The MTT S2000 uses a 12NM process, uses 4096 MUSA cores, configures a maximum of 32GB of video memory, has a single-precision computing power of up to 12TFlops, supports H.264, H.265, AV1 multi-channel HD video encoding and decoding, and a wide range of AI model computing power acceleration. The MTT S2000 adopts passive cooling, single-slot design to meet the high-density GPU configuration mode of data center.
Moore Threads has launched a special unified programming model, runtime, driver and other software tools for hardware architecture for MTT S2000 series products, which can facilitate developers to complete the porting and adaptation of applications and fully call MTT OpenGL ES, DirectX, Vulkan and other graphics APIs; Support audio and video processing ecology through FFAMPEG and VA-API/DXVA compatibility; And through OpenCL and Vulkan to meet the AI and scientific computing program compatibility.
Moore Threads MTT S2000 is compatible with CPU architectures such as X86 and ARM and mainstream Linux operating system distributions, and has begun to cooperate with a number of server partners to adapt to a variety of OEM manufacturers' general-purpose server and GPU server models, which can be deployed in many hardware and application environments.
Specifications | |
MUSA Number of cores | 4096 |
FP32 Hash | 10.6 TFLOPS |
INT8 Hash | 42.4 TOPS |
Encode | 2 MUSA Gen1 Encoder |
Decode | 2 MUSA Gen1 Dncoder |
Memory Capacity | 16/32 GB |
Memory Bit Width | 256 bit |
TDP | 150W |
Dimensions | L 266mm X H 111mm |
Device | PCIE |
![]() | Support for modern graphics rendering engines |
![]() | Support H.264, H.265 and AVI encoding formats Support AVI, H.264, H.265 and other encoding formats while multi-channel concurrent decoding |
![]() | Supports AI computing such as Pytorch, TensorFlow, paddle, etc |
![]() | Supports OpenCL 3.0 parallel computing |
![]() | Support X86, ARM architecture CPU |
![]() | Support Ubuntu, Kirin, Tongxin and other operating systems |
![]() | MT GMI status and performance monitoring |