When the AI Gold List is titled, how much is the amount of the MLPERF list?

In the era of digitalization, data is new production data, and computing power is new productivity. Many friends are curious, where does the computing power come from? An important answer is the server.

It is not difficult to find that the GDP (GDP) is significantly positive as a positive linearity. The server shipment can already be used as an important indicator of the level of computing power. In the server market, we noticed a relatively low -key name: Ning Chang.

Not long ago, in the international authoritative AI performance benchmark test MLPERF Training V2.0, Ning Chang's number of artificial intelligence servers has won the world's number one test score and breaks 2 world records. This is not accidental. In fact, Ning Chang has won the world championship five times in a row.

The importance of AI for the digital economy is self -evident, and the needs of the AI ​​algorithm on the computing power also exponentially increased. The progress of the field of artificial intelligence server provides more solid computing power support for the development and application of AI technology. On this basis, it may be possible to answer a question: the trophy that Ning Chang recovered on the artificial intelligence server was geometric in the era of computing power?

The amount of the trophy: the logic behind the international AI evaluation

Computing power evaluation, everyone is not unfamiliar, but the performance of artificial intelligence servers is not as simple as "running a score". On the one hand, AI computing is a collection of various tasks and architectures. The test ability is compared to the entire system's processing capacity, not a single hard piece; in addition, different tasks, different scenarios, and the performance of the AI ​​server may be different, so the single task may not be outstanding. It can represent the overall level. The AI ​​performance evaluation must be divided into tasks and tracks, which is only fair and has reference significance.

Therefore, in the CLOSED track, based on the same set of parameters, in 8 AI scenarios such as DLRM, Mask R-CNN, ResNet, U-Net3D, each computing system is evaluated. The evaluation of different scenarios is different, which means that the AI ​​hardware facing multiple scenes is more advantageous and withstand the test of different AI tasks. For example, Mask R-CNN (Heavyweight Image Target Detection) test and U-Net3D (image segmentation model) test are mainly used in autonomous driving, medical image analysis, monitoring, industrial testing and other scenarios. The training tasks of these models are data centers. Typical applications have become the main battlefield of various manufacturers PK. This year, Ningchang Artificial Intelligence Server X640 G40 broke the world record that he set, and equipped with 16 and 20 NVIDIA A100 Tensor Core 80GB GPUs, and then won the Mask R-CNN and U-Net3D championship.

In addition, the application of AI models to the industry cannot increase resources without restrictions. The best performance is performed under the same configuration. It is very reference for industry users. The path that stood out. It can be seen that under the same configuration of the NVIDIA A30 Tensor Core GPU, the Ningchang server X640 G40 also shows superior performance, in BERT (natural language processing), RNN-T (voice recognition), U-Net3D (image image The division model), DLRM (recommendation system), RETINANET (target detection) and ResNet (image classification) were won first.

Green and low carbonization have become the general trend, and energy consumption should also be included in the key consideration of AI hardware. In terms of heat dissipation, liquid cooling technology is more advantageous than the air -conditioning system of traditional data centers, and it has also become the focus of innovation of server products. This time, Ningchang X660 G45 LP Solid Cold Server has also participated in the MLPERF Training test for the first time, showing the component -level precise refrigeration of the component level The level makes people look forward to reducing the data center PUE and energy saving and emission reduction.

Zhi in Baying: From the international arena to thousands of industries

For server manufacturers, the competition is only a training ground. In the end, it still has to go to the vast market and broadcast AI's technical dividends to all directions. Thousands of industries are the real weight behind the grades.

With excellent products and services, Ningchang artificial intelligence server has also been widely recognized by corporate customers. The "Second Movement of 2021 (H2) China Accelerated Computing Services Market Analysis Report" released by IDC shows that Ning Chang ranked second in the Chinese GPU and accelerated server market with 8.5%of the market sales proportion of 8.5%.

From the field to the market, what chain reactions will happen around the computing power?

First of all, the supply of intelligence computing power is more abundant, the cost of unit is reduced, and enterprises that benefit intelligent transformation directly. Taking Ning Chang as an example, its artificial intelligence server products and services have covered more than a thousand TOP -level users in many industries such as the Internet, automobiles, telecommunications, finance, and cloud computing.

Secondly, promoting AI -related application innovation, new applications such as the Yuan universe, digital twin, and scientific computing have computing power base support, which may be a reality in the near future.

Furthermore, the intelligent computing power and the digital economy complement each other, helping the high -speed sustainable development of digital China. The digital and intelligent upgrade of China's industry constitutes a certain incremental space. Ning Chang's artificial intelligence server provides computing power support for the digital economy, and it will also gain more value.

The digital economy needs to aggregate a variety of industrial forces and technical forces, which is the core value behind MLPERF. I believe that Chinese manufacturers will continue to write legends in the field and the market.

- END -