Have you ever thought of buying a server CPU? These have a much greater power than the desktop ones, but also a higher price and are accompanied by special plates, different RAM memory and all this for much higher prices than a desktop CPU. We explain the differences and the reason for them.
We know that there are CPUs for all types of sizes and computers. From the one in a smartphone to the most powerful supercomputer. But while there is a clear difference between processors for smartphones and for PCs, it does not seem that there is that much difference between that of a computer and a server except in the number of cores and the fact that they are more expensive. It really is not so.
More cores in a server CPU
Because servers have to run a greater number of processes than a conventional PC at the same time, their CPU needs to have a greater number of cores. As you well know, the programs scale in performance under two metrics. The first one is the parallel performance that grows with the number of cores in the system, the second is the serial performance that depends on the power of each processor alone.
The reason why a server CPU has more cores, as well as more cache, actually has to do with economics. A company that builds a server has enough capital to pay for the purchase of a large processor, with a large number of cores and a large amount of cache memory inside.
Due to the greater economic capacity, when building CPUs for servers, manufacturers have a greater number of cores and cache memory.
Multiple CPUs on servers
Before the advent of multi-core CPUs, we had a small era where we saw multi-processor configurations on the same interposer or sharing the same socket, but that was short-lived as soon as dual-core CPUs appeared.
In the server market it is normal to see several CPUs mounted on a motherboard, each with its socket. In PC we have seen CPUs with chiplets, but it is not the same, since they continue to use a single socket and it is not the same size as the sockets for servers.
The socket and motherboard of a server CPU is different
The other point is the Socket, which is where the processor is connected. The fact that the processor of a server is much larger than that of a PC makes it necessary to have a special socket and with it special motherboards that are designed for that market, which are obviously much more expensive.
Server motherboards have things that conventional PCs do not have, since they can have things such as a greater number of Gigabit Ethernet interfaces, SFP transceivers that allow the use of fiber optic interconnections, SAS ports for connecting disks in servers, etc. . Regarding RAM, many server CPUs use ECC memory, which cannot be used with conventional CPUs and traditional RAM cannot be used with motherboards and therefore with server CPUs.
ECC memory includes additional memory bits, which are handled by additional hardware in the memory controller and interface. These extra bits are used to store an encryption code, which is written when data is stored in memory. When the data is read from the server CPU, both codes are compared. In this way, the integrity of the data and therefore of the information can be maintained.
A server CPU is always running
The CPU of a server will be on for 24 hours a day, 7 days a week and 365 days a year. That is, from the moment they are installed until they stop working, they will be in operation all the time. What is translated is that when it comes to doing quality tests on a CPU for servers, they are much more demanding and the number of processors that pass the cut is much lower. That is why they are so expensive and rare.
This is why server CPUs tend to run at a lower average clock speed per core than their desktop counterparts. The reason is that they are designed so that they can work without ever having to turn off the server and that it does not reach critical consumption levels. That is why they are not designed for overclocking and all hardware for this function is completely ruled out.
The lower clock speed they make up for with a higher number of cores. This allows the CPU to run a greater number of threads in parallel. It must be taken into account that contemporary operating systems handle not tens, but hundreds of execution threads simultaneously, which will be assigned to the cores that are free from the server’s CPU.
No GPUs or integrated media coprocessors
Server CPUs are not used to reproduce full graphics, what’s more, the designers don’t put any graphics hardware on the same chip as the CPU itself, so they don’t have an integrated GPU. One reason is because of the thermal choking that a high-performance GPU can put on the CPU by sharing space with it. We have already said that the important thing for a server CPU is the fact that it can always be on.
That is why while the current CPUs in PCs are really SoCs in which we have several types of processor with heterogeneous architecture, CPUs for servers are still homogeneous configurations that in general tend to use a single type of core, those of the CPU . And they do not include any other type of processor inside.
Many current servers perform cloud computing services and often carry dedicated graphics cards with virtualization, this means that the GPU can function as several virtual GPUs asynchronously for several different remote clients.