Domestic cloud service providers need parallel computing services

The author of this paper, Qi Haijiang, Qingdao Five-Pulse Spring Information Co., Ltd. Technical director, University of Pennsylvania Bioengineering, Ph. D., Nanjing University. For many years engaged in graphic images, 3D vision, neural computing, machine learning algorithms such as research.

"Abstract" cloud computing services is essentially a sharing of social intelligence resources, through the cloud of technology packets, reducing the difficulty threshold, so that more users can use "very advanced" technology. China's mobile interconnection new economy is highly prosperous, need to have the corresponding technology high cloud computing service as keel support. The obvious trend of today's computing is:

Video audio Graphics +3d+ scale machine learning + large data analysis = "High-intensity computing task =" cloud GPU parallel operation

And China's cloud service providers, the majority of the current use of too simple and extensive "remote computer room + mobile large hard drive" mode, can not meet the needs of parallel graphics processing calculation. The famous American graphics card manufacturer Nvidia has already successfully launched the GPU virtualization technology based on its CUDA system, and by the Amazon and other U.S. cloud service providers to provide convenient commercial leasing services. I hope that Chinese operators adhering to the "value-oriented" competition route, recognize the technology trend, integration of cutting-edge computing tools, as soon as possible to promote the cloud GPU parallel computing services to promote the overall technical level of China Mobile Internet climbing.

1. The current graphics, image and 3D computing are widely used in various video games, film industry, industrial design, medical imaging, space exploration and remote communication.

With the development of computer technology, people's graphics and image processing requirements are more and more high, especially the emerging 3D technology, so that graphics and image processing and 3D computing has been applied to a variety of video games, film industry, medical imaging, space exploration, remote communications and other aspects.

Now popular large 3D games, such as "Call of Duty" "need for Speed" and so on, these game picture lifelike, 3D special effects strong, so the demand for computer graphics and image processing capacity is very high. The 2010 film "Avatar" created an animated image instead of the actor's 3D film, it perfectly uses the 3D stereo screen to create a realistic effect so that the picture is magnificent. In industrial design, there are many widely known 3D processing software, such as Autocad,maya,solidworks and other well-known software. In medical imaging, 3D/4D Stereo imaging technology enables medical staff to obtain information data which can not be captured from the traditional plane display, can 360-degree omni-Directional stereoscopic reading image information, provides richer and accurate image data for clinical diagnosis, greatly reduces the missed diagnosis of the lesion, improves the diagnosis and treatment quality, will set off a technological revolution in medical imaging information processing.

Along with the development and popularization of it internet and handheld terminals, the explosive growth of data quantity to deal with, the development trend of 3D game on mobile phone also appeared, these all put forward more demand to the data image and 3D computation.

In view of this, the current large demand for graphic images and 3D computing requires the computer to have a strong 3D modeling capability, but the CPU's serial processing capability is far from satisfying the efficient processing of image and 3D computing capabilities, so the use of parallel computing technology is increasingly widespread.

2. The GPU parallel computing technology, represented by the Cuda operation package of the American Nvidia company graphic display card, has become the standard component of workstations, servers and PCs.

The GPU is the microprocessor that is responsible for image operation in computer graphics display card. The famous display card company Nvidia designed a dedicated GPU Parallel computing toolkit for its mainstream graphics products, called Cuda (Compute Unified Device Architecture, unified Computing architecture).

Taking GeForce 8800 GTX as an example, its core has 128 internal processors. With Cuda technology, those internal processors can be colluded into thread processors to solve data-intensive computations. Each internal processor can exchange, synchronize, and share data. Using Nvidia's C-compiler, these features can be leveraged through drivers. can also become a stream processor, allowing applications to perform operations. GeForce 8800 GTX display card can reach 520GFlops, if the SLI system is built, it will reach 1TFlops.

Software vendors have developed an Adobe Premiere Pro plug-in using Cuda technology. Through Plug-ins, users can use the display core to speed up the h.264/mpeg-4 AVC coding speed. Speed is about 7 times times faster than using CPU for software acceleration.

NVIDIA supports CUDA technology from all civilian and professional graphics or operational modules based on G80 and after architecture. The overall computing power is 7 times times higher than the speed of using the CPU alone. Tesla GPU is for workstations and server accelerator, compared with consumer graphics cards and professional graphics card, with a complete double-precision floating-point operation performance, with dual DMA engine to meet two-way PCIe communication, onboard memory up to 12G (Tesla K40 GPU), with specialized Linux patches, InfiniBand drivers and CUDA drivers, CUDA drivers for Windows operating systems can achieve higher performance, TCC drivers can reduce the system overhead of the CUDA kernel and Support Remote Desktop (Windows remote Desktop) and Windows Services

3. GPU Parallel computing, represented by Cuda, has played an important role in many fields

N in the field of scientific research, CUDA is widely used. For example, Cuda is now able to accelerate amber. Amber is a molecular dynamics simulation program that is used by more than 60,000 researchers worldwide in academia and pharmaceutical companies to speed up the exploration of new drugs.

N in financial markets, Numerix and Compatibl released Cuda support for a new counterparty risk application and achieved a 18 times-fold increase in speed. Numerix is widely used by nearly 400 financial institutions.

N in the consumer market, almost every major consumer-level video application has been used by cuda to accelerate or soon use Cuda to accelerate, including elemental technologies, MOTIONDSP and Loilo companies.

4.NVIDIA attaches great importance to the parallel GPU computing on the cloud server, the United States has a number of cloud service providers to provide parallel GPU cloud computing services.

N October 20, 2009, Nvidia and mental images jointly launched a cloud-based high-end server--realityserver.

N May 17, 2012, Nvidia launched the use of GPU to accelerate cloud computing technology.

N October 17, 2012, Nvidia unveiled the first cloud computing virtual GPU Accelerator platform--VGX K2.

N 2013 GTC Conference, NVIDIA brought the latest product server platform--nvidia GRID in the cloud computing field.

In the following years, several server vendors in the United States launched their own cloud services platform based on parallel GPU computing. The service providers that now offer GPU cloud computing are amazon,nimbix,peer 1 Hosting.

Cloud service Provider

Cloud services Provided

Service Price

Service Introduction

Amazon.com

1. Amazon Cluster Super strong graphics calculation example

2. WorkSpace

3. AppStream

1. A small calculation example is USD 0.085 per hour;

2. High memory double Super Large calculation example charge 1.00 dollars per hour;

3. Cluster calculation four times times the large calculation example of 1.60 dollars per hour.

1. This example provides GPU processing capabilities in the cloud. It provides developers and businesses with instant access to highly coordinated GPU computing performance.

2. Cloud Desktop Computing services.

3. Enable graphics-intensive applications to operate on devices that do not have special GPU tools.

Nimbix

High performance computing infrastructure and applications based on cloud computing

Each time around is about to start 2.00 dollars per node/hour, dedicated node 899 dollars a month, free data transmission

HPC applications accelerate computing cloud services through Nimbix

Peer 1 Hosting

GPU Cloud Services (HPC Cloud)

HPC Cloud delivers final professional-level High-performance computing based on NVIDIA GPU card delivery

5. A very puzzling situation is that domestic major cloud service providers (such as Aliyun, Sheng, Wan) seem to have no action on GPU parallel computing.

Since the concept of cloud computing has moved rapidly in China