzal-Ahmad argue that while it may cost hundreds of millions of dollars to train the big language model behind ChatGPT, its operating costs or the content production behind it will be much higher, even with any reasonable deployment size that far exceeds its training costs.
Microsoft is rumored to be developing an AI chip codenamed "Athena" to reduce the cost of running generative AI models. The report says the project has been in production since 2019 and is available for testing by a small group of Microsoft and OpenAI employees.
Microsoft previously reached a $1 billion investment agreement with OpenAI that requires OpenAI to run its models only on Microsoft's Azure cloud servers. This follows news that shortages have led Microsoft to ration GPUs for some internal teams, and that NVIDIA's processors sell for a premium, so Microsoft expects to run them more cheaply for the same workload.
In addition to powerful performance, Nvidia's chips have significant software advantages, with most AI workloads designed for them and decades of developer experience. Microsoft currently has about 300-plus employees working on the chip.
Sources said the chip could be released as early as next year for internal use by Microsoft and OpenAI, to which Microsoft did not officially respond, but whether it will also be used by Azure customers is still under discussion. Google has developed its own line of AI chips, TPU, and is currently the only competitor chip developing LLM, while Amazon has its own alternative product line, Trainium.

Previous : clearly needs to improve its AI development tools and processes.

Next : Sources say AMD has transferred some orders for 4nm CPU chips from TSMC to Samsung

LSM115JE3/TR13 Microchip - Schottky Diodes

Apr .02.2025

The Microchip LSM115JE3/TR13 is a 1 Amp Schottky rectifier in the DO - 214BA package. It has a 15V working and repetitive peak reverse voltage. With features like a guard ring for reverse protection, low power loss, and high efficiency, it can handle a 50

LT1767EMS8-1.8#PBF Analog Devices Inc. | Power Management (PMIC)

Mar .31.2025

The Analog Devices Inc LT1767EMS8 - 1.8#PBF is a monolithic buck switching regulator. Operating at 1.25MHz, it has a 1.5A switch in an 8 - pin MSOP package. With a 3V - 25V input range, it offers a fixed 1.8V output with 2% tolerance. Its high - efficienc

DS1746-70+ Analog Devices / Maxim Integrated-Real Time Clocks ICS

Mar .28.2025

The DS1746-70+ is a highly advanced and fully functional real-time clock/calendar and nonvolatile static RAM. With its year-2000-compliant design, this product ensures accurate timekeeping and data storage without any risk of Y2K-related issues. The byte-

MBRS540T3G by onsemi Schottky Rectifiers

Mar .26.2025

Onsemi MBRS540T3G is a surface - mounted Schottky power rectifier. It features a compact J - bent pin package, ideal for space - saving, high - density PCB assembly. With a stable oxide - passivated chip, it ensures reliability and low leakage. It's RoHS

About Us

About Us What we do Quality People

Products

CHIP/IC Active devices Passive device Components OTHER

China electronic components supplier----Wah Luen Shing Electronic.----www.wlschip.com

Leave a message