Tachyum Details AI Training Using Its Breakthrough Super-Sparsity Technology

The AI training models leveraged are all fully functional, ensuring Prodigy remains on the leading edge of the industry.

Metadata
Provider: 
Business Wire, Inc.Business Wire, Inc.
Retrieved on: 
Tuesday, October 18, 2022
Information
Tags: 
Technology, Semiconductor, Security, Professional Services, Software, Hardware, Data Analytics, Electronic Design Automation, Artificial Intelligence, Memory, Deep learning, Efficiency, Carbon, CPU, OPEX, CAPEX, HPC, Industry, FP8, CEO, Economics, TPU, Prodigy, GPU, GPGPU, Medical device, Video game console, Computer data storage, AI
Geotags: 
UNITED STATES, NORTH AMERICA, NEVADA
⚠ generated by AI ↴
Organisation: 
TACHYUM
Summary
Key Points: 
  • The AI training models leveraged are all fully functional, ensuring Prodigy remains on the leading edge of the industry.
  • Among the features described in the latest Tachyum white paper are AI training techniques that include:
    Quantization Aware Training Used with low-precision data types for training deep neural network models.
  • Prodigy processors allow deep neural network training using just 8-bit floating point (FP8) representations for all matrix and tensor computations weights, activations, errors, and gradients.
  • Per-Layer Gradient Scaling Tachyum employs a per-layer gradient scale instead of a global loss scale for the training of deep neural networks using FP8.


Tachyum™ today released a white paper that examines how its Prodigy processor addresses trends in AI, enabling deep learning workloads that are environmentally friendly by achieving less energy consumption and carbon emissions. The AI training models leveraged are all fully functional, ensuring Prodigy remains on the leading edge of the industry.

Prodigy, the world’s first Universal Processor, delivers a revolutionary new architecture that unifies the functionality of CPU, GPGPU, and TPU into a single monolithic device, enabling Prodigy processors to address the high demands of cloud and HPC/AI workloads, without expensive and power-hungry accelerators, by using a simple homogenous software model that is aligned with software composability and dynamic reallocation of server resources to maximize utilization.

Among the features described in the latest Tachyum white paper are AI training techniques that include:

  • Quantization Aware Training – Used with low-precision data types for training deep neural network models. Prodigy processors allow deep neural network training using just 8-bit floating point (FP8) representations for all matrix and tensor computations – weights, activations, errors, and gradients. This leads to dramatic improvements in computational efficiency and energy efficiency without impacting model convergence and accuracy, lessening memory transfer times and enabling the use of fast low-precision computing operations.
  • Per-Layer Gradient Scaling – Tachyum employs a per-layer gradient scale instead of a global loss scale for the training of deep neural networks using FP8.
  • Sparsity – Tachyum’s Prodigy Processor provides native support for block structure sparsity and super-sparsity. It supports sparse matrix multiplication optimized for compressed neural network models, thus reducing memory and computation requirements.
  • Results for Image Classification and Instance Segmentation – the company trains a wide range of computer vision models, object detection and instance segmentation models using the FP8 format.

“With our recently released architectural white paper detailing the Prodigy processor, we have received tremendous interest and were asked for more details about our AI features,” said Dr. Radoslav Danilak, founder and CEO of Tachyum. “This latest deep dive provides insight into all the AI training models that can run on Prodigy and how each of those approaches allows system developers to tackle increasingly complex deep learning calculations, delivering an order of magnitude better performance than is currently available from competitors.”

Prodigy delivers unprecedented data center performance, power, and economics, reducing CAPEX and OPEX significantly. Because of its utility for both high-performance and line-of-business applications, Prodigy-powered data center servers can seamlessly and dynamically switch between workloads, eliminating the need for expensive dedicated AI hardware and dramatically increasing server utilization. Tachyum's Prodigy delivers performance up to 4x that of the highest performing x86 processors (for cloud workloads) and up to 3x that of the highest performing GPU for HPC and 6x for AI applications.

Those interested in learning more about Tachyum’s FP8 can download the Tachyum Prodigy on the Leading Edge of AI Industry Trends White Paper at https://www.tachyum.com/resources/whitepapers/2022/10/18/tachyum-prodigy-artificial-intelligence/.

Follow Tachyum

https://twitter.com/tachyum

https://www.linkedin.com/company/tachyum

https://www.facebook.com/Tachyum/

About Tachyum

Tachyum is transforming AI, HPC, public and private cloud data center markets with its recently launched flagship product. Prodigy, the world’s first Universal Processor, unifies the functionality of a CPU, a GPU, and a TPU into a single processor that delivers industry-leading performance, cost, and power efficiency for both specialty and general-purpose computing. When Prodigy processors are provisioned in a hyperscale data center, they enable all AI, HPC, and general-purpose applications to run on one hardware infrastructure, saving companies billions of dollars per year. With data centers currently consuming over 4% of the planet’s electricity, predicted to be 10% by 2030, the ultra-low power Prodigy Universal Processor is critical to continue doubling worldwide data center capacity every four years. Tachyum, co-founded by Dr. Radoslav Danilak is building the world’s fastest AI supercomputer (128 AI exaflops) in the EU based on Prodigy processors. Tachyum has offices in the United States and Slovakia. For more information, visit https://www.tachyum.com/.

Copyright: 
Copyright Business Wire 2022