Editing 4-bit computing (section)

{{Short description|Computer architecture bit width}}
{{Use dmy dates|date=February 2020|cs1-dates=y}}
{{Computer architecture bit widths}}

'''4-bit computing''' is the use of [[computer architecture]]s in which [[integer (computer science)|integers]] and other [[data (computer science)|data]] units are 4 [[bit]]s wide. 4-bit [[central processing unit]] (CPU) and [[arithmetic logic unit]] (ALU) architectures are those that are based on [[processor register|registers]] or [[bus (computing)|data buses]] of that size. A group of four bits is also called a [[nibble]] and has 2<sup>4</sup> = 16 possible values, with a range of 0 to 15.

4-bit computation is obsolete, i.e. CPUs supporting 4-bit as the maximum size. However, 4-bit integers (or smaller), and 4-bit floating point is gaining ground for AI, large-language models.<ref>{{cite conference |arxiv=2310.16836 |doi=10.18653/v1/2023.emnlp-main.39 |title=LLM-FP4: 4-Bit Floating-Point Quantized Transformers |book-title=Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing |date=2023 |last1=Liu |first1=Shih-Yang |last2=Liu |first2=Zechun |last3=Huang |first3=Xijie |last4=Dong |first4=Pingcheng |last5=Cheng |first5=Kwang-Ting |pages=592–605 }}</ref>

4-bit processors were widely used in [[electronic calculator]]s and other roles where decimal math was used, like electronic [[cash register]]s, [[microwave oven]] timers, and so forth. This is because a 4-bit value holds a single [[binary-coded decimal]] (BCD) digit, making it a natural size for directly processing decimal values. As a 4-bit value is generally too small to hold a [[memory address]] for real-world programs or data, the [[address bus]] of these systems was generally larger. For instance, the canonical 4-bit [[microprocessor]], the [[Intel 4004]], had a 12-bit address format.

4-bit designs were used only for a short period when [[integrated circuit]]s were still expensive, and were found primarily in cost-sensitive roles. While 4-bit computing is mostly obsolete, 4-bit values are still used in the same decimal-centric roles they were developed for, and modern implementations are generally much wider and process multiple 4-bit values in parallel. An example of such a system is the [[HP Saturn]] design of the 1980s. By the 1990s, most such uses had been replaced by general purpose binary designs.