Technology

Why Vorticity?

All modern computer processors are based on a design developed by a brilliant scientist named John Von Neuman in the 1940s. Its versatility led to an explosion in computing in the decades following. However, this architecture has always fallen short for data heavy compute intensive scientific simulations. John Backus in 1977 accepting his Turing award, coined the term “von Neumann bottleneck” to describe this problem. In the past, improvements in speed and ability of processors from one generation to the next made this problem less relevant. But with the end of Dennard scaling and plateauing of Moore’s Law, humanity needs a new solution to tackle more complex scientific and technological problems. As a result, Vorticity built a computer system for scientific computing, including breakthrough improvements from system design, all the way down to the processor architecture. The result is a system that solves scientific compute problems up to five orders of magnitude faster than existing state of the art GPU/CPU based computers.

How does Vorticity achieve significant improvements in scientific compute speed?

Stage 1

Compute Speedsignal

Generic software

arrow downarrow up

High-performance hardware

Businesses invest in expensive hardware to increase performance, but fail to drastically improve compute speed

figure

Stage 2

Compute Speedsignal

Custom written software

arrow downarrow up

High-performance hardware

Custom written software is designed to optimize compute on high-performance hardware

small logo
figure

Stage 3

Compute Speedsignal

Custom written software

arrow downarrow up

Custom built hardware

Vertically integrated compute design fully integrating software and hardware to achieve maximum compute speed improvement

small logo
figure

The Problem

The Problem
Modern data-parallel applications for scientific computing are placing an increasing demand for floating point operations per second and memory bandwidth. With Moore' Law for general purpose computing slowing and Dennard Scaling ending, compute architectures must be redesigned to meet rising performance needs.

The Limitation of Current Solutions

The Limitation of Current Solutions
The growing number of hardware accelerators addresses specific problems but lacks the scalability for a wider range of scientific applications. AI architectures target these challenges in the machine learning domain, but are often too specialized for broader scientific tasks.

The Solution

The Solution
We propose a Scientific Processing Unit (SPU), a domain-specific architecture for scientific computing reaching extreme high performance and energy efficiency, while maintaining the flexibility of a general-purpose engine. Vorticity combines hardware innovations with optimized software to deliver unmatched computational speed for solving complex scientific problems.