How much does the mlperf list weigh when AI is named?

Digital age , Data are new means of production , Computing power is a new productivity . Many friends are curious , Where does computational power come from ？ An important answer is the server .

It's not hard to find out ,GDP（ GDP ） There is an obvious positive linear correlation with the number of servers , Server shipments can already be used as an important indicator to measure the development level of computing power . In the server market , We noticed a relatively low-key name ： Ning Chang .

not long ago , International authority AI Performance benchmarking MLPerf Training V2.0 in , Ningchang's AI server won the first place in the world in many tests , break 2 World Records . And this is no accident , in fact , Ning Chang has won the world championship five times in a row .

AI The importance of digital economy is self-evident , and AI The demand of algorithm for computing power also increases exponentially . Progress in the field of AI servers , by AI The development and application of technology provide a more solid computing support . On this basis , Maybe you should answer a question ： Ning Chang won the trophy on the AI server , In the age of computing power, the weight is geometric ？

The weight of the trophy ：

The international AI The logic behind the evaluation

Calculation force evaluation , Everyone is no stranger , But the performance of AI server is high or low , But not “ A word doesn't agree with a point ” So simple . One side ,AI Computing is a variety of tasks 、 A collection of schemas , The test is the processing capacity of the whole system , Instead of a single piece of hardware ; in addition , Different tasks 、 Different scenes ,AI The performance of servers may vary widely , Therefore, excellent single task may not represent the overall level , AI Performance evaluation should be divided into tasks 、 Sub track , Just now , It has reference significance .

therefore ,MLPerf Fixed tasks （closed） On the track , Based on the same set of parameters , stay DLRM、MASK R-CNN、ResNet、U-NET3D etc. 8 individual AI scenario , Evaluate each computing system . Evaluate according to different scenes , It means multi Scenario Oriented AI Hardware has more advantages , Endure the difference AI The test of the task . such as MASK R-CNN（ Heavyweight image target detection ） Testing and U-Net3D（ Image segmentation model ） test , It is mainly used in automatic driving 、 Medical image analysis 、 monitor 、 Industrial detection and other scenes , The training task of these models is a typical application of data center , Become various manufacturers PK Main battlefield . This year, , Ningchang AI server X640 G40 Broke his own world record , Respectively carry 16 star 、20 star NVIDIA A100 Tensor Core 80GB GPU, Recapture MASK R-CNN、U-Net3D Two Champions .

in addition ,AI The model is applied to the industry , It is impossible to increase resources without limitation , The best performance under the same configuration , It is of great reference significance for industry users , And this is also AI Unicorns or start-ups stand out from the giants . You can see , On board NVIDIA A30 Tensor Core GPU Under the same configuration , Ningchang server X640 G40 It also shows superior performance , Respectively in Bert（ natural language processing ）、RNN-T（ speech recognition ）、U-Net3D（ Image segmentation model ）、DLRM（ Recommendation system ）、RetinaNet（ object detection ） as well as ResNet（ Image classification ）6 Get the first place in the test .

greening 、 Low carbon has become the general trend , Energy consumption should also be included AI Key consideration areas of hardware . Liquid cooling technology has more advantages than traditional data center air conditioning systems in terms of heat dissipation , It has also become the innovation focus of server products , This time Ningchang X660 G45 LP The liquid cooling server also participated for the first time MLPerf Training test , It shows the accurate refrigeration level at the component level , In lowering the data center PUE、 People are looking forward to energy conservation and emission reduction .

Wisdom lies in all directions ：

From the international arena to a thousand industries

For server vendors , The game is just a training ground , Finally, we should go to the broad market , take AI The technological dividends of spread in all directions . A thousand lines and a hundred industries , Is the real weight behind the results .

With excellent products and services , Ningchang AI server has also been widely recognized by enterprise customers .IDC released 《2021 The second half of the year （H2） China accelerated computing server market analysis report 》 Show , Ning Changyi 8.5% Share of market sales , Ranked in China GPU And the second place in the accelerated server market .

From the stadium to the market , Next, what chain reactions will occur around computational power ？

First , Is the supply of intelligent computing power more abundant , Unit cost reduction , Directly benefit enterprises with intelligent transformation . Take Ningchang as an example , Its AI server products and services have covered the Internet 、 automobile 、 telecom 、 Finance 、 There are more than a thousand industries such as cloud computing TOP User level .

secondly , Push AI Relevant application innovation , Meta universe 、 Number twin 、 New applications such as scientific computing are supported by a computational base , Maybe it will become a reality in the near future .

Further more , Intelligent computing power and digital economy complement each other , Boost the high-speed sustainable development of Digital China . And China's industrial digitalization 、 Intelligent upgrade , It constitutes a deterministic incremental space , Ningchang's AI server provides computing support for the digital economy at the same time , It will also gain more value .

The digital economy requires the convergence of various industrial and technological forces , It's just MLPerf The core value behind , I believe that Chinese manufacturers will continue to write legends in the field and market .