Improvements in CPU features help shape selection (original) (raw)

When purchasing server hardware, organizations have always had to decide on CPU type. In the early days, this decision might have been as simple as choosing between an Intel or AMD processor.

There are now many considerations around CPU features, such as the number of cores, architecture and speed. These choices must be weighed carefully, as CPU selection can affect both performance and licensing costs. This is because specific CPU features have changed -- and advanced -- over the last 10 years.

Single core vs. multiprocessor cores

The number of physical cores is one of the most obvious things IT buyers should look for when evaluating CPU features.

Multicore CPUs are widespread, and the number of cores available is far higher than what could be found 10 years ago. In 2010, for example, Advanced Micro Devices (AMD) released the Phenom II X6, the first hex-core desktop CPU. Today, AMD's EPYC 7742 is equipped with 64 cores while Intel's Xeon Platinum 9282 processor has 56 cores.

When it comes to the number of cores within a CPU, many IT buyers assume that more is better. While higher core counts are generally regarded as a good thing, it is also necessary to consider how the cores affect software licensing costs.

Windows Server 2019 Datacenter and Standard licenses, for example, are priced based on a 16-core system. License stacking may be required to properly license servers with additional CPU cores.

Threads

At one point, the number of threads a system could simultaneously execute was limited by the number of physical processors. A single CPU system, for instance, could execute a single thread.

However, it isn't just the CPU count that determines the number of threads that can be simultaneously executed. The number of CPU cores plays a role, as does the use of Multithreading technology. AMD and Intel both support multithreading to enable each CPU core to simultaneously execute multiple threads.

Intel CPUs accomplish multithreading using Intel Hyper-Threading Technology. That means the previously mentioned Intel Xeon Platinum 9282 processor and its 56 cores can handle 112 threads because each core can support two execution threads.

AMD uses simultaneous multithreading. This means the AMD EPYC 7742 noted earlier can support 128 threads in its 64 cores.

Though the practice of licensing software based on the number of available cores has become relatively common, software vendors generally do not license software based on threads.

Frequency

CPU frequency, also referred to as clock speed, indicates how many cycles the processor can perform each second. Most modern processors operate at a frequency in the range of 2 Ghz to 4 GHz. By comparison, the original IBM PC -- which ran on Intel's 8088 processor -- had a clock speed of 4.77 MHz.

Traditionally, frequency was a direct indication of CPU performance. A higher clock speed meant the CPU ran faster. Today, you must also factor in the instructions per clock or cycle (IPC). The total number of instructions per second can be determined by multiplying a CPU's frequency by its IPC value.

Yet CPUs do not always run at their maximum frequency. Manufacturers use CPU throttling, or even shut down cores, during periods of low activity to reduce power consumption.

Thermal considerations

Regardless of the manufacturer or architecture, CPUs require adequate cooling to avoid damage. Intel does not provide temperature ranges for its processors because they can vary. However, Intel does perform thermal junction and case temperature monitoring to ensure CPUs do not overheat.

Current processors from AMD and Intel include protections to prevent damage related to overheating. A CPU will run slower than normal if it begins to overheat. This process, known as thermal throttling, is designed to help cool the CPU.

X86 vs. ARM servers

Most server CPU technology has historically been based on the x86/x64 architecture, but advanced RISC machine (ARM) processors are gaining traction. ARM processers adhere to RISC architecture. RISC processors have existed for decades but have gained far greater traction in the last 10 years because of their use in various mobile devices.

ARM servers offer many potential benefits over x86 systems. They are generally far more efficient than x86 systems and consume significantly less power. This makes ARM servers particularly attractive because power and cooling tend to be some of the biggest expenses associated with running a data center.

For now, there are few ARM-based servers available. Hewlett Packard Enterprise offers the Apollo 70 System equipped with Cavium's ThunderX2 ARM processor, for example, but many other server vendors do not currently offer ARM hardware.

Another issue with running ARM servers is the lack of software. A couple of years ago, Microsoft announced that it was experimenting with Windows Server on ARM. To date, an ARM version of Windows Server has yet to be released. Open source OSes such as Ubuntu are available for ARM systems, but finding ARM versions of popular applications may still be problematic.