Moore Threads released the first full-stack functional GPU graphics card for the data center class, the MTT S200.
It fully supports graphics image rendering, video cloud processing, AI, scientific computing, and is also based on the MUSA unified system architecture.
MTT S2000 also uses a 12nm process, MUSA core doubled to 4096, the maximum memory of 32GB, single precision computing power also doubled to 12TFlops, passive heat dissipation, single slot design to meet the high-density configuration of data centers.
It supports OpenGL, OpenGL ES, DirectX, Vulkan and other graphics APIs, supports audio and video processing through FFMPEG, VA-API/DXVA, etc., and supports H.264, H.265, AV1 multi-channel HD video codec, and meets the program compatibility of AI and scientific computing through OpenCL and Vulkan.
In terms of PC cloud desktop, MTT S2000 is the first to have hardware-based MT Mesh 1.0 GPU virtualization technology, which meets the industry standard SR-IOV and can accommodate multiple virtualized users to work remotely on each physical GPU.
Users can bind a GPU to a single virtual machine according to the traditional GPU pass-through scheme to obtain the maximum performance of a single application, or allocate the GPU to many virtual machines through GPU virtualization to improve the utilization rate of computing power resources.
In the virtual machine, MTT S200 supports Both Windows and Linux operating systems, compatible with RenderingX, OpenGL, Vulkan and other rendering APIs, as well as DirectStream technology can quickly complete rendering encoding and concurrently support multi-channel video encoding.
In terms of Android cloud game acceleration, MTT S200 supports ARM architecture, supports kunpeng, ampere, feiteng and other server CPU platforms, natively supports openGL ES interface, ETC/ETC2/ASTC/S3TC and other texture material hardware processing, and the memory bandwidth usage rate is increased by 2.4 times.
It is also equipped with the first generation of Moore Thread Android Cloud Acceleration Platform (ACX), which greatly improves the computing efficiency and system compatibility of Android cloud mobile phones and cloud games through Android driver transmission and GPU hardware encoding.
In terms of video cloud computing, MTT S200 has independent hardware encoders and decoders, supports multi-channel concurrent hardware codecing, supports H.264, H.265, AV1 encoding formats, and supports H.264, H.265, AV1, VP8, VP9.
In terms of cloud application rendering, MTT S2000 has been deeply adapted and optimized for application rendering frameworks such as Unreal and Unity.
In terms of AI-accelerated computing, MTT S2000 supports DBNet, CRNN, FastRCNN, Yolo V2/V3/V5, PSENet, Mask RCNN, Resnet 50/101, Inception, Vgg, Alexnet, Densenet, Unet and other mainstream deep learning algorithms, and supports deep learning frameworks such as PyTorch, TensorFlow, paddlePaddle and so on , support OpenCL, CUDA development environment.
Moore Thread's server partners include: Baixin, Great Wall Chaoyun, Xinhuasan, Yangtze River Computing, Huanghe, Broad Technology, Lenovo, Tsinghua Tongfang, Inspur, Siteng Heli, SanyiYongdao, etc.
Cloud desktop partners include: Ruijie, Xinxinfu, New H3C, etc.