on-chip-accelerated-ai-processor-announced-by-ibm

On-Chip Accelerated AI Processor Announced by IBM

IBM announced its upcoming IBM Telum Processor at the annual Hot Chips conference, the first processor of the company to contain on-chip acceleration for AI inferencing during transactions. Designed to introduce deep learning inference to the enterprise-scale to prevent fraud in real-time.

The IBM Telum Processor allows applications to run effectively in a local environment, eliminating the need for significant memory and data movement capabilities to process inferencing. The breakthrough of on-chip hardware acceleration enables enterprises to process high-volume inferencing for real-time sensitive transactions without external AI solutions, helping customers gain business insights at scale across banking, finance, trading, insurance applications, and customer interactions. This greatly increases performance and allows clients to build and train individual AI models.

Currently used detection techniques are applied after a transaction occurs. This process requires significantly more time and computing power, especially with analysis and detection being carried out far away from critical data. This hinders real-time detection, impeding the efficiency of prevention.

The Federal Trade Commission’s 2020 Consumer Sentinel Network Databook reveals that in 2020, consumers reported more than $3.3 billion loss to fraud, up from $1.8 billion in 2019. The new IBM Telum Processor opens up the possibility for fraud prevention instead of fraud detection.

The chip’s centralized design opens up the full power of the AI processor for a range of AI-specific processes, like financial services workloads, fraud detection, loan processing, anti-money laundering, and risk analysis. Existing rules-based fraud detection will be enhanced by accelerated deep learning processes, facilitating credit approval, improving customer service and profitability, and helping identify failing trades or transactions.

Telum’s innovative design features hardware and software co-development and easy integration into leading software frameworks. With 8 processor cores, the chip runs with more than 5GHz clock frequency. It contains 22 billion transistors and 19 miles of wire on 17 metal layers.

Watch the video below:

Roboticulized digests/handpicks the latest news about the artificial intelligence/machine learning industry and serves them to you daily. We provide you with the latest news and press releases straight from the AI/ML industry.
Hipther

FREE
VIEW