On October 17, Bitmain officially released the terminal artificial intelligence chip BM1880, along with the SA3 intelligent server based on the cloud artificial intelligence chip BM1682, the embedded AI mini machine SE3, the 3D face recognition intelligent terminal, and development boards, AI modules, computing power sticks and other products based on BM1880.
Bitmain's product strategy director Tang Weiwei said that Bitmain's cloud-based artificial intelligence chips are rapidly iterating at a speed of 9 months. The related products based on the chips have performed well in actual operation, and customers have given positive feedback. Cooperation is being carried out extensively, and many security projects continue to be implemented. At the same time, since the target market applications require end-cloud integrated AI solutions, in order to better meet the needs of customers and the market, we decided to develop and launch end-side AI chips and solutions.
Bitmain's second-generation cloud AI chip BM1682
From cloud to end, fast iteration of chip products in 9 months
The BM1682 chip was released into mass production in the first quarter of 2018. It has a peak computing power of 3TFlops and a power consumption of 30W. It is the second-generation artificial intelligence chip launched by Bitmain for the field of deep learning. It can exist independently from the X86 CPU and support secondary development by customers. It has a single-chip eight-way H.264 and H.265 decoding capability, supports hardware acceleration for video image post-processing, supports Ethernet and PCIE multi-chip interconnection, and is easy to expand horizontally. Compared with the first generation, it has higher density and greatly improved actual performance.
Based on the BM1682 chip, Bitmain has developed the deep learning acceleration card SC3, intelligent server SA3, face recognition server SS3-C3 and embedded AI mini machine SE3, which are widely used in security, Internet, campus and other fields. Bitmain terminal AI chip BM1880
The BM1880 chip was successfully taped out in July 2018. It is a deep learning inference artificial intelligence chip focused on edge applications. It can provide 1TOPS computing power for 8-bit integer operations. Under Winograd convolution acceleration, it supports up to 2TOPS@INT8. The specially designed TPU scheduling engine can effectively provide extremely high-bandwidth data flow for all tensor processor cores. The chip contains 2MB of memory, which can provide the best programming flexibility for performance optimization and data reuse. At the same time, BM1880 also provides users with a powerful deep learning model compiler and software SDK development kit. Mainstream deep learning frameworks such as Caffe and Tensorflow can be easily ported to the BM1880 platform, and common neural network models such as CNN/RNN/DNN can also run.
The BM1880 chip can be used as a coprocessor for deep learning reasoning acceleration, or as a main processor to receive video streams, images or other data from the Ethernet interface or USB interface to perform reasoning and other computer vision tasks; other hosts can also send video streams or image data to BM1880, which performs reasoning and returns the results to the host.
At the media communication meeting, Tang Weiwei also revealed the progress of the next-generation cloud chip BM1684. BM1684 is expected to be launched at the end of the year, with greatly improved performance. The server SA5 based on this chip will also support a significant increase in the number of video structured paths, far exceeding similar products.
Focusing on AI+security, the smart server SA3 has excellent performance Bitmain's SA3 intelligent server is developed based on the second-generation artificial intelligence chip BM1682. The SA3-23 server is a 19-inch standard 2U high-density server with three intelligent processing units, each containing six BM1682 chips. SA3 has the following advantages:
SA3 uses an interconnected cluster architecture of multiple BM1682 chips and has low-power master control technology. It provides services such as cluster management, task scheduling, load balancing, and video image API interfaces. It has parallel computing capabilities for massive video data, and can respond to external requests in a stable, reliable, and timely manner, and is manageable.
The SA3 server based on BM1682 has a higher performance-to-power ratio than servers designed by traditional manufacturers that rely on GPU general-purpose graphics processors. It reduces power consumption while meeting the requirements of high-performance video analysis, and the performance-to-power ratio is improved by more than 2 times.
Smart Server SA3 provides super computing power and reduces the power consumption of the whole machine, while also greatly reducing the spatial density of computing. Compared with the mainstream 4U servers with equivalent computing power in the industry, SA3 has a 2U height, which occupies less cabinet space in actual computer room deployment, allowing customers to choose more flexible deployment solutions, thereby reducing the total cost of ownership (TCO) of deployment in the data center, enabling the industry while reducing the cost of use.
It is worth mentioning that the SA3 video structured server (2U) can support up to 90 video structured channels, with a performance 2.2 times that of traditional industry servers (based on GPU); its face recognition performance is three times that of traditional servers (based on GPU). Suanfeng Intelligent Server SA3
At the same time, Bitmain is also actively expanding the smart campus market and empowering campus scenarios with AI. The embedded AI mini machine SE3 based on BM1682 is compact, with dimensions of only 210mm*115mm*45mm. Its single-precision peak computing power can reach 3 TFLOPS, and its rated power consumption is 60W. A single device can support 4-way dynamic or 10-way static face recognition, far exceeding similar products on the market and widely recognized by customers. Embedded AI Mini PC SE3
In addition, Bitmain has also launched a 3D face recognition smart terminal with an extremely high security level, providing multiple system-level biometric verification technologies such as 3D face recognition, voiceprint recognition, and speech recognition , to solve the security issues of traditional biometric technology due to high false recognition rate, non-living attacks , and insufficient anti-counterfeiting capabilities. It is also the industry's first 3D face recognition smart terminal that supports multi-mode hybrid biometrics . 3D face recognition smart terminal
Directly targeting edge computing, several AI terminals are launched
In addition to cloud-based AI chips, Bitmain is also actively expanding new product lines. Focusing on the terminal AI chip BM1880, it has launched a number of AI terminals for edge computing at one time.
The Suanfeng edge computing development board is designed for a variety of edge computing applications that require powerful deep learning capabilities. With its fast and complete prototyping design, it can quickly help developers complete various types of application development. The Suanfeng edge computing development board contains the BM1880 chip and supports development modules for Arduino and Rasperry Pi. Developers can use ready-made modules to develop leading deep learning or machine vision-related applications, including face detection and recognition, facial expression analysis, object detection and recognition, license plate recognition, voiceprint recognition, etc.
Edge computing development board
In addition, Bitmain has also designed two AI modules and AI computing sticks for various edge applications through USB interfaces. Both products contain BM1880 chips, which add AI functions to traditional products through USB interfaces and can be applied to smart industrial computers, robots, industrial computers and other products.
USB AI computing stick
Actual combat stress testing, gaining market recognition
While rapidly developing and iterating chips and related products, Bitmain is also actively promoting industry cooperation and project implementation.
At the AI Computing System forum of the 2018 Baidu AI Developer Conference, Baidu released its latest AI reasoning acceleration engine, Anakin v0.1.0. The latency data of Bitmain's AI chip BM1682 when running Mobilenet was the best among all the tested hardware, and its performance was about 2-4 times that of mainstream chips in the industry optimized by TensorRT or Anakin.
In terms of security, in 2018, Bitmain signed a city-level strategic cooperation agreement with Fuzhou City, using artificial intelligence technology to assist Fuzhou in building a smart city and a safe city; Bitmain's artificial intelligence products participated in the security algorithm test of the Hubei Provincial Public Security Department, and the test results ranked first; in addition, Bitmain products have also been gradually implemented in security and artificial intelligence projects in other provinces and cities.
In terms of parks, the facial recognition solutions provided by Bitmain have been continuously used for large exhibitions such as the "China Straits Project Achievement Fair" and the "20th China International Investment and Trade Fair", serving as many as 400,000 people, and have performed stably and reliably in such "stress tests". At present, Bitmain's AI+ Park products and solutions have been widely used in factories, industrial parks, office buildings, new retail and other scenarios, and have worked closely with high-end real estate companies.
On October 23, Bitmain will also appear at the Beijing Security Expo, not only showcasing its technological progress and latest products in the security field, but also showcasing solutions and implementation cases jointly implemented with partners. Welcome to visit the E1B04 Bitmain booth during the Beijing Security Expo (October 23-October 26) for guidance. About Bitmain
Bitmain is a leading high-performance computing chip company, whose products are mainly used in the fields of blockchain and artificial intelligence. In the field of blockchain, Bitmain has an absolute leading position in the global market share, and about half of its products are sold abroad. High-performance chip design is Bitmain's core competitiveness, and it has the world's most advanced 7nm process design capabilities.
Bitmain started developing AI chips at the end of 2015, and launched the AI chip BM1680 in the first half of 2017. This is a dedicated computing acceleration chip for deep learning in the cloud, and is the world's first publicly available dedicated chip for this application field. In March this year, the second-generation AI chip BM1682 was launched, which has a performance improvement of about 5 times compared with the first-generation product, leading the international peers.
About Bitmain's AI brand Suanfeng Official website: https://sophon.ai |
>>: Monero will enable Bulletproofs technology after upgrade, transaction size may be reduced by 80%
This week, an unnamed Caribbean island became a s...
Palm Lines Diagram: What to do if there is no lif...
The face of a wealthy man What kind of man is the...
In the past few weeks, the price of Bitcoin has d...
Blockchain startup Digital Asset Holdings has hir...
What does it mean for people who don't have m...
I set out to discover some of the top altcoins th...
What are the types of palm lines that indicate we...
What does it mean if there is no career line on t...
Crazy Review : Crypto Facilities is headquartered...
Physiognomy: The nose tells you your fortune The ...
The cryptocurrency market has entered a free fall...
Women born under the Chengtoutu sign and with the...
As we age, women over 35 and men over 40, the yan...
In fact, hair is related to a person's fortun...