-
TF7000 Series High Performance Central Processor
TF7000 is designed for “general purpose computing + AI inference” cloud / edge computing scenario. The processor uses heterogeneous computing and compute-storage coupled architecture to improve performance. It supports all main stream Linux OS and virtualization solution.
40 / 80 cores ARMv8.2 64bit CPU@2.5~3.0GHzBuilt-in inference NPU with patented ManyCore architecture, 40/80TOPS@INT84x/8x DDR4@3200 / DDR5@4800 channel, support 1DCP/2DCP64x/96x lane PCIe 4.0 bus64x/128x 1080P@30FPS H.264/H.265 video hardware decoder10x/20x 1080P@30FPS H.264/H.265 video hardware encoderUp to 4x NUMA interconnection -
Single Processor Motherboard
Standard Micro-ATX / ATX model, easy to fit in standard rack / tower server chassis
Integrate single TF7000 series central processor4x/8x DDR4 RDIMM slotsMulitple PCIe 4.0 slots, support GPU and peripherals1x M.2 PCIe 4.0 x4 NVMe SSD slotIntegrate dual GbE data portsIntegrate server-level BMC chip, single GbE management port with IPMI protocolATX power socket and PMBus for power reading -
Customized Server Motherboard
Customized motherboard design adapted to client's server chassis
Integrate dual TF7000 series central processors16x DDR4 RDIMM slotsMulitple PCIe 4.0 slots, support GPU and peripherals1x M.2 PCIe 4.0 x4 NVMe SSD slotIntegrate dual GbE data portsIntegrate server-level BMC chip, single GbE management port with IPMI protocolCRPS redundant PSU sockets -
TF16000 Series Central Processor
The series is designed for “general purpose computing + visual analytics” cloud / edge computing scenario. Multiple computing clusters support various AI algorithms running in parallel.
12/20 cores ARMv8 64bit CPU@1.8~2.4GHzBuilt-in inference NPU with patented ManyCore architecture, up to 12.8TOPS@INT85x LPDDR4X@4266 channel8x lane PCIe 3.0 bus64x 1080P@25FPS H.264/H.265 video hardware decoderRGMII, USB 2.0, UART, SPI, I2C, GPIO -
Inference Acceleration Module
NNM100The module is designed for scalable deep learning inference computing at cloud. It integrates one of our high performance heterogenous processor TF16110, which supports various AI algorithms under mainstream deep learning frameworks and virtualization. AI computing server clusters can be elasticly built upon this compact computing cell.
Quad-core ARMv8 64bit CPU@2.4GHz4x ManyCoreTM heterogenous computing clusters, 12.8TOPS@INT818GB LPDDR4X-4266 + 64/128GB eMMC Flash64x 1080P@25FPS / 16x 4K@25FPS hardware video decodingHigh-speed on board connector with PCIE 3.0 and GbEPeak power consumption 25W, operating temperature 0~55°C -
Inference Acceleration Card
NNC100Integrating the high performance heterogenous processor TF16110, this card is compatible with all kinds of standard scale-out rack servers and can provide application layer service API.
12.8TOPS@INT818GB LPDDR4X-4266 + 64GB eMMC Flash64x 1080P@25FPS /16x 4K@25FPS video decodingPCIe 3.0 x8 half height half length single slot cardCompatible with Windows and Linux host operating systemPeak power consumption 25W, operating temperature 0~55°C -
Video Analytics Server
VAS100VAS100 is a dedicated video analytics sever built upon NNM100 modules, providing ultimate video analysis throughput and power efficiency. The server can be flexibly deployed in either data centers or edge computing nodes.
Standard 19 inch 1U rack serverUp to 6x built-in NNM100 modules2x GbE for data exchange + 1x GbE for managementUp to 144 HD video streams’ real-time facial recognitionSupport seamless system upgrade and expansion with virtualizationDual power hot backup, peak power consumption 200W -
Edge Analytics Workstation
EAS50/EAS100Integrating the versatile heterogenous processor TF16005, EAS50/EAS100 is designed for mixed workload of transaction processing and deep learning. It has rich peripheral interfaces and strong environment adaptability. Industry customized models are available upon request.
Quad-core ARMv8 64bit CPU@1.8GHz2x ManyCoreTM heterogenous computing clusters, 4.8TOPS@INT810GB LPDDR4X-3200 + 64/128GB Flash16x 1080P@25FPS / 4x 4K@25FPS hardware video decoding1x GbE, USB2.0/3.0, RS232/RS485, GPIO, HDMI video outputPeak power consumption 15~25W, operating temperature -20~85°C
