50
R. Lulli et al.
7.1.2 PC Hardware Evolution of the 2000s
In the 2000s, several problems appeared on the CPU market: the development of
silicon foundries with higher integration scale required increasingly important investments that only a large market with the high added value could allow to amortize. So
the number of companies able to have a state of the art silicon foundry was reduced
to a few units.
In that period, it was also understood that the increase in the computing capacity
of the microprocessors no longer grew linearly with the clock frequency. Moreover,
in 2005, Intel failed to market its latest microprocessor (Pentium 4 at 4 GHz clock
frequency) because it was unstable when used with air-cooled heat sinks. The most
computing capable Intel microprocessors of that period absorbed over 100 W of
power and produced a quantity of heat very near to the package dissipation technical
limit. These problems led Intel and other companies to stop the MHz race (which
had characterized previous decades and which in recent years had turned into the
race for GHz) and to change technology roadmaps.
The solution to the problem was found, considering that the increase factor of the
power dissipated by a microprocessor grows approximately with the square of the
clock frequency growth factor. Theoretically, it was possible to replace a high clock
frequency capacity and significantly reducing energy consumption. This solution
introduced the so-called multi-core CPU, a CPU composed of two microprocessors in the same chip at half clock frequency (concerning to the previous one) still
maintaining the same order of computing power, but significantly reducing energy
consumption.
Although considering the technological limits of Moore’s law [5], from the second
half of the 2000s, these CPUs have evolved rapidly both in their internal architecture
and in the number of cores integrated into a single chip.
The CPU computing power has evolved dramatically even if the clock frequencies
of the individual cores have remained in the order of 4 GHz. However, according to
Amdahl’s law [6], to make the most of the computing capacity of the new CPUs, the
application software has to be modified through the integration of algorithms and
methods of parallel computing.
In the same period, there was an introduction to the market of the so-called manycore architectures. “Multi-core simply means “more than one core,” but when the
number of cores grows well beyond the reach of finger counting, we use another name.
Many-core chips are multi-cores that contain tens, hundreds, or even thousands of
cores. While there is no hard threshold beyond which a multicore becomes a manycore, an easy distinction is that you probably have a many-core if you no longer care
about losing one or two cores” [7].
Examples of many-core architectures are Nvidia and AMD GPUs and the Intel
Xeon Phi. GPUs were initially being developed to raise the computing power of PC
video adapters, but their architecture it also proved suitable for developing parallelizable algorithms typical of DSP applications. In the GPUs, the Control Unit (CU) and
the cache management are elementary respect to the one of CPU. A single Control
R. Lulli et al.
7.1.2 PC Hardware Evolution of the 2000s
In the 2000s, several problems appeared on the CPU market: the development of
silicon foundries with higher integration scale required increasingly important investments that only a large market with the high added value could allow to amortize. So
the number of companies able to have a state of the art silicon foundry was reduced
to a few units.
In that period, it was also understood that the increase in the computing capacity
of the microprocessors no longer grew linearly with the clock frequency. Moreover,
in 2005, Intel failed to market its latest microprocessor (Pentium 4 at 4 GHz clock
frequency) because it was unstable when used with air-cooled heat sinks. The most
computing capable Intel microprocessors of that period absorbed over 100 W of
power and produced a quantity of heat very near to the package dissipation technical
limit. These problems led Intel and other companies to stop the MHz race (which
had characterized previous decades and which in recent years had turned into the
race for GHz) and to change technology roadmaps.
The solution to the problem was found, considering that the increase factor of the
power dissipated by a microprocessor grows approximately with the square of the
clock frequency growth factor. Theoretically, it was possible to replace a high clock
frequency capacity and significantly reducing energy consumption. This solution
introduced the so-called multi-core CPU, a CPU composed of two microprocessors in the same chip at half clock frequency (concerning to the previous one) still
maintaining the same order of computing power, but significantly reducing energy
consumption.
Although considering the technological limits of Moore’s law [5], from the second
half of the 2000s, these CPUs have evolved rapidly both in their internal architecture
and in the number of cores integrated into a single chip.
The CPU computing power has evolved dramatically even if the clock frequencies
of the individual cores have remained in the order of 4 GHz. However, according to
Amdahl’s law [6], to make the most of the computing capacity of the new CPUs, the
application software has to be modified through the integration of algorithms and
methods of parallel computing.
In the same period, there was an introduction to the market of the so-called manycore architectures. “Multi-core simply means “more than one core,” but when the
number of cores grows well beyond the reach of finger counting, we use another name.
Many-core chips are multi-cores that contain tens, hundreds, or even thousands of
cores. While there is no hard threshold beyond which a multicore becomes a manycore, an easy distinction is that you probably have a many-core if you no longer care
about losing one or two cores” [7].
Examples of many-core architectures are Nvidia and AMD GPUs and the Intel
Xeon Phi. GPUs were initially being developed to raise the computing power of PC
video adapters, but their architecture it also proved suitable for developing parallelizable algorithms typical of DSP applications. In the GPUs, the Control Unit (CU) and
the cache management are elementary respect to the one of CPU. A single Control