512-bit computing

In computer architecture, 512-bit integers, memory addresses, or other data units are those that are 512 bits (64 octets) wide.

The maximum value of a signed 512-bit integer is 2511 − 1, written in decimal as 6,​703,​903,​964,​971,​298,​549,​787,​012,​499,​102,​923,​063,​739,​682,​910,​296,​196,​688,​861,​780,​721,​860,​882,​015,​036,​773,​488,​400,​937,​149,​083,​451,​713,​845,​015,​929,​093,​243,​025,​426,​876,​941,​405,​973,​284,​973,​216,​824,​503,​042,​047 (approximately 6.7039×10153).

However, the Xeon Phi's vector processing unit does not operate on individual numbers that are 512 bits long.

[1] Some GPUs, such as the Advanced Micro Devices (AMD) Radeon HD 2900XT, the Nvidia GTX 280,[2] GTX 285,[3] Quadro FX 5800,[4] and several Nvidia Tesla products, move data across a 512-bit memory bus.

AVX-512 are 512-bit extensions to the 256-bit Advanced Vector Extensions SIMD instructions for x86 instruction set architecture proposed by Intel in July 2013, and released in 2016 with Knights Landing, and in 2017 on the HEDT and consumer server platform, with Skylake-X and Skylake-SP respectively.