POSIT unum universal number system | Beyond Floating Point | 2 decimal places more accurate than 32 bit single precision FP | 3 decimal places more accurate than 64 bit double precision FP | High hardware efficiency

The End of Error

Floating point computation is only accurate to 6 decimal places in 32-bit single precision, or to 16 decimal places in 64-bit double precision. Values beyond that are rounded or truncated.

The POSIT universal number system, or unum, is more accurate than single precision floating point by 2 decimal places, and by 3 decimal places over double precision.

With the POSIT unum, you will achieve a higher dynamic range, and the accuracy of your computations will improve by 2 or 3 decimal places. This is particularly significant when carrying the result of one computation into the next, as the inaccuracy of the floating point computation becomes exponential.

There are a handful of academic open source implementations of POSIT, but VividSparks’s embodiment as a soft IP core represents decades of refinement of the number system for commercial, industrial, and research applications.

Beyond Floating Point

VividSparks extracts unmatched power and performance from your hardware. The higher dynamic range means less silicon for the same application, combined with higher computational accuracy from the POSIT number system versus Floating Point.

Applications include HPC High Performance Computing, Graphics, Machine Learning, Artificial Intelligence, DSP digital signal processing, audio and video.

Sky Blue Microsystems GmbH of Munich, Germany, provides advanced electronics products for science and industry, and is an international distributor for Vivid Sparks. 

VividSparks POSIT delivery embodiment implementation formats

Vivid Sparks is delivered in the most flexible and economical form of your choice:

• in the Verilog HPL as a soft IP core for inclusion in your ASIC or FPGA
• as one of our or FPGA board level solutions
• in the Cloud as a service, with time-slice rentals
• as one of the following turnkey chips:
  • Supersonik HPC coprocessor for High Performance Computing
  • Racer coprocessor for Graphics applications
  • Falkon coprocessor for AI Artificial Intelligence
  • Tez coprocessor for automotive ECU calculations

VividSparks soft IP core is delivered with an API that is embedded in your source code that calls the VividSparks IP to perform computations, as required. You can think of this as a virtual math co-processor, “virtual” is that it is incorporated in your design and code rather than as external hardware, and “math co-processor” in that it performs your computations faster, more exactly, and more efficiently than any other hardware.

VividSparks POSIT mathematical operations for extreme accuracy and efficiency

VivdSparks has implemented a full range of operations using CFA Carry Free Addition, including:

• Add, subtract, multiply and divide
• multiply-add (for example: A+/B*C)
• sum of multiplications (A*B+/- C*D)
• reciprocals
• inverse square root
• logarithms
• exponents (exp(x))
• trigonometric functions such as sine, cosine, tan
• inverse trig functions
• hyperbolic functions
• inverse hyperbolic functions
• compare functions
• convert functions