server case; server chassis; computer case; 4u case; computer case best; Universal server chassis
computer chassis; 4u chassis; computer case cover; Universal server chassis; Storage server chassis; AI server chassis; Edge server chassis; Heterogeneous architecture chassis; Node server chassis
It’s time to reconstruct the server!
Basic idea of modularization
Modular customized data center will become one of the trends in the development of data centers in the future. With the expansion of demand for modular data centers, the demand for “modular customized” servers, storage and heterogeneous computing power is becoming increasingly strong, and modular servers will become the development trend in the future.
In traditional rack servers, designers usually arrange storage, PCIe, power supply, heat dissipation, CPU and other components from front to back in the chassis. At present, the chassis depth of general servers has exceeded 800mm. By adjusting the layout of various functional components in the chassis, servers with different functional focuses can be realized.
The MGX server released by NVIDIA fully adopts modular design. The NVIDIA MGX server provides a modular reference architecture for system manufacturers to manufacture more than 100 server models quickly, economically and efficiently. MGX has designed 8 “Bays” including 1U, 2U, long and short to load different components, and finally install these Bays into the cages in 1U or 2U chassis.
In the context of heterogeneous computing becoming mainstream, the decoupling and coupling ideas of traditional servers around traditional performance subsystems such as CPU and storage have been challenged. BYD Electronics has reconstructed modular servers by adjusting the granularity of functional modules. In particular, the idea of immersion liquid cooling module servers is very different from the idea that the industry has been accustomed to emphasizing in recent years that traditional rack servers can be applied to immersion liquid cooling as a whole with low transformation, which is impressive.
Future-oriented modular server requirements
Based on the understanding of data center application scenarios and server products, as well as the prediction of chip technology and software technology development trends, we have proposed some basic requirements for modular servers:
1. Reduce the types of module mechanical dimensions while meeting the flexible combination of various devices. By defining a size with a larger granularity and meeting the existing data center scenarios, a wider range of versatility can be achieved through fewer types; if it can install CPU memory, 2.5-inch, 3.5-inch, E3.S, E3.L disk bodies, and single-width, double-width, or even 4-width (two double-width cards interconnected) full-length PCIe cards can also be installed, and half-height cards and full-height cards, as well as full-height and full-length cards, etc. must be supported.
2. The module can simultaneously meet the cooling solutions such as air cooling, cold plate, and immersion liquid cooling. At present, modularization usually considers the design of air cooling, including the air duct direction of the heat sink, etc. However, when the components in the module need to support cold plate or immersion liquid cooling, more challenges need to be faced.
3. The module can adopt a simpler and more flexible coupling method. When the module is placed in a chassis, the size of the chassis is the maximum limit of server resource expansion. The coupling method is closely related to the size of the module and the granularity of the module;
4. Module interconnection. When the signal rate is not high, in the scenario of blade servers, the middle board is often used to realize signal interconnection, including PCIe, SAS, Ethernet and other protocols, as well as signal and power supply. In the era of PCIe Gen5, including the future Gen6 and Gen7, continuing to use the middle board interconnection may bring higher costs or poor signal quality. Using copper cable connection within a distance of 1 meter may be a more appropriate solution. In the future, when silicon photonics matures, silicon photonics interconnection may be adopted.
5. Module power supply. Can the module support independent power supply? Can it draw power from other modules? Can it support Power Bus power supply?
6. Smaller wind resistance or flow resistance. When modules are arranged in sequence in the heat dissipation air path or liquid flow path, wind resistance or flow resistance will inevitably increase. Reducing wind resistance also means reducing the power of the cooling fan. At the same time, increasing the ventilation cross-sectional area and reducing wind resistance can further reduce the power consumption of the cooling fan.
