As small as a pocket MP3 player, all the way to system on a chip, network switch, router, car computer, mobile phone, smart TV, robot, medical systems, ATM machine, laptop, AL/ML servers, etc. As far as you can think of, there are a lot of systems(or you can call them servers) running related to our daily life.
CPU, memory, I/O, and power are essential components for all systems to be operational. Of course, you can add on storage(SD card, HDD, SSD or NVMe), Ethernet, Wi-Fi, display, other input/output or other peripherals depends on your need.
Depending on your application, target market and customer usage scenario, your systems can vary a lot. Couple examples as below.
- NAS: Network-attached storage
- You see the keywords “network” & “storage”. It means you have to have at least one NIC and one or more drives. And of course, you need to have CPU, memory, and a storage(could be a dedicated USB flash or using SATA drive) for OS installation for NAS to be operational then it can share attached drives in your home network. Intel Celeron J4125 could be a good choice for the NAS. It has very lower power consumption and the cost is low as well. It has SATA and USB built in already. Hence, you only need to find a PCIe Gen 2 x1 ethernet chip then you can pretty much put together all components for system board design. Of course, you need to get reference design and AVL from CPU/chipset vendor. It is not just that you also need to define how many drives you are going to support, connectors/interface, how much memory to be installed, features and chassis, etc. You’d better having target marketing requirement -> NAS product specification -> engineering specification then develop and validate it.
- Rock chip RK3588(ARM base) is another good choice. It already has SATA, USB and NIC included. And actually you can it make as a linux PC since it has display output and audio.
- If you would like to have one NAS but you don’t want to buy it, your retired PC/laptop is perfect for this purpose.
Enterprise storage: A server system with good amount of storage devices either HDDs or NVMes. Support FC, iSCSI, RAID controller and data protection. Leverage a shipping reliable server from tier 1 vendor will be a good way to develop your storage server. On top of shipping HW from tier 1 vendors, put the OS(Suse or Redhat) to manage storage devices then develop a fancy and practical user interface. Or you start everything from scratch, build your own server then customize OS you want to use for the storage server and develop user interface.
Gaming PC: Architecture wise, it is same as regular PC. But it needs to have powerful CPU and hi-end GPU for gaming.
General purpose server: TBU
AI/ML Servers: Servers built for Artificial Intelligence and Machine Learning. AI/ML servers are servers which can plug in as many GPGPUs as they can. You can call it as a GPGPU box or GPGPU server. Of course, this server needs to provide high speed network connection to exchange data as well. Now Nvidia provides whole rack AI/ML solution which is GB200 NVL72. It is specially designed for AI/ML and it has very high throughput and performance. Also the price is very high. Apparently, most of companies are using GPGPU for AI/ML training. Actually, some big companies(Google, Amazon, Microsoft) can afford to develop their own ASIC to replace partial or all training which are done by GPGPU. These ASICs are designed to handle AI/ML training unlike General Purpose GPUs. These ASICs are much cheaper than GPGPU and they have lower power consumption. To these companies, they save a lot of money and will not be bounded by Nvidia GPGPU.
Server block diagram: Gigabyte MA34-CP0
PC block diagram: Intel Core desktop 700 series