Moore thread: Full-featured GPUs provide comprehensive computing power for future metaverse computing

After 7 months, the domestic GPU unicorn company “Moore Thread” launched the second-generation multi-function GPU “Chunxiao” based on the self-developed MUSA architecture, as well as its first national trend game graphics card MTT S80, MTT S3000 for server applications, meta-computing all-in-one MCCX, series GPU software stacks and application tools and a series of new software and hardware products.

As early as the end of March this year, Moore Thread launched the first generation of GPU “Sudi”, positioned as a low-end processor, mainly for the To B market. In order to further meet the needs of more mainstream users, the second-generation GPU “Chunxiao” launched by the company will focus on the consumer market, meet the needs of high-end gamers and more users for graphics and computing, and further cover the needs of mid-to-high-end and low-end applications.

As the core computing power foundation of artificial intelligence, data centers and other fields in the future, the development of GPU performance and ecology is crucial, but there is still a big gap with international top players. “As a global industry in the IT market, whether the GPU is localized or globalized, the first thing to exercise is the technical strength of the enterprise itself.” Moore Thread CEO Zhang Jianzhong said.

So, after the launch of the first GPU “Sudi”, what market pain points did Moore thread find? What targeted optimizations and upgrades have been made to the second GPU “Chunxiao”, which was launched only 7 months apart? After the press conference, 36Kr and a few other media conducted an exclusive interview with Moore Thread CEO Zhang Jianzhong and CTO Zhang Yubo, while deeply discussing the above issues, we also further understood the company’s development strategy in domestic GPU technology and ecological path.

The following is an exclusive interview conversation between 36Kr and other media and Moore thread, edited:

Q: After the launch of Sudi, what is the feedback of users? What are the key optimizations and upgrades in response to user feedback in this new release?

Zhang Jianzhong: Sudi has a wide range of users and has been used in telecommunications, finance, education and other fields. Customers expect our GPUs to not only support the domestic ecosystem, but also the internationally common ecosystem as soon as possible, which is why we want to launch the second generation of products as soon as possible to meet the needs of mainstream game users.

After the release of the first generation of the product, we showed League of Legends at the press conference and received a lot of letters and feedback from gamers, asking when can I buy a card to play the game? Therefore, we also know that Chinese users especially want to have their own domestic graphics card to satisfy their patriotism.

The MTT S80 we released is equipped with the fastest interface PCIe Gen5, paired with 16GB of GDDR6 high-speed video memory, for gamers, seeing the hardware itself knows that it is a good gaming graphics card. With it, with the constant upgrading of the software driver, it is possible to be competent in many games.

This graphics card is genuine, using advanced processes, are very high configuration, coupled with our use of software-driven non-stop iteration, to provide users with a lot of automatic upgrade startup program methods, such as PES control center, automatic push driver updates after installation, without the user to download themselves to improve the game experience.

Q: Which product is the Chunxiao GPU benchmarking?

Zhang Jianzhong: In fact, we have not benchmarked the products of any company, our product positioning is a full-function GPU, most enterprises do not do full-function, some may focus on games, some focus on AI, is divided into different majors.

However, we believe that the future meta-computing (Metaverse computing) requires comprehensive computing power, and it is difficult to use a single computing power to meet the needs of these users, which is also the most important core of our company’s creation of Moore thread, we want to build a multi-computing platform that empowers the next generation of the Internet to meet the multi-functional application scenarios of “AI + graphics + codec + physical simulation”.

So in theory, our products are not benchmarked against any one company. Our product application functions are very comprehensive, a gamer can not only use their own computer to play games, but also can do AI applications, AI content creation, and even professional picture design. We cannot limit the usage scenarios of users.

Moore thread: Full-featured GPUs provide comprehensive computing power for future metaverse computing

Moore thread MTT S80

Q: Moore thread has optimized the cloud native scene, provided software support in order to avoid GPU waste, is this pain point long-lasting?

Zhang Yubo: First of all, the computing power of a GPU is fixed, there are many GPUs in the cloud, but the demand for GPUs by all cloud users is large and small, which may be much smaller than the computing power that one GPU can provide, or it may be large enough that many GPUs can be combined to meet the demand. In this case, it is necessary to arbitrarily partition a single GPU so that a GPU can serve more users and reduce waste.

For example, a small demand monopolizes a GPU, may only use 20% of the computing power, and the remaining 80% is wasted. If you can put 5 such users together and share a GPU, its needs are just met; Or one user needs 30% of computing power, another user needs 70% of computing power, and they can also be combined to share one GPU instead of occupying two GPUs, which can greatly reduce the cost of data centers, which is just needed for enterprise-level users.

Many existing solutions do not support dynamic arbitrary slicing, many are static, such as only one split in two or one into four, but there is no way to achieve an arbitrary ratio.

Q: Moore thread’s MUSA software stack is compatible with NVIDIA CUDA, will the company continue to follow this path in the future or will it find its own ecology?

Zhang Jianzhong: We are walking two roads at the same time. First of all, we are compatible with industry standard ecosystems, such as Windows, DX, OpenGL, OpenCL, and OpenGL ES, which are standard industry APIs. We are compatible with commonly used AI development platforms such as CUDA, and on the basis of compatible international general development platforms, we are also building our own MUSA platform.

The MUSA developer software stack we launched this time is for developers. Developers can use the MUSA system architecture directly on our platform to develop their own MUSA software. I believe that users have their own choices, and most users can more easily use the development platform of MUSA itself if they want to get the most out of our GPU hardware performance.

Q: Moore thread first-generation GPU Sudi focuses more on the To B field, this first time for the consumer side released the national trend graphics card, so for 2023, what are the company’s key goals or landing plans in the To B and To C markets can be shared with us?

Zhang Jianzhong: Our goal is the same, we are committed to becoming a metacomputing platform company, and we hope to build a metacomputing platform with metacomputing as the core. The company’s development direction is mainly to provide the core technology of Metaverse computing.

The four engines built into the Moore thread full-function GPU will continue to evolve and develop, and we will continue to iterate and innovate our own IP, in graphics computing, AI computing, physical simulation and multimedia encoding and decoding, constantly improve computing performance, to meet the needs of gamers, content creators, designers, programmers, developers, cloud service providers, and even the needs of customers in other industries at the cloud edge, applied in automobiles, and even in some industrial computers or edge servers, the application scenarios are very wide.

Our product application scenarios will continue to expand with the development direction of partner business, and each OEM and ODM, as well as partners, will have their own product lines, and they can serve customers in various industries based on our GPUs.

Q: How do you think the technical route of domestic GPUs should go in the future? If the road of advanced technology does not work, will it consider other ways?

Zhang Jianzhong: We are also looking for a variety of different research and development directions, the good part is that the GPU does not actually need advanced processes, even many GPUs of an international manufacturer stay at 8 nanometers, and do not use the most advanced 3 nanometers. Therefore, there is no need for state-of-the-art in terms of process.

In addition, GPU development is a long road. In addition to the architecture of the hardware itself, the functions of the software are very large, rather than catching up with the hardware itself or adopting the most advanced process, it is better to spend more time to fully build our software strength. Therefore, we will hope to really do a better job of soft power, so that the GPU can play the comprehensive computing power of the GPU in the real application.

In addition, in the entire new application industry, Metaverse applications are the starting point for empowering the next generation of the Internet, and if we can seize these opportunities and create a new computing platform in new application scenarios that others do not have, it may be more beneficial to us.

Q: How does the company define the development stage of domestic GPUs now? What is the most difficult and urgent bottleneck to break through?

Zhang Jianzhong: I think we are still in an initial stage, after all, we only have two years, and long-term development will definitely take a lot of time to polish, whether it is in hardware, software, architecture, or our R&D team. Therefore, we also hope that in the future development process, we can get the support of more industry and ecosystem partners to work with us to jointly promote the development of domestic GPUs.

Posted by:CoinYuppie,Reprinted with attribution to:
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.

Like (0)
Donate Buy me a coffee Buy me a coffee
Previous 2022-11-06 21:56
Next 2022-11-06 22:01

Related articles