What is the core technology of supercomputer

CPU is an important part of supercomputer, but it is not all supercomputer technology. Some people on the Internet equate CPU technology with supercomputer technology, believing that if there is no independent CPU, the core technology will not be mastered. (Recommended learning: PHPSTORM )

This statement is not accurate. In fact, compared to the famous CPU, the unknown architecture design is the key technology that has been ignored by the public and the media.

For example, the Dawn Nebula adopts a self-developed ultra-parallel processing architecture, the Sunway Blu-ray supercomputer adopts a massively parallel processing architecture, and Tianhe-1 adopts a multi-array Configuring a collaborative parallel architecture, Tianhe-2 adopts a self-innovated new heterogeneous polymorphic architecture.

How important is the ability to design architecture?

For example, if a supercomputer is an army, then the architecture design is the army’s historical inheritance, glorious tradition, organizational management level, military training level, weapons and equipment, logistics capabilities, advanced The strategic command level of generals and the tactical command capabilities of middle and lower-level officers, while CPUs are the soldiers of the army.

In addition, supercomputers are not simply stacks of CPUs. Because stacking CPUs is also a technical task. If the system structure is not designed well, the high-speed Internet is not good, the system software is not good, and the storage array is not good, no matter how many CPUs are stacked, the performance of the supercomputer will not be improved. go.

It is simply impossible to create a supercomputer by simply stacking CPUs, let alone a supercomputer that can rival Tianhe-2.

In addition, even if you master the correct method of stacking CPUs, you will not be able to obtain a supercomputer that can rival Tianhe-2 just by stacking the number of CPUs.

What is the reason? Because supercomputer construction is not a simple matter of stacking CPUs like building blocks - even if a large number of computing cards are stacked, it is restricted by other aspects, such as the technical level of high-speed Internet, which makes it impossible to achieve theoretical calculations. performance.

Specifically, the difficulty of high-speed Internet lies in the huge amount of data transmitted between computing nodes of supercomputers and strict delay requirements. When the efficiency of the Internet is insufficient, it will lead to data congestion and greatly reduce the overall performance of supercomputers. machine system efficiency. The more computing nodes a supercomputer has, the higher the requirements for the Internet.

Therefore, even if you want to increase computing power by stacking CPUs, it will be limited by the performance of the Internet. As a result, this approach cannot uncontrollably improve the performance of supercomputers - it is limited by factors such as the Internet and Due to other bottlenecks, the efficiency of the entire machine is lowered, resulting in actual performance not being improved by stacking more CPUs.

In addition, stacking too many CPUs also has problems such as excessive power consumption and excessive chassis size, which is very detrimental to future operation, maintenance and use, and is basically uncompetitive in the supercomputer market.

In terms of software systems, the requirements for software systems to control a small number of computing nodes and to control a large number of computing nodes are almost completely different. The software system must ensure that the performance of each supercomputer computing node is maximized in order to fully tap the potential of the hardware. Otherwise, it will affect the overall efficiency of the supercomputer.

Therefore, without a good architecture, the performance of the CPU will not be fully utilized, and the more CPUs are stacked, the more complex the entire system will be. For high-speed Internet networks, storage arrays, and monitoring The higher the system, cooling system and software requirements, the more difficult it is to improve the overall machine efficiency.

When the level of architecture design capabilities is not high enough, simply stacking the number of CPUs will reduce the overall machine efficiency and fail to improve the overall machine performance.

The above is the detailed content of What is the core technology of supercomputer. For more information, please follow other related articles on the PHP Chinese website!