AMD Turbo Core

from Wikipedia, the free encyclopedia

AMD Turbo Core is a function for the automatic overclocking of main processors from AMD , which was first used in the six- core processors of the Phenom II series with K10 micro-architecture. Turbo Core allows the demand-oriented dynamic increase of the processor speed .

Turbo Core is automatically activated by the processor when the operating system requests the highest possible performance (“P0 state”) on at least one core. No additional drivers or software are required for this technology to work. This function is also known as automatic or dynamic overclocking function.


Many (often older) programs are not designed to use multiple processor cores and, even with a modern multi-core processor, only use one core because of the execution of only one thread . As a result, a single-core processor with a higher clock rate could execute such an application faster than a multi-core processor with a lower clock rate, even if the total computing power of the multi-core processor is higher than that of the single-core processor. The thermal design power ( TDP for short ; also thermal budget ) of a processor describes the maximum dissipated power loss of the entire processor through the cooling system. A processor in which only one core is fully utilized does not fully utilize the TDP. If the processor does not completely use its thermal budget due to inactive cores, a core can use part of the budget that has not yet been used without the cooling device having to dissipate more heat than it would in a state in which all cores are fully active, must be able to dissipate.

First version in the Phenom II processors

So that a processor core is considered inactive and parts of its TDP can be used by other cores, the core must at least be in the Halt state ( referred to as "C1 state" in the ACPI table). If at least half of the cores are in this state and the operating system requests the highest possible performance for at least one core, the remaining active cores can increase their clock rate by a fixed level (400 or 500 MHz depending on the processor).

Whether a processor (especially under full load, when the operating system reports a load of 100% for all cores) actually fully utilizes its TDP depends not least on the program used. For example, if a program only uses the integer unit but leaves the floating point arithmetic units unused, less heat is generated than with code, which uses all arithmetic units equally. Processors of the so-called "Black Edition" offer some setting options for Turbo Core . Among other things, you can define how many cores must be considered inactive so that the processor can increase its clock rate. The height of the turbo level can also be adjusted. Experienced overclockers can also determine the settings for the P0 state of the processor themselves.

Processors with Turbo Core

  • Simple version with restriction to half the cores:
    • Phenom II series (K10 microarchitecture) - recognizable by the “T” at the end of the version number
      • all processors Phenom II X6 1xxx T (6 cores)
      • Phenom II X4 960 T (4 cores)
  • Improved version with power measurement and dynamic clocking
    • AMD A-Series ( Llano )
      • all mobile variants in the FS-1 base, e.g. B. A8-3500 M
      • some desktop versions in the FM-1 socket: A8-3820, A8-3800, A6-3620, A6-3600, A6-3500
    • AMD C and E series ( Bobcat )
      • AMD C-60
      • AMD E-450
    • AMD FX Series ( Bulldozer )
      • all CPUs (in socket AM3 +)
    • Opterons
      • all CPUs of the 42xx and 62xx series except the Opteron 6204


While Intel launched its overclocking function called Turbo Boost with the Nehalem microarchitecture in 2008 , AMD introduced the first processors with automatic overclocking function in April 2010 with the first six-core processors Phenom II X6 1055T and the Phenom II X6 1090T. This simple version of Turbo Core was initially reserved for the Phenom II X6 models. The Phenom II X4 960T quad-core processor, which is based on the same chip as the six-core, followed in November 2011.

In June 2011 the first processors of the A-series (Llano) were sold, which use the improved Turbo-Core variant. There is no restriction to a fixed number of inactive cores, and multiple turbo levels are now possible. The AMD FX CPUs with Bulldozer core presented in October 2011 also implement this more flexible variant.


Turbo Core offers the theoretical possibility of executing programs that only use one thread faster, but in practice this advantage can only be used poorly on a normal system. To run half of the processors in turbo mode, at least half of the cores must be inactive. Modern operating systems distribute the applications to be calculated and executed evenly across all available cores, which means that a processor core does not often switch to the C1 state, which in turn means that it is not considered inactive and thus the other cores cannot clock higher. Another disadvantage is shown in Windows, which processes that are not optimized for multiple threads ( single thread applications ), in z. B. a quad-core processor and then processes 25% of the code on each core instead of calculating 100% on one core. Although only one core is active at the same time, the core needs time to switch to the C1 state, and the remaining cores need time to increase the clock frequency. Technologies such as core parking can eliminate these disadvantages, but must first establish themselves or establish themselves in the individual operating systems. In reality, Turbo Core can hardly deliver a gain in performance due to the existing problems without core parking .

Differences to Turbo Boost

P-states with Turbo Core
P0 P0 Turbo level
P0 P1 Nominal operation
P1 P2 Savings level 1
P2 P3 Savings level 2
P3 P4 Savings level 3
P4 P5 Savings level 4

AMD's Turbo Core differs from Intel's Turbo Boost in several ways . One of the big differences is that with Turbo Core at least half of the cores must be considered inactive so that the processor can turn on Turbo at all and thus achieve a higher clock rate. Unlike Intel's Turbo version, Turbo Core cannot use the TDP part of two cores in a six-core processor. While Turbo Core can only increase the processor's clock rate by a fixed level, some mobile quad-core processors from Intel can increase the clock rate by up to 9 multiplier levels of 133.33 MHz each with only one active core. Since it is seldom the case with modern operating systems that only one core of the processor is loaded, the highest multiplier level is rarely or not at all achieved in practice.

While with Turbo Boost a core is only considered inactive when it remains in the C3 state, with Turbo Core cores in the C1 state are already inactive. The advantage of this is, on the one hand, that the processor can switch to the C1 state more quickly and "wake up" from it again. On the other hand, it is more likely that the processor is not in C1 or in a deeper sleep state when there is no processor load than in the ≥ C3 state. Turbo Core also - unlike Turbo Boost - does not require an additional entry in the BIOS . AMD has further separated the P states of the processor from the P states of the operating system (table on the right). If the operating system requests full performance (P0 state), however, the processor first goes to P1 instead of P0. The P1 state does not stand for the first power-saving level, as usual, but for working with the nominal cycle, which means that the turbo cycle has its own entry as the highest possible performance. This creates a separate entry for the turbo stage in the ACPI table, with which the processor can, for example, set a higher core voltage in the turbo stage . The separation of the P0 status of the operating system and the P0 status of the processor in the Linux kernel with version numbers 2.6.31, 2.6.32 and 2.6.33 ensured that the kernels read out an incorrect clock rate and, as a result, the system up to Runs 27% slower. Switching off the energy saving function Cool'n'Quiet helps , but this leads to an increased power consumption of the system when idling by 10 to 20 watts. The Linux kernels from 2.6.35 and the patched versions of the old kernel versions no longer have this problem.

See also

Individual evidence

  1. Core Parking in Windows Server 2008 R2 and Windows 7 | Dr. Dobb's and Intel Go Parallel Programming
  2. If the Windows operating system governs thread distribution, Turbo CORE fizzles out.