CPUs (Processors)

Server processors - scalable computing performance for high requirements

As the central processing unit of servers and workstations, the processor is also known as the CPU. Through the further development of chip technology, the performance has been constantly increased, but the purpose of these components has remained largely unchanged with the processing of data and computing instructions.
The two US manufacturers Intel and AMD (Advanced Micro Devices) compete in the market for desktop and server processors. With its Xeon CPUs for professional applications, however, Intel is the market leader in this segment despite the now strong competition from AMD's EPYC processors. Due to their scalability, these CPUs often allow the operation of several processors in one system, which can further increase the computing power.

The processor generations - increased performance through modern manufacturing technology

Depending on the model series and generation, CPUs differ in the structure of the semiconductor chip (die) used and therefore in their performance. The process architecture, which depends on the manufacturing process, determines how many transistors and units can be integrated on a chip. This size is specified in nanometres (nm), and it generally applies here that smaller structures enable higher computing performance and energy efficiency. Since modern manufacturing technology can usually also process larger data streams, newer processor generations support faster main memory and provide a larger number of PCIe lanes (PCI Express) for connecting graphic cards and NVMe SSDs (Non-Volatile Memory Express) with high bandwidth.

How the sockets for installing the processors differentiate?

A carrier board is used to install the chips in the system. It establishes the connection between the processor and the mainboard through electrical contacts on the underside. In the technical implementation of these contacts, a distinction is made between two common types of processor sockets in the sector of servers, workstations and PCs:

• PGA Sockets (Pin Grid Array)

With CPUs for PGA-type sockets, the contacts on the underside of the processor are designed as contact pins. In return, the motherboard has a mounting with contact surfaces in correspondingly arranged holes. Since the pins can be easily bent, PGA CPUs tend to be more vulnerable to damage, whereas the motherboard socket is comparatively insensitive. To simplify installation and prevent damage, both processor and socket are often marked with appropriate markings to indicate correct alignment.

• LGA Sockets (Land Grid Array)

In case of the LGA socket type, the situation is the opposite. The processor has the contact surfaces and the pins for establishing the electrical connection are located in the mounting of the mainboard. This shifts the vulnerability to damage from the CPU to the socket. The advantage of this variant, however, is that the contact pins can be made smaller and therefore a larger number of contact points is possible on the same surface. Sockets 3647 and 4189 for Intel Xeon Scalable processors and socket SP3 for AMD EPYC CPUs, which are relevant for server systems, are LGA sockets.

The various processor generations of the manufacturers often vary in the number and arrangement of the contacts, which often makes it necessary to replace the entire mainboard when upgrading the CPU to a new generation. Therefore, it is essential to check the compatibility of the hardware before changing the processor. Under certain circumstances, it may make more sense to replace the existing CPU with a more powerful model of the same generation, thereby avoiding high costs due to the change of platform.

The high performance of server processors requires appropriate cooling

To ensure consistently high performance and prevent damage due to overheating, the temperature of the chips generated during operation must be reliably dissipated. For this purpose, most processors have an integrated heat spreader (IHS) as a cover, which both protects against mechanical damage and ensures good heat transfer to the CPU heatsink mounted above it. The heat spreader can either be glued to the carrier board or soldered to the chip. The advantage of a fixed connection with solder is the permanently constant conductivity, as there is no thermal paste between the chip and the IHS, which can age under certain circumstances.

The required cooling of a CPU always depends on its performance, which is why processor manufacturers specify the thermal power dissipation (TDP: Thermal Design Power) in watts for the various models. This value can also be found in the technical data of CPU coolers and simplifies the selection of a proper cooling solution for the respective processor. If the cooling is not sufficiently dimensioned and a CPU therefore exceeds its intended maximum operating temperature, the integrated protection mechanisms intervene and reduce the power to lower the temperature. This prevents damage due to overheating, but the performance of a system decreases noticeably. In addition, a permanently too high temperature can have a negative effect on the lifespan. Therefore, cooling is a decisive factor for the reliable and efficient operation of servers and workstations.

What advantages offer processors with a high core count?

All modern CPUs are multicore processors that combine several cores in one semiconductor chip. Since each core represents a largely independent computing unit, the workload of a system can be better distributed and processed in parallel as the number of cores increases. Furthermore, the models of the higher performance classes often support multithreading, which means they have additional virtual cores. These threads share the existing units of the chip with the real cores, which means that the performance increase is somewhat lower in direct comparison. Especially for applications that benefit from a large number of processing cores, such as in the field of virtualisation, this function represents a considerable added value. The most powerful Intel Xeon Scalable processors have up to 40 cores and 80 threads; the competitor AMD even achieves up to 64 cores and 128 threads with the top models of the EPYC CPUs. Through refined production processes, manufacturers will increase the number of cores even further in the coming processor generations.

What impact has the clock frequency on the processor's performance?

Another important factor is the clock frequency. This value, usually in the unit gigahertz (GHz), indicates the number of clock cycles that a processor can execute per second. At a clock rate of 2.5 GHz, for example, 2.5 billion clock cycles are executed. Therefore, the higher the clock frequency of the CPU, the faster computing instructions can be processed.  The technical base here is the system's basic clock defined by the mainboard, which is increased by a multiplier and therefore results in the processor's effective frequency. In most cases, CPUs with a smaller number of cores operate at a higher clock rate than models with a large number of processing cores. So if applications benefit more from faster processing of instructions than from distributing the load over many cores, models with a higher clock frequency often offer better performance.

In addition to the processor's standard clock, a second value is often specified, the turbo clock frequency. For processing demanding workloads, many CPUs can dynamically increase their frequency to enable higher computing performance. In addition to the thermal limits of the processor, this also takes into account how many cores are being utilised to achieve the highest possible safe clock speed. The performance increase is therefore lower the more cores are active due to the higher temperature development.
However, the evaluation of a processor on the clock rate is generally only useful within the same product generation of the respective manufacturer. Due to the further development of chip technology and the resulting increase in efficiency, newer models can often outperform the higher clocked previous generation despite a lower clock frequency. The effective performance of a CPU depends on how many instructions can be processed in one clock cycle. This value is called IPC (Instructions per Cycle) and is usually higher in newer microarchitectures.

Cache as buffer memory - fast data access for the processor

For high-performance processing to be possible at all, data and instructions must be accessible to the processor quickly. For this purpose, CPUs have cache as buffer memory in several levels, which differ in size and performance. Here, instructions from the main memory are buffered, as access to the cache can be significantly faster, resulting in lower waiting times. The distance between the cache and the computing core determines the speed. The shorter this distance, the faster but also smaller is the cache. In most processors, the cache is divided into three levels:

• L1 Cache (First-Level Cache)

The L1 cache is used to temporarily store the most frequently used data and instructions. It is operated with the processor clock and located in the processor core. This makes it the fastest cache level, but with its size in the kilobyte (KB) range, it is also the smallest.

• L2 cache (second-level cache)

The L2 cache is used to temporarily store data from the main memory. Due to its position outside the processor core, it can be significantly larger with sizes in the single-digit megabyte (MB) range, but also works slower due to its greater distance from the core. 

• L3 cache (third-level cache)

All cores of the processor can access the L3 cache, which simplifies the exchange of data between the cores. It is also used to synchronise data between the caches of the individual cores. With a size in the double-digit megabyte range, the third cache level is the largest and in comparison the slowest buffer memory.

Furthermore, different cache hierarchies are classified for multicore processors. Inclusive cache offers the advantage that data from the L1 cache is also stored in the L2 and L3 cache. Although these occupy additional memory space there, the exchange and synchronisation of data is made easier. In the case of exclusive cache, the data of the first-level cache is only available to the specific processor core, which provides more memory space in the further levels. However, this means that cores can only exchange this data with each other in a detoured way.

Refurbished high-performance server processors available at low prices - ServerShop24

The further development of semiconductor technology has constantly increased the performance of processors, but at the same time has led to sometimes astronomical prices for models of the latest generations. In contrast, however, the requirements for servers have hardly changed in many areas. Therefore, processors of previous generations, which are available as carefully tested used parts at affordable prices, can often fully meet the requirements of your application. With over 10 years of experience, we are your reliable partner for professionally refurbished used servers, storage solutions and network equipment. In addition to pre-configured systems, you will also find a large selection of high-performance processors and high-quality components in our online shop to individually adapt your IT infrastructure. Thanks to fast shipping from our extensive stock, you will receive your order with the shortest possible delivery times. If there are any questions about products and your order, our competent and friendly support team will assist you. Contact us - we will take care of your request!