CPU speed, or the speed of the central processing unit in a computer, is essentially the speed at which the computer can perform calculations which are fed to it through software program instructions loaded into volatile random access memory (RAM). Processor speed is limited by the number of transistors built into a processor, parallel connections to other processors, the capacity of the bus to transmit data back and forth from the CPU to memory, and other hardware specifications. Most CPUs also have their own memory registers for performing core calculations locally, without having to transmit them across a bus to another hardware component and back.
Computer processors on current systems are capable of operating at such a rapid pace that performance limitations in most personal computers are tied much more to the bottleneck of bus capacity. The amount of RAM available and the design of the software that is accessing the system are also more critical than the actual CPU performance itself. Multithreading capacity in CPU design is another key speed factor, which is the ability of the CPU to perform multiple tasks in a shared execution environment on the CPU, so less information has to be stored and retrieved from memory during program operations.
Hobbyists will often change what is known as the clock speed on a CPU, by overclocking the device. Part of what determines the CPU speed on a computer is its clock rate, or clock speed, which is the number of clock cycles, based on the computer's internal clock, that the CPU needs to perform one instruction. Identical CPUs can have much different performance rates if one is clocked, for instance, to add two numbers together in 10 cycles, where the other CPU does the same calculation in 2-clock cycles.
While overclocking a computer's CPU will take it out of synchronization with the speed of the bus, it can increase CPU performance considerably on older systems that have been enhanced with new bus architectures. Newer processors won't benefit from changes in clock speed, however, since they are already operating at a level far above what the bus and computer memory can handle. With CPU speed in the multiple gigahertz range, billions of calculations are performed per second. A 2.4 gigahertz CPU therefore can run 2.4 billion calculations per second, whereas a typical 32- or 64-bit Peripheral Component Interconnect (PCI) bus will run in the 127–508 megabyte (millions of bytes) per second range.
Another limiting factor for CPU speed, whether overclocked or not, involves the ability of the entire computer system to dissipate heat away from the processor, as increased heat generates a thermal barrier for the transmission of electrical signals in metal oxide semiconductor field-effect transistor (MOSFET) CPU designs. Faster processors require higher wattage power supplies, which translates to greater heat generation. Heat sinks, which act as mini-radiators, are built onto the surface of processors to dissipate heat by conduction, and fan systems within the computer housing carry it away as well by convection.
Running multiple processors in parallel to share data calculations on one computer is now a common approach with most computers to increase CPU speed. On advanced systems, liquid cooling is also involved to keep the CPU at a stable temperature setting. Very advanced supercomputers use thousands of processors operating in parallel, and are cooled with liquid nitrogen or liquid helium to temperatures around -452° Fahrenheit (-269° Celsius), with clock speeds reaching above 500 gigahertz, or 500 billion calculations per second.