Recently, the Internet celebrity robot has become popular, and a large number of men, women, old and young have been captured by him. Every day, many people rush to brush its short videos and leave comments to express their love for it.

But at the same time, it also attracted many doubts. Some netizens said that someone was manipulating in the background. What was the matter? Will it be operated artificially, as questioned by netizens? In fact, these are typical cases of such disciplines as deep learning and machine language. The so-called "learning" is a process of acquiring and storing data. What is obtained here is voice data, and behind it is a series of algorithms. These algorithms are actually a process of deep learning. In the process of dialogue, robot Kuaibao will obtain the answer content from cloud services through big data, and give the corresponding dialogue content. So Kuaibao is "feeding" big data. After the machine is finished, it needs to "train" to achieve optimization. Otherwise, it will be impossible to answer questions and talk.

When Kuaibao hears someone talking, it enables the voice recognition function and compares the person's words with the voice data in its own database. If the comparison is successful, it means that you have "understood". Then it calls the existing voice data to reply. If it is unsuccessful, it will reply to something else. If it digs from the topic, it is all human words anyway. Here, Kuaibao can know people. This is machine vision and machine language. The camera is his eye, but the eye can only see things, not understand them. Only when it corresponds to the image data in the database can he understand. He knows that this is a person or a cat.. This is the same as the learning process of children. They rely on the existing knowledge in their heads to reason and understand new things. However, people have thinking, but Kuaibao has no thinking. It just calls the database from "learning" to achieve a "communication" function. It makes people feel that it can think, just like cheating in an exam. The teacher looks at the exam paper and thinks that the students will, but in fact, the students will not be the same.

What are the above applications implemented by? That is the truth we want to decrypt today. In fact, it is very simple. With a high-performance server, we can easily implement the application of the above disciplines. Today, we recommend a server 4029GP-TVRT used by Beiliande in the field of deep learning and machine language applications.

Front view                                                                                         Rear view

4029GP-TVRT adopts 4U rack structure, which is divided into upper and lower layers and two 2U nodes.

Up to 8 NVIDIA® can be installed on the upper node; Tesla® For the V100 GPU, the channels between each GPU and the GPU are interconnected, and the maximum bandwidth of 300GB/s is guaranteed, providing our servers with powerful parallel computing capabilities.

PCB boards are painted with classic green paint, symbolizing energy conservation and environmental protection. There are 16 interfaces on the motherboard, and one Tesla® can be installed for every 2 groups of interfaces; V100; Each GPU is equipped with 8 passive radiators; In front of the node are four 4Pin 80cm 13500 rpm intelligent temperature controlled cooling fans; There is a black wind deflector on both sides, which can better heat the GPU. Behind the node is the card slot interface that connects to the chassis backplane, eliminating the trouble of various wires.

The lower node is the control node. The same green panel supports dual Intel ®; Xeon® Scalable processor; It supports up to 6TB DDR4-2933MHz full buffer memory and 24 memory slots.

In the middle of the node is the hard disk backplane, which can support 16 2.5 "SAS/SATA hard disks or SSDs, of which the last 8 drawers can choose NVME interfaces.

On the left side of the front end of the node is the I/O output port. There are two RJ45 10 Gigabit Ethernet interfaces and one RJ45 IPMI special interface for remote management; 2 USB 3.0 interfaces and 1 VGA display output interface. On the right are the panels for switches and status lights, as well as two expansion slots. There are 16 hot swappable hard disk bays in the middle. The back end is the card slot interface like the upper node.

After looking at the two nodes, let's finally take a look at the box. The box is 4U high. Looking inside the box, it is the backplane that connects the entire system. It mainly connects two nodes, the cooling system and the power supply. The rear end of the box is a cooling system composed of eight 92cm cooling fans, four of which have the function of power supply. The 2200W titanium level certified (96%+) redundant power supply adopts the 2+2 mode, which means that only two power supplies work at ordinary times, and the other two power supplies play a backup role. When the power supply fails, the backup power supply will automatically take over the work to ensure the stable operation of the server.

It is such an invisible server that can make Kuaibao a popular online robot. Of course, such servers can not only be used for deep learning or machine language, but also serve people in many cutting-edge fields.

If you want to pay more attention to the server application market, please join Beiliande.