The current world has entered the "Reading" era, and the picture data takes up a huge storage resource of the Internet data center. For this purpose, Google launches compression than the higher WEBP image format, but because the algorithm is more complex, the WEBP image is longer. The wave introduces the WEBP image loss compressed FPGA heterogeneous speed, enable fast conversion of JPEG-WEBP image format, can be upgraded by 14 times compared to the maximum conversion efficiency of traditional programs, and can support higher concurrent density pictures real-time retrieval, transmission, etc. Task.
Image compression acceleration scheme based on the most powerful performance of the most powerful performance of the industry, F10AFPGA
This is a "reading" era
With the development of mobile Internet, the picture is becoming important components in daily information. From a certain extent, we have entered the "Reading" era. According to research, the Internet data generated by global equipment is expected to reach 24.3EB per month in 2019, of which 60% of data traffic comes from the picture. "Reading the era", especially the "big picture era", where is so high quality? How to solve storage costs sharply?
Use computing power to "exchange" storage space
At present, the mainstream of the Internet includes JPEG, GIF, etc., but due to small compression, the image data storage takes up excessive server storage resources.
Webp format greatly narrows the picture size
To this end, Google released a new picture loss compressed format --Webp, which can reduce the data size by 39.8% compared to JPEG, 26% compared to PNG, 26% compared to GIF. After using the webp format picture, the average transmission time is reduced by 33%, and the webpage load time is reduced by 10%.
In China, Tencent's news client, QQ space, etc. also started to switch to WEBP, and traffic peak bandwidth can reduce 9GB, the delay and data download delay is reduced by 100ms. However, the WEBP's application still has thresholds because the WEBP actually requires the calculation capacity to "exchange" storage space.
Due to the more complex compression algorithm, the codec process of the Webp format image needs to consume more computing resources than JPEG, resulting in less than 10 times the processing efficiency. Also, the CPU is not so efficient when processing high parallel tasks, so it is the key to developing a more efficient picture-coded acceleration hardware is the key to the fast landing of WebP.
FPGA accelerated compression algorithm, overall performance can increase 14 times
The codec of a single picture is a small computing amount for computing devices, but this task has a lot of things, and the number of images to be processed at the same time. If you use a serial calculated CPU to process, the processing efficiency is limited to the clock frequency.
Therefore, the wave uses a parallel computational power computing platform FPGA to perform picture codes, and the application acceleration is achieved by hardware the inherent parallel computing in the high performance calculation algorithm. When processing large-scale concurrency tasks, more pipes can be provided, that is, to perform a large number of instructions in parallel. Thus, by reasonable configuration, FPGA can perform a large amount of data throughput at the same time, in which case the device is equivalent to a plurality of conventional CPUs.
Waves WEBP Image Compressed FPGA heterogeneous acceleration scheme, based on the high high density of the industry developed by the wave - F10A. For the compression task of the picture data, embedded the WEBP codec optimization algorithm based on the FPGA computing environment, and greatly enhance the processing performance of the webp image compression coding algorithm by making full use of hardware stream design and task-level parallel.
Wave measured data showing the FPGA scheme than the traditional program efficiency is more than 14 times
According to the wave of wave, use the wave of webp images to compress the FPGA solution, compared to the traditional implementation method, this solution is over the traditional CPU scheme, the overall processing efficiency is increased by 9.13 times, the highest performance Can increase by 14 times from the CPU! This means that the number of servers required for image data storage is reduced by nearly half, and the tasks such as higher concurrent density are supported, transmission and other tasks.
For example, enter a JPEG picture of 1200 20481536 resolution, use dual Xeone5-2690v3 servers to perform WEBP format transcoding, a total of 33.4 seconds, 35 per second. The use of the wave of FPGA acceleration is only 2.39 seconds, 502 sheets can be handled per second, and the acceleration ratio is 14.37 times!
In today's society, the picture has become an indispensable element of the Internet, in the fields of e-commerce, instant messaging, picture search, life service, social platform, etc., the image is gradually increased, and the picture storage and distribution for the operation cost of data center. The pressure is also high, and the wave webp image is used to compress the FPGA program, which will effectively improve the WEBP format image transcoding efficiency, and improve the picture speed of the website, the app, and reduce the hardware resources required for background storage.
At present, the wave has launched the corresponding FPGA solution in the direction of WEBP image compression, Gzip data compression, and residual network-RESNET, compared to traditional application implementations, the wave FPGA scheme has a high performance power consumption ratio. Subsequent, the wave will continue to develop FPGA-based general system solutions, including deep learning, network acceleration, storage optimization, etc., and promote program to other applications and customers. In the future, CPU + FPGA may be used as new heterogeneous acceleration modes, more and more data centers, internet depth learning, etc.
Original link: https://www.eeboard.com/news/cpu-23/
Search for the panel network, pay attention, daily update development board, intelligent hardware, open source hardware, activity and other information can make you master. Recommended attention!
[WeChat scanning picture can be directly paid]
Technology early know:
After the old two, the boss, relied on ridicule skills to preheat MATE 10, Huawei this time is easy to play.
The new generation of domestic baseband and radio frequency integrated new generation of intellectual property rights, officially released
This 90 Chinese guy developed the world's first vertical take-off flight car
Sweep one, smarter! Xiaomi released a new rice scan robot
Also 10nm process, transistor density is twice that of Samsung and Tablet, Intel's 10nm process is invincible!
Our other product:
