Connect with us


Tachyum Details AI Training Using Its Breakthrough Super-Sparsity Technology




LAS VEGAS, October 18, 2022–(BUSINESS WIRE)–Tachyum™ today released a white paper that examines how its Prodigy processor responds to AI trends, enabling environmentally friendly deep learning workloads by reducing the energy consumption and carbon emissions. The AI ​​training models operated are all fully functional, ensuring that Prodigy remains at the forefront of the industry.

Prodigy, the world’s first universal processor, offers a revolutionary new architecture that unifies CPU, GPGPU, and TPU functionality into a single monolithic device, enabling Prodigy processors to meet the high demands of cloud and HPC/AI workloads , with no cost or power-hungry accelerators, using a simple homogeneous software model that is aligned with software composability and dynamic reallocation of server resources to maximize utilization.

Among the features outlined in the latest Tachyum whitepaper are AI training techniques that include:

  • Conscious Quantification Training – Used with low precision data types to train deep neural network models. Prodigy processors enable extensive neural network training using only 8-bit floating-point (FP8) representations for all matrix and tensor calculations – weights, activations, errors, and gradients. This leads to dramatic improvements in computational efficiency and power efficiency without affecting model convergence and accuracy, reducing memory transfer times, and enabling the use of fast computational operations at low precision.

  • Gradient scale per channel – Tachyum uses a per-layer gradient scale instead of a global loss scale for training deep neural networks using FP8.

  • parsimony – Tachyum’s Prodigy processor provides native support for block structure parsimony and super-parsity. It supports sparse matrix multiplication optimized for compressed neural network models, reducing memory and compute requirements.

  • Results for image classification and instance segmentation – the company trains a wide range of computer vision, object detection, and instance segmentation models using the FP8 format.

“With our recently released architectural white paper detailing the Prodigy processor, we have generated tremendous interest and have been asked for more details on our AI capabilities,” said Dr. Radoslav Danilak, Founder and CEO of Tachyum. “This latest deep dive provides an overview of all the AI ​​training models that can work on Prodigy and how each of these approaches enables system developers to tackle ever-increasing deep learning computations. complex, offering performances an order of magnitude better than those currently available from competitors.”

Prodigy delivers unprecedented data center performance, power and economy, dramatically reducing CAPEX and OPEX. Due to their utility for high-performance and 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 utilization. servers. Tachyum’s Prodigy delivers up to 4x the performance of the highest performing x86 CPUs (for cloud workloads) and up to 3x the performance of the highest performing GPU for HPC and 6x for AI apps.

Those interested in learning more about Tachyum’s FP8 can download the white paper Tachyum Prodigy on the Leading Edge of AI Industry Trends at tachyum-prodigy-artificial-intelligence/.

Follow Tachyum

About Tachyum

Tachyum is transforming the AI, HPC, and public and private cloud data center markets with its recently launched flagship product. Prodigy, the world’s first general-purpose processor, unifies the functionality of a CPU, GPU, and TPU into a single processor that delivers industry-leading performance, cost, and power efficiency for specialty and high-end computing. general use. When Prodigy processors are delivered in a large-scale data center, they enable all AI, HPC, and general-purpose applications to run on hardware infrastructure, saving businesses billions of dollars. dollars per year. With data centers currently consuming more than 4% of the world’s electricity, up from 10% by 2030, the ultra-low-power Prodigy Universal Processor is essential to continue to double global 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

See the source version on


Mark Smith
JPR Communications