The localization of personal computers and servers is intensifying.
From operating systems to memory chips, CPUs and other components, products created by Chinese manufacturers are entering the field of vision of more people. Today, the GPU designed and developed by Chinese companies officially debuted, and Moore Thread officially released the first full-featured GPU chip Sudi with MUSA architecture, and demonstrated the upcoming industry-wide application prospects around graphics computing and AI computing.
Moore Thread's first full-featured GPU is available
Just 18 months after its inception, Moore Thread brought its first GPU product, Sudi.
This is the first GPU (graphics processing unit) based on the Moore ThreadSa architecture, which supports a variety of modern graphics rendering interfaces, including Microsoft's DirectX, which is also the most widely used in desktop environments, Vulkan, which is popular on multiple platforms, the most versatile OpenGL and openGL ES supporting mobile platforms, so it has better compatibility with existing software games.
At present, Su Di has achieved support for global illumination, space-time anti-aliasing, physical menus, soft shadows, reflections, volumetric light and other characteristics. Run League of Legends smoothly without any screen errors. At present, the mainstream AI training and inference engine, AI network model and various applications can run stably and efficiently on devices using Sudi chip.
Multimedia codec performance is also an important indicator of GPU evaluation, Su Di easily supports 4K, 8K resolution video codec, compatible with AV1, H.264, H.265 encoding. It is worth mentioning that Su Di is currently the first GPU to support THE AV1 encoding format, which can transmit a clearer video picture while the size is significantly reduced.
Based on the Sudi GPU, Moore Threads introduced two graphics products for PCs and workstations, as well as for servers (data centers).
The first is the desktop graphics card MTT S60 for PCs and workstations, which is also the first domestic graphics computing card to support the Windows platform. With 2048 MUSA cores up to 1.5GHz, it provides 6T FLOPS of single-precision floating-point computing power, on-board 8GB of video memory, on-board DP1.4 and HDMI 2.1 interfaces and supports maximum 8K video output.
Server-oriented is the MTT S2000, which has 4096 MUSA cores, has 12T FLOPS single-precision floating-point computing capabilities, and supports mainstream deep learning frameworks such as PyTorch, Tensorflow, paddlePaddle and so on. The MTT S2000 is cloud-native, virtualizing every gpu core, monitoring through the cloud and compute center, supporting mainstream container operations and K8s plug-ins, and distributed rendering and computing.
Multi-party partners to build high-quality hardware and applications
For the new hardware products that come out of nowhere, the biggest difficulty is often not the market position of performance specifications, but whether it can use the whole machine products and application scenarios to go to consumers, if there is no actual value, it will be ignored. Moore Thread obviously took this into account, gathered a "circle of friends" for graphics card products equipped with Sudi GPU, and jointly launched products with upstream and downstream partners in the industry.
In order to make moore thread's GPU and graphics card easy to apply in more scenarios, Moore thread has cooperated with a number of partners to establish the PES Perfect Experience System Alliance, which aims to bring high-quality desktop experience to end users.
Moore Thread MTT S60 supports mainstream CPUs such as Intel, AMD, Loongson, Feiteng, and Mega Core, as well as operating systems such as Windows, Kirin, Tongxin, Ubuntu, etc. The pc manufacturers include Lenovo, Inspur, Tsinghua Tongfang, Great Wall Super Cloud, etc. Kingsoft Office, Xiaoyu Yilian, Taiji Graphics, Zhongwang CAD, Guanglianda, D5, Cangqiong Digital, Supermap Software, Yizhirui, Zhongdi Digital, OSG Community and Gala Sports, etc., have brought industry applications based on this graphics card.
Moore Thread MTT S2000 supports x86 and ARM architecture CPUs, and server partners include Inspur, New H3C, Lenovo, Tsinghua Tongfang, Great Wall Super Cloud, Steng Heli, etc. Moore Thread will work with partners in cloud gaming, audio and video codecs and heterogeneous distributed computing to create industry solutions with partners such as WEILing Times, SoundNet and First-class Technology.
Moore Thread has also cooperated with Ruijie, Xinxinfu and H3C to create a variety of cloud desktop environments, which can use mobile phones, notebooks and other devices to access Windows 10 cloud desktops in the cloud in the future. The native Android cloud game acceleration of the graphics card allows the Weiling era to use the MTT S2000 to create a cloud gaming environment with low latency, low power consumption and high cost performance.
Metacomputing is coming, and the Moore thread is ready?
The release of the MUSA architecture and the Sudi GPU and two graphics card products is eye-catching, but this does not seem to be the biggest highlight of Moore Thread's spring 2022 conference. Founded and CEO by former NVIDIA executive Jianzhong Zhang, the company brings together top semiconductor industry talent and wants to bring reliable graphics computing solutions to help digitalize across the industry.
For example, AlphaCore, a next-generation multi-platform GPU physics simulation system independently designed and developed by Moore Thread, has powerful material mechanics calculation performance, which is deeply integrated with existing graphics engines and interfaces, such as Vulkan, CUDA, DirectX, or Unity and Unreal Engine. Whether it is a physical simulation experiment, or the game, film and television, animation industry, you can get a considerable increase in work efficiency.
Zhang Jianzhong also proposed the concept of metacomputer, which is a universal computing power platform to support the next generation of Internet applications, including metacosm. He said: "The era of metacomputing has begun, and multi-functional GPUs are the computing infrastructure of metacomputing and the origin of our innovation. ”
At the same time, a series of full-chain, end-to-end Moore thread meta-computing solutions are presented in order to help the development of the "14th Five-Year Plan" digital economy across the industry.
In terms of digital energy, Moore Thread and State Grid E-commerce Technology jointly promote the application of "artificial intelligence in the power industry" and "industrial meta-universe power scenarios", and jointly build a laboratory to explore the application of GPUs in the field of power. In terms of digital agriculture, Moore Thread and Escher Technology jointly built a digital demonstration planting base, built a digital twin model of agricultural base, and realized standardized and intelligent management.
For digital cities, 3D modeling and rendering can restore the status quo of cities, and Moore Thread is exploring cloud-native rendering and digital twins with Ray Cloud and 51 World. Coming to the current hot digital life, the Moore thread GPU multi-core parallel architecture can accelerate the three-dimensional reconstruction process of cryo-EM proteins, and help medical and pathological research.
There is also the DIGITALME digital human solution, moore thread through visual deep learning technology and TTS speech generation technology, to build a single photo and a small number of audio samples to generate digital human, lightweight solutions. At the press conference, the Moore thread digital person demonstrated that it currently has functions such as expression learning, talent imitation, and explanation of the natural state of a given text.
It can be seen that Moore Threading brings not only high-performance GPUs or graphics products, but also provides digital computing prospects for efficiency and workflow improvement in various industries.
GPUs power a new wave of computing
The chip field has attracted a large number of people of insight to join, but the products are often concentrated in general computing architecture products such as CPUs, few delve into GPU parallel processing architecture products, and the Moore thread that has been joined by a number of industry-rich experienced people is naturally to fill the long-term gap of the GPU, and what surprises us more is that it has brought such a competitive product with one shot.
In the consumer graphics card market, due to well-known reasons, consumers have not been able to buy products with reasonable prices and consistent performance for a long time, and the market gap is huge. Moore Thread with a wide range of compatibility of the product to kill, undoubtedly for the long-awaited graphics card consumers to bring new hope, from the press conference live demonstration, can meet the common game office needs.
Coming to the wider field of accelerated computing, the Moore thread is catching up with the giants. The AI computing acceleration, physical simulation and scientific computing features brought by the MUSA architecture and the Sudi GPU all support mainstream application scenarios, so that various industries seeking alternatives have a foothold, and also provide enterprises that choose a single product but want to find more options, providing new options worth looking forward to.
Moore Thread's products have shown good compatibility with existing general-purpose computing needs, software, and platforms, and computing requirements in various industries can be transferred fairly quickly and at low cost. Perhaps in the near future, the field of GPU computing will form a new pattern, with NVIDIA founded by Chinese on the other side of the ocean, and Moore threads founded by Chinese in the motherland to help China's digital transformation.
From its inception to its official release of the GPU, Moore Threads has shown speeds unprecedented in the semiconductor field. Zhang Jianzhong said that investors did not believe that the first product would be launched in a year and a half, but the full-featured GPU of the real knife and gun has made all this a reality. He expressed his gratitude to partners from all walks of life, co-developed with clear needs, and had a miraculous speed.
On November 25, 2021, Moore Thread announced that it had completed a round of 2 billion yuan of financing, which was jointly led by Shanghai Guosheng Capital, Wuyuan Capital, bohai Zhongsheng Fund under BOC International, and jointly invested by nine well-known institutions such as CCB International, Qianhai Fund, China Merchants Securities and Hubei High-quality Development Industry Fund.
Such a "luxurious" lineup of investors has appeared in previous financing, a startup that has completed three rounds of financing within a year, and behind each round of financing is the full support of strategic investment, state-owned capital and famous financial investment. The financing announced in February last year was jointly led by Shenzhen Venture Capital, Sequoia Capital China Fund and GGV Jiyuan Capital.
According to the previous plan, the funds raised by Moore Thread will focus on the mass production and manufacturing of the first GPU chip, the IP research and development associated with GPU SoCs, and the expansion of the GPU ecosystem. Today, we've seen all of this become a reality, and Moore Thread is creating the power of GPU-based digital computing that belongs to China, and is working together step by step to release energy.
How many more advances and surprises can moore threads bring in the future? In what areas will Chinese GPUs create brilliance? It's worth holding your breath and looking forward to.