Author | ZeR0
Edit | Shadow of indifference
Chipthing reported on June 1 that the GPU track is becoming more and more lively, today, the domestic GPU has added a new Ding - full-featured GPU unicorn Moore thread for games and video scenes to create MTT S70 graphics card officially went on sale, priced at 2499 yuan.
After the explosion of generative AI, the domestic market has become more and more urgent for domestic GPU products. On this occasion, Moore Thread released 19 new GPU products or technology updates for four major application scenarios of digital office, entertainment and creation, AI and cloud computing, and metaverse, and the commercialization process has begun to bear fruit.
Oriented to gaming and creation, Moore Thread launched the new MTT S70 graphics card and became the first domestic GPU company that can support DirectX 11 games, based on the DirectX 11 Community Edition driver, allowing users to play 3A-level games such as Genshin and Dota 2. In addition, Moore Thread also released the first intelligent terminal product "Smart Entertainment Motorcycle Fang" based on domestic GPUs and integrating games and generative AI applications, and its GPU physics engine AlphaCore was also newly upgraded and opened for test download.
For localized digital office, Moore Thread GPU adds support for OpenGL4.0 and Vulkan 1.3, DKMS technology can quickly adapt to various CPU+ operating system version combinations, and launched cloud desktop products MT vGPU 2.1 and MCCX VDI cloud desktop all-in-one.
Facing AI and cloud computing, Moore Thread launched the MUSA Community Developer Program, released the MUSA Toolkit 1.0 software toolkit and code porting tool MUSIFY, which is fully compatible with CUDA, and open-source MT PyTorch, supporting large-scale model inference and training of typical NLP language models.
Facing the metaverse, Moore thread announced that the MTVERSE metaverse platform has been upgraded, which can support real-time rendering in the cloud; AIGC content creation platform Mobi Maliang internal test launched; AND DEMONSTRATED THE UPGRADED VERSION OF DIGITALME'S DIGITAL HUMAN SOLUTION CAPABILITIES.
In the online live broadcast of this summer conference, Moore Thread also showed the strength of "AIGC generation", and the whole process was made by Zhang Jianzhong, founder and CEO of Moore Thread, as a 2D digital human stand-in.
▲ Zhang Jianzhong, founder and CEO of Moore Thread, hosted the online conference as a 2D digital human double
After the press conference, Zhang Jianzhong had in-depth exchanges with media such as Xindong and further interpreted his thoughts on new products and technological upgrades. When talking about product positioning, he said that enthusiasts who are not poor in money will definitely buy 4090 this flagship product, considering the market demand, Moore thread first cut from the mainstream products, in the application scenario to meet the performance and experience of user needs. Providing a better cost-effective solution is where the opportunity for Moore threads lies. "Artificial intelligence + 3D graphics + cloud native + cloud computing power, this is our goal." Zhang Jianzhong said.
01.
Game: The first to support DX11 in China
The whole product supports AI learning and application development
According to Zhang Jianzhong, since Moore Thread released the first game graphics card MTT S80, in the past six months, its graphics card Windows driver has been upgraded 5 times, supporting more than 60 games, completing support for more than 50 motherboards and more than 30 monitors, and the compatibility has been greatly improved; Game performance has improved by about 50% on average, and mainstream online games such as League of Legends and CrossFire perform even better on their game graphics cards.
On this basis, Moore Thread released a new game graphics card with a slightly lower configuration and cheaper price than the S80, MTT S70.
MTT S70 is equipped with Moore Thread's second full-featured GPU chip "Chunxiao", which contains 3584 MUSA computing cores, can provide 11.2T FP32 computing power at 1.6G main frequency, is equipped with 7GB of high-speed GDDR6 video memory, theoretical bandwidth of 392GB/s, supports up to four 8K30 frames of ultra-high-definition display output, supports clipping, OBS and other video editing and live broadcast software to enable hardware-accelerated encoding and decoding.
It is worth mentioning that its Windows driver, version number v221.31, released yesterday, has implemented support for MTT S70 gaming graphics cards. The DirectX 11 Community Edition driver, which will be released in late June, will also support MTT S70 and S80 users to experience triple-A games such as Genshin and Dota 2 in early access.
Supporting DirectX 11 games, Moore Thread is the first domestic GPU. Zhang Jianzhong told Xindong that in order to support DirectX 11, it is first necessary to achieve hardware-level support on the GPU architecture, and it also takes a huge amount of software workload to achieve one-to-one adaptation to mainstream games.
At the same time, the Moore thread launched "Alpha Action" in the "Mocha Player" community to gather more players to use and feedback on the DirectX 11 Community Edition driver to accelerate the iteration speed of the driver.
Moore thread launched the out-of-the-box product "Smart Entertainment Motorcycle Fang". Equipped with MTT S80 and MTT S70 full-featured GPU graphics cards, "Smart Entertainment Motorcycle" is pre-installed with a selected game center and PES system management center, which can not only play games, but also support generative AI learning and application development. Moore Thread will also open up Ubuntu driver downloads for users.
Moore Thread also upgraded its self-developed next-generation GPU multi-platform physics engine, AlphaCore.
The new version of Catalyst FX, a real-time fluid simulation tool developed based on this engine, can directly produce fluid effects in Houdini without changing the original workflow, and the performance is improved by 5-10 times compared to the original PyroFX; Compared to the traditional Houdini Vellume production pipeline, VeraFiber, a flexible body simulation tool accelerated by Moore thread GPUs, can improve the solution efficiency by 3-5 times.
AlphaCore has also made deep optimizations to the Compute Shader compute version of DirectX 11, and in terms of fluid dynamics simulation, the actual performance of Catalyst FX on the MTT S80 has reached more than twice that of mainstream high-end graphics cards in the market.
At present, Catalyst FX and VeraFiber have completed the development of the Houdini plug-in interface, and the beta version of the Houdini plug-in will be available for download on June 6.
It is worth mentioning that the Catalyst FX Houdini version plug-in has been delivered and docked with MOREVFX, a famous domestic film post-effects production company. AlphaCore also co-developed with DOVFX, a CG animation production team invested by NetEase Games, and successfully applied VeraFiber to the fabric and hair simulation of complex characters in the game's CG title.
02.
Localized digital office: accelerate nearly 100 domestic applications
Launched GPU Cloud Desktop All-in-One Machine
Facing the field of localized digital office, Moore Thread hopes to provide a wide range of compatibility and new solutions to promote domestic PC products from "usable" to "easy to use".
Moore thread has successively completed the product introduction of more than 10 ODMs and 18 OEM manufacturers, and jointly launched more than 30 PC models, which are used in key industries of the digital economy such as telecommunications, finance, and energy. Zhang Jianzhong said that the main demand for domestic digital office applications is smooth operation, faster speed and better experience.
In response to the graphics effect that is widely concerned by domestic software, Moore Thread GPU has added support for the full functions of OpenGL4.0 and Vulkan 1.3, and has passed the interface compatibility test 100%, which can achieve more efficient graphics rendering. With new graphics features such as tessellation, Moore-threaded GPUs also provide finer geometry for applications.
▲ Comparison of the picture effect of turning Tessellation on and off
Due to the rapid development of the current domestic office ecology, there are hundreds of combinations of various CPU and operating system versions, and it is extremely difficult and complex to ensure that users maintain long-term stable operation when using different office applications. In this regard, the driver software of Moore thread adds DKMS technology, which can quickly adapt to the cores of different operating systems, and improve the development efficiency by dozens of times.
Moore Thread has recently successfully adapted to domestic operating systems such as Kirin , UnionTech , openKylin, deepin, Ningsi, Zhongke Fangde, and Puhua, and took the lead in completing comprehensive compatibility certification with UnionTech and Kirin OS, becoming the first GPU company in China to pass UnionTech UHQL quality certification.
Through the continuous optimization of the driver, the overall performance of Moore Thread GPU has been improved by more than 2 times compared with the initial release, and its GPU hardware products have accelerated nearly 100 domestic applications, including office, video conference, audio-visual, browser, video editing, design and GIS software.
Zhang Jianzhong said that at present, based on the national production platform, compared with similar representative products at home and abroad, the performance of Moore thread full-featured GPU has been improved by 2-3 times in various standard evaluation items.
With the increasing demand of enterprises for cloud migration, traditional CPU cloud desktop solutions are difficult to meet the needs of new office and rich media applications. In this regard, the China Academy of Information and Communications Technology, together with Mobile Cloud, China Telecom Research Institute, and Moore Thread, released the White Paper on the Development of New GPU Cloud Desktops, which clearly defines the experience standards for cloud desktop scenarios.
Based on this standard, in the four main scenarios of video playback, web browsing, Office office and education and teaching software, a single server equipped with Moore thread MTT S2000 GPU can support more than 40 high-definition users at the same time, compared with the traditional CPU-only solution, the performance is improved by nearly 5 times, and the overall TCO is reduced by more than 60%.
In terms of vGPU innovation, Moore Thread launched cloud desktop products MT vGPU 2.1 and MCCX VDI Cloud Desktop All-in-One.
MT vGPU 2.1 cloud desktop products add support for Moore thread MTT S3000 graphics card, the number of concurrent virtual machines on a single card reaches 28, the concurrent performance is increased by 40% compared with the previous one, support for 4K picture quality, adds support for Windows Server Edition, expands the scope of hardware capabilities and application support, and adds GPU super-resolution technology and support for SRIOV.
Over-partitioning technology doubles the number of virtual machines without changing hardware resources, significantly reducing customer costs, while SRIOV technology provides better QoS, isolation, and security to protect customer data from compromise.
Moore Thread MCCX VDI Cloud Desktop All-in-One is mainly used in education and office scenarios, is an end-to-end delivery solution including servers, thin terminals and software, and specially launched the special edition of education reform and the enhanced version of office experience. The special edition program can effectively accelerate 19 types of education reform software, and the enhanced version of office experience has customized and optimized more than 60 office software such as WPS, PDF reader, video conferencing and decompression software.
At present, Moore Thread has completed the adaptation of products with more than 10 customers, including Tianyi Cloud Computer and Mobile Cloud PC, and has completed product introduction with SINF, New H3C Information, Huayun, Kuzhan and Tingyu Technology, and cloud desktop products and solutions based on Moore Thread vGPU will be implemented in all walks of life.
03.
AI+Cloud Computing: CUDA compatible
Zero-cost CUDA code migration
GPU application innovation is inseparable from the contribution and support of developers. Moore Thread last year released the MUSA Metacomputing Unified System Architecture, including a unified programming model, software runtime, driver framework, instruction set architecture, and chip architecture.
Yesterday, Moore Thread announced a series of technical updates around the MUSA architecture, including the launch of the MUSA Community Developer Program, the launch of the MUSA Toolkit 1.0 software toolkit and the "MUSIFY" code porting tool, and the open source MT PyTorch.
After Moore Thread launches the MUSA Community Developer Program, it will provide partners and developers with a full set of MUSA development tools, programming guides, series of tutorials, open source frameworks and model libraries and other resources; It will also work with third-party communities to drive the development of new algorithmic models, computing systems, and platforms.
In order to make full use of the computing power of the Moore Thread GPU, Moore Thread launched the MUSA Toolkit 1.0 software toolkit, which includes MUSA drivers, runtime libraries, compilers, AI acceleration libraries, math libraries, communication libraries and other tools.
Since MUSA is architecturally compatible with CUDA, its code porting tool MUSIFY can quickly migrate existing CUDA programs to MUSA, complete the automatic migration of CUDA code at zero cost, and then users can complete hot spot analysis and targeted optimization in a short time, greatly shortening the migration optimization cycle.
In addition, Moore thread will open source MT PyTorch, support inference of multiple models, covering CV, NLP, speech and other fields, and be able to run typical large-model distributed multi-card inference such as ChatGLM, Stable Diffusion, LLaMA, etc.
Using distributed training techniques such as data parallelism, model parallelism, and ZERO, MT PyTorch can also complete the training of simple basic models and NLP language models with typical transformer structures.
Based on MUSA, developers can reuse a large number of model operators from the PyTorch open source community to reduce development costs.
04.
Metaverse: AI painting platform closed beta online
Generate digital humans to fake the real thing
Combining capabilities such as graphics rendering, physical simulation, AI training, and reasoning, Moore Thread continues to improve core elements such as people, scenes, and content in metaverse applications, and brings a series of major upgrades to better empower industries such as digital cities, digital education, digital healthcare, digital energy, and digital office.
ZHANG JIANZHONG ANNOUNCED THAT MOORE THREAD UPGRADED THE MOVERSE METAVERSE PLATFORM AND DIGITALME DIGITAL HUMAN SOLUTION, AND ANNOUNCED THE LAUNCH OF ITS AIGC CONTENT CREATION PLATFORM, THE MABI MALANG INTERNAL BETA.
The MTVERSE metaverse platform provides scalable performance, real-time rendering and simulation in the cloud, and AI-driven diversified computing power support.
VNET Internet, a leading third-party IDC service provider, took the lead in deploying a Moore thread kilocard-level GPU computing power cluster in the cloud, and successfully combined the MTVERSE platform with Unreal Engine and cloud rendering streaming technology to provide computing acceleration for 51WORLD's 51Meet metaverse high-precision open platform.
According to reports, this is also the first metaverse application to achieve a domestic closed-loop. With multiple people concurrently, users can enjoy a low-latency, high-fidelity, and immersive metaverse experience. In addition, including Migu Metaverse, Zhihui Yunzhou video twin, panoramic view 3D reconstruction, etc. are all being updated one after another.
Moore's AIGC content creation platform, which integrates software and hardware, was launched.
The platform provides a variety of access methods, which can be logged in through the web and mini programs, and can also be called remotely by users through rich APIs. For users who need a complete solution.
Moore Thread provides private deployment capabilities, including GPU clusters, heterogeneous computing power scheduling platforms, API interfaces, and boilerplate applications similar to Mobi Ma Liang.
Zhang Jianzhong told Xindong that the AIGC content creation platform is a value-added service provided by Moore Thread to provide users, and Moore Thread does not consider making a profit, but hopes to show product capabilities to more developers, so that developers can use these software technologies and hardware product solutions to develop their own business.
IN ADDITION, FOR DIGITAL HUMAN GENERATION APPLICATIONS, THE DIGITALME DIGITAL HUMAN SOLUTION CREATED BY MOORE THREAD HAS BEEN FURTHER UPGRADED, INCLUDING 2D AND 3D.
DIGITALME includes the "Nuwa" digital human generator, the "Paint Skin" expression-driven engine, the "Suiying" action-driven engine and the "Follow-and-answer" dialogue system. Among them, "Suianswer" ushered in the upgrade of two major main capabilities: one is to interact with people naturally through voice, realizing the function of "being able to hear and speak"; The second is intelligent question answering based on the big language model, which realizes the function of "thinking and saying things".
Moore Thread also demonstrated two digital human product solutions based on the DIGITALME technology platform at the press conference - 2D broadcast digital human and 3D interactive digital human, and said that it looks forward to working with partners in various industries to create digital humans with multiple identity functions, which are widely used in live broadcast, social networking, film and television animation, office, entertainment and other scenarios.
▲Based on the DIGITALME platform, the digital human is reproduced with real people and timbre
05.
Conclusion: GPU development and application is a long-term business
The application of GPU is the "touchstone" to test technological innovation. From the perspective of the rhythm of releasing new products and technical updates, Moore Thread is not a boring route, but chooses a more practical way of accumulating experience while iteratively optimizing products, and verifies the stability, reliability and ease of use of its software and hardware products through cooperation with a wide range of ecological partners.
Zhang Jianzhong said that by integrating more distinctive self-developed technologies in GPUs, and investing heavily in adaptation support for many domestic game applications, game engines, operating systems and office software, Moore threads are building differentiated competitive barriers and providing users with some experiences that are not inferior to professional high-end graphics cards at lower prices.
In Zhang Jianzhong's view, AIGC, digital humans, metaverse 3D image generation, autonomous driving, robots, digital twins, etc. are all related to "AI+graphics + metaverse" application scenarios. The development of these applications is inseparable from the GPU, which has become the "infrastructure" of all scientific research enterprises and the necessary hardware in the research and development process of enterprises. This is also the comprehensive application direction he encourages developers to explore more.
He believes that compared with dedicated chips, the versatility of GPUs is more conducive to opening up the creativity of many developers, and a full-featured GPU can take into account the development of large models, virtual digital humans, 3D graphics rendering, industrial digital twins and other multi-modal tasks that have high demand for graphics and AI computing, thus becoming an innovation engine for various industries.
"Our strategy is to provide customers with more efficient, reliable, and better-to-use GPU solutions through continuous and rapid meta-computing innovation and the support of ecosystem partners." Zhang Jianzhong looks forward to working with partners from all walks of life to apply innovative achievements to practical scenarios and promote the wide application of GPUs in the digital economy era."