What will the "chip giants 2019 AI chip contending? "In the article, the author Karl Freund detail the AI chip of giants. In addition, dozens of Silicon Valley Entrepreneurship and China Unicorn have valued more than $ 1 billion, and also participated in the competition of AI chips. In this article, the author will introduce the world's most outstanding, or at least the most attention AI chip startup.
Wave computing
Wave Computing has made a lot of progress in 2018, launched its first DataFlow processing unit, acquiring MIPS, creating MIPS Open, and delivers the first batch of systems to a few customers. Although Wave architecture has some very interesting features, we are more looking forward to the user's large-scale real experience feedback.
Wave is not an accelerator that is inserted into the server, which is a standalone processor for graphics computing. This method is advantageous. From a positive aspect, Wave will not be affected by the memory bottleneck facing the accelerator such as GPU. In terms of negative, the installation WAVE will be a new upgrade, which requires full replacement of the traditional X86 server, which also makes it a competitor of all server manufacturers.
I don't think WAV can beat NVIDIA from some point, but the architecture is very good, the company has said it will soon have a customer feedback.
Figure 1: Wave is a system built from the 4-node "DPU" displayed above. Wave computing
Graphcore
Graphcore is a team of globalized (financing 310 million US dollars, currently estimated $ 1.7 billion), has a global team. It is building a new type of graphics processor architecture that is located on the same chip with its logic unit, which should be capable of achieving higher performance. The posting time of the team's product is not clear, but they expressed "almost ready to release" in April last year, the latest information in December shows that it will soon start production.
Graphcore's investor list is impressive, including sewed capital, BMW, Microsoft, Bosch and Dell Technology.
I learned the company's architecture, it is very impressive. Extended from edge devices to "Colossus" double chip package for training and reasoning for data centers. In the nearest neurips activity, Graphcore demonstrates its Rackscale IPU POD, which provides more than 16 PETAFLOPS in a 32-server rack. Although the company often claims to provide 100 times better GPU stronger than the same class.
Graphcore said 4 "Colossus" GC2 (8 chip) servers can provide 500 TFLOPS (tens of thousands of hours per second) mixed precision performance. A single NVIDIA V100 can provide 125 TFLOPS, so 4 V100 can provide the same performance as its same performance.
As usual, details can be found in the details, and V100 peak performance is only available when the reconstruction code executes Tensorcore's 4x4 matrix multiplication, which is a smart-definition of the GraphCore architecture. Needless to say that V100 consumes 300 watts of electricity and a large amount of cash.
In addition, GraphCore supports the interconnect and "processor memory" method, you can get excellent performance recognized by TFLOPS benchmarks. In some neural networks, such as Geneative Adversarial Networks, memory is bottleneck.
Once again, we will have to wait for real users to use actual applications to evaluate this architecture. Despite this, Graphcore's investor list, expert list and Taiwan price valuation tell me that this may be a good thing.
Figure 2: Graphcore demonstrates photos of ImageNet data set. Visualization helps developers understand their training processing occupation period.
Habana Labs
Habana Labs is an Israeli Entrepreneur Company, which announced in the first AI hardware summit last September that it has been prepared to launch a chip for reasoning, and its record performance is used to convolutional neural network image processing. The result is displayed in the RESNET50 image classification database, which is classified by 15,000 images per second, which is about 50% higher than the T4 of NVIDIA, and the power consumption is only 100 watts.
In December 2018, the latest round of financing of Habana Labs, which is compliant with Intel venture capital, WRV Capital, Bessemer Venture Partners and Battery Ventures, and the company's financing has also increased from $ 45 million. $ 75 million.
It is reported that Habana Labs new financing will partially use a second chip called "Gaudi", which will focus on training markets, allegedly extending to more than 1,000 processors.
Other entrepreneurial companies
I know that more than 40 companies in the world are training and reasoning chips for artificial intelligence design. I have found that most companies are ongoing simple FMA (floating point multiplication) and mixed precision math (integer 8 and floating point 16 and 32). I will not be surprised here because this approach is relatively easy and will get some results, but it does not provide a long-lasting architecture advantage for making a different architecture like NVIDIA, Intel and a few startups.
Here are some companies that causing me to pay attention to:
GROQ: Former Google staff founded, engaged in TPU, they have ambitions to rule the rest of the world.
TenStorrent: Canadian AMD staff founded in the confidential phase. I can only say that his vision and architecture of his CEO have left a deep impression.
Thinci: Indian company focuses on edge equipment and automatic driving, with Samsung and DENSO have established partnerships.
Cerebras: By Seamicro (Lei Feng Network Note, AMD subsidiary, focusing on super-intensive computer server, stopping operations on April 16, 2015) Employee leaders, including Andrew Feldman, is still in depth "stealth" mode.
Mythic: An entrepreneurial company with a unique method for marginalization, similar to analog processing on non-volatile memory; there should be a chip in 2019.
China AI chip startup
China has been trying to find a way to get rid of American semiconductors, and artificial intelligence accelerators may provide opportunities that they have been seeking. China set up a goal of establishing a 10,000-billion US dollars in 2030. Since 2012, investors have invested more than $ 4 billion in funding to entrepreneurship.
The US Congress said this is an artificial intelligence alarm competition. The US science and technology industry may lag behind in China and research institutions because they don't pay much attention to the privacy issues that hinder the progress of Western progress.
Cambricon (Hanwu Technology) and SENSetime may be the most worthy of Chinese artificial intelligent players, but players like Edge Ai are more concerned about Horizon Robotics (Skyline). In addition, large-scale technology companies such as Baidu, Huawei, Tencent and Alibaba are also worthy of attention, all of which have a large investment in artificial intelligence software and hardware.
Han Wu Ji Technology is estimated to be $ 2.5 billion, which has released a Chinese unicorn company in the third generation AI chip. Cambrian said that it can provide better AI performance than NVIDIA V100, and they also sell their AI IP, which is equipped in the processors of Huawei Kirin 970 and Kirin 980 as AI accelerated hardware.
Shang Tang Technology may be the highest value of Ai Entrepreneurship to promote intelligent monitoring cameras in China. These security cameras have exceeded 175 million, including cameras produced by other companies. Shang Tang Technology was established in Hong Kong, and the number of financing has reached $ 600 million in Alibaba. According to reports, the value of this startup is currently $ 4.5 billion.
Shang Tang Technology and Alibaba, Gao Tong, Honda and Even NVIDIA have established strategic partnerships. The company has a supercomputer today, runs approximately 8,000 pieces (may be NVIDIA?) GPU, and plan to build 5 supercomputers to handle the surface identification data acquired by millions of cameras.
Hot recommendation:
NXP LPC54018 Internet of Things Module OM40007, let your schedule fall more quickly
Texas Instrument MSP-EXP430FR2355 LaunchPad Development Kit, allows design development to be simple and intuitive
Microchip Technology SR087 Power Demo Board, Switching Power Supply New Selection
Bosch BMI088 High Performance IMU, improve the flight and navigation experience with you
Roma Semiconductor Sensorshield-Evk-003 Evaluation Kit, a "available"
Intel RealSense Depth Camera D400 Series "More Convenient Secret"
Analog Devices Eval-ADXL362 Evaluation Board to Develop Designers More Choice
Cypress Semiconductor PSoc6 BLE development kit, beautiful design is about to happen
STMicroelectronics VL53L1X flight time ranging sensor, absolute ranging fearless color and reflection
NXP I.MX 8MQUAD Evaluation Kit MCIMX8M-EVK, a high-performance development tool for unpacking
Infineon IM69D120 and IM69D130 Xensiv MEMS microphone, designed for low distortion and high signal-to-noise ratio
Original address: https://www.eeboard.com/news/2018-ai-11/
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]
Our other product:
