Shulou(Shulou.com)06/03 Report--

Now the integrator's sales to see customers and then talk about cloud desktops do not have to waste their breath to explain what cloud desktops, should they go to cloud desktops? All we need to talk about is how much more this time. Because everything that can be sold has been sold (including those scenarios that are not suitable for cloud desktops at all), if you want to sell a little more, you have to pile up selling points on the product. So domestic manufacturers began a new round of conceptual innovation. The three biggest slot points are GPU virtualization, superconvergence and front-end computing.

Today, the first thing I want to talk about is GPU virtualization, which is the most hyped concept recently. What is GPU? GPU is the abbreviation of Graphics Processing Unit (graphic processor); GPU is also called display core, visual processor and display chip. The advent of GPU has improved the graphics processing ability of computers and liberated CPU. Equipped with high-performance GPU display card has been widely used in graphics and image processing, industrial design and other professional fields, but also the favorite of gamers.

Students who have a little knowledge of VDI technology all know that VDI transmits the video output of the remote virtual machine to the terminal screen, and the terminal is only responsible for inputting and outputting the information of the remote virtual machine, and the internal principle is similar to the remote desktop. Both Citrix and VMware have made great efforts on the remote desktop transport protocol, constantly updating and improving the transmission efficiency. However, due to the influence of bandwidth and other factors, when we really carry out industrial design, watch high-definition video, 3D games and other applications on the cloud desktop, we will find that the cloud desktop screen will frequently skip frames, pause or even blur screen phenomenon. Therefore, in the early days, it is a consensus that the virtual desktop technology based on VDI is not suitable for image processing-intensive application scenarios.

With the rise of cloud computing, machine learning artificial intelligence, high computing and block chain in recent years, GPU has a huge market demand for its excellent performance in floating-point and parallel computing. The difference between GPU and CPU is that CPU consists of several cores optimized for sequential serial processing, while GPU has a massively parallel computing architecture of thousands of smaller, more efficient cores designed to handle multitasking at the same time. GPU is no longer just a display card, but has become a larger computing module outside of CPU. Many new high-computing servers in university labs are often equipped with 4-8 cards, in order to improve computing power. Some so-called professional excavators also use high-performance GPU as the computing core. The new generation of operating systems and software are more and more dependent on GPU. According to Lakeside, GPU usage of Windows 10 is 32% higher than that of Windows 7, and the number of applications using GPU acceleration has doubled over the past five years, while the number of users using these applications has also increased by 60%. Here we can clearly find that the infrastructure without GPU computing power is difficult to meet the requirements of future software and systems.

The manufacturer exaggerates the effect of VGPU on cloud desktop applications is actually a kind of conceptual hype. Some salespeople even advertise that the thin terminal cloud desktop can be qualified for the work of the original professional graphics workstation through VGPU. Claim that cloud desktops can be used in highly intensive graphics and image processing and design work. But what is the actual situation?

As early as ten years ago, NVIDIA launched the virtual GPU project, that is, physical GPU is cut on demand through the underlying management software of NVIDIA and allocated to multiple virtual machines at the same time. The GPU used by virtual machines is called virtual GPU. The vGPU computing power used in the virtual machine comes from the quota split by the physical GPU, which greatly improves the floating-point computing and parallel computing performance of the virtual machine with the support of strong back-end GPU. This technology is originally designed to serve the professional users of high-computing hosting who rent cloud hosts, and is mainly used in virtual servers and application host servers. Today, we all have VGPU pools to choose from when leasing CVMs from cloud computing providers such as Aliyun and AWS. Users can choose multiple VGPU according to their own needs to improve the computing speed of their hosted application services. It can be said that this technology was not developed by desktop users from the beginning. Although VGPU can theoretically accelerate the graphics and image processing ability of the virtual machine, it is not helpful to the output ability of graphics and images in the remote terminal.

Once again, VGPU technology is used to improve the floating-point computing and parallel computing ability of the virtual machine, and it is not substantially helpful for the remote output of the virtual machine image to the terminal. CVM cloud services focus on computing power; cloud desktops focus on the interactive experience of front-end desktops, and all remote desktop transport protocols lossy compress the image. when the frame output by the remote virtual machine is transmitted to the terminal (box) through the network, the image has actually been cut. Even if the VGPU on the back-end virtual machine renders a 4K 32-bit true color image, it is actually a shrunk copy when it is echoed by the remote Desktop transfer Protocol (the terminal box receives the encrypted and compressed data and restores the data into an image) to the terminal display when it is seen by the user's naked eyes. In fact, the cloud desktop of VDI does not perform well in high-density image processing, and its bottleneck does not lie in the lack of back-end GPU, and the loss of network transmission and the lack of front-end capacity. It is more obvious in narrow bandwidth and thin terminal environment.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.