An Introduction to Floating-Point Arithmetic - Alan Zucconi

Glitchified on Instagram: "Modern GPUs perform 10^12–10^15 arithmetic operations per second depending on class and precision. Explanation and examples: FLOPS = floating-point operations per second. 1 TFLOP = 10^12 FLOPS. Midrange/mobile GPUs: hundreds of GFLOPS to a few TFLOPS (10^9–10^12 FLOPS). High-end consumer desktop GPUs: tens of TFLOPS. Example: an RTX 4090 ≈ 82.6 TFLOPS (≈8.26×10^13 FLOPS) for FP32 peak. Data-center AI/HPC GPUs using low-precision matrix engines: hundreds to nearly a tho

Converting Floating Point Values in the Binary Numerical System

[D] Consider a 16-bit, floating-point number in a format similar to that discussed in Problem 9.21, with a 6-bit exponent and a 9-bit mantissa fraction. The base of the scale factor is 2 and the exponent is represented in excess- 31 format. (a) Add the numbers A and B, formatted as follows: A= 0 100001 11111110 B= 0 011111 001010101 Give the answer in normalized form. Remember that an implicit 1 is to the left of the binary point but is not included in the A and B formats.