Teddy ParkClick here to view Teddy Park’s profile Teddy Park MEDIUM Head of business development team Published Jun 1, 2020
Accelerated development of artificial intelligence semiconductor chip technology
Graph Core (UK)
Developed machine learning optimized semiconductor IPU, up to 100 times faster than CPU or GPU… Delivery to MS
Medium Inc. develops hardware based blockchain accelerator BPU 10,000 times faster than Bitcoin. It has become ambiguous to complement and overcome the limitations of software performance through hardware, and to differentiate between software and hardware. New Start-ups have emerged solving market performance problems with hardware and they leaded innovation.
CPUs that are evolving based on hardware
Learn about the new startups that lead the world innovation.
Thanks to AlgohGo, the absolute leader of Go, the world was surrounded by AI craze. GPUs optimized for 'parallel processing', a high-speed computing operation required for machine learning, deep learning, and artificial intelligence, have also been spotlighted.
In response, Google made and released a chip optimized for this high-speed computing. It is a Tensor Processing Unit. To improve the algorithm performance, of course, the source code needs to be improved, but Google's TPU has implemented it with hardware algorithm. Using TPU, if we just elevate the chip performance, we can increase total performance. The concept of upgrade of software algorithm and the hardware chip upgrade were battled.
Graphcore IPU (Intelligent Processing Unit)
UK's artificial intelligence (AI) semiconductor startup 'Graphcore' announced that it has decided to deliver an AI-only semiconductor to Microsoft (MS), an American IT technology giant. Microsoft said it plans to mount it on its cloud computing platform 'Azure' to provide customers with a more convenient AI development environment. As a new startup founded in 2016, Graphcore's cooperation with Microsoft is a monumental event for Graphcore. UK Semiconductor Startup Graphcore's IPU Graphcore's AI semiconductor is an IPU specialized in machine learning and the AI data processing speed is 10 to 100 times higher than existing system semiconductors such as CPU or GPU. In fact, the famous French search engine company 'Qwant' announced that the image search speed has increased 3.5 times since applying GraphCore's IPU to the system. Hedge Fund Carmot Capital also showed a 26x improvement in financial information analysis speed through IPU.
When IPU is applied to MS Azure, users can develop new services or products by utilizing machine learning or natural language processing (NLP) in the Azure platform. It means using advanced AI infrastructure without additional investment. "Graphcore will continue to expand Azure's capabilities and remain the best cloud service for AI," said Microsoft Vice President Girish Bablani. Currently, Graphcore's corporate value is valued at over $ 1.5 billion (about 1.75 trillion won). In December last year, it also attracted an additional investment of $ 200 million (about 234.1 billion won). Microsoft and German automaker BMW have joined as investors. Currently, Graphcore is developing with Samsung participating in equity investment.
Medium Inc. is developing BPU.
Medium, a hardware-based blockchain accelerator development company in Korea. Medium inc. solves the problem of blockchain performance scalability through the hardware algorithm. In hardware through BPU (Blockchain Process Unit) in hardware, improving the performance to about 10,000 times more than Bitcoin, Korea And it is rapidly gaining attention from the Chinese government and large enterprises.
Medium BPU
This is the same context that Google developed Tensor Processing Unit (TPU), a hardware chip optimized for TensorFlow. Like Google, which has implemented performance through hardware, Medium has implemented it in the blockchain field.
In addition, Medium and Graphcore are similar in that they are startups that have innovatively improved market problems through hardware. BPU and IPU are quite similar in that they take the same problem-solving method to deal with the performance degradation problem repeatedly occurring in the existing CPU-based data operation processing method through a dedicated processor and a dedicated chip. Medium is currently developing with the goal of implementing MDL5.0 which will be released in Dec. Pieter Willems, Marketing Director of Silex, a global security IP company, said, “In collaboration with Medium, we have advanced the processing speed of blockchain and have been able to further strengthen the stability of the use level. "Medium is a company that can theoretically implement 2 million TPS."
News 
Market Data 
Discover 
AI
