Meeting the Industry Challenge for Fast Inference Hardware
- Centaur Technology’s AI Coprocessor (AIC): Industry’s FIRST high-performance, deep-learning coprocessor integrated into server-class x86 processor
- CT-AIC provides 20 tera-operations-per-second, low-latency compute capability for AI workloads
- Eight high-performance x86 CPU cores and fast memory system support general-purpose computing and boost AI performance/flexibility
- Targets Edge Analytics Servers to run x86 software and DL Inference on multiple streams of data
Innovative Cutting-Edge Technology
Centaur’s microprocessor-design technology targets secure, cost-effective SoC applications. Talented team has over 26 years of experience designing high-performance x86 cores and specialized-processing cores for efficient computing on heterogeneous workloads.
The x86 microprocessor cores deliver high instructions/clock (IPC) for server-class applications
- Supports latest x86 extensions such as AVX 512
- Adds new instructions for fast transfer of AI data
The CT-AI Coprocessor requires minimal extra die area and delivers fast, low-latency inference by sharing the x86 SoC platform. Up to 44 PCIe lanes support optional off-chip inference/training acceleration.
- MLCommonsTM founding member with certified MLPerfTM performance*
- AIC-accelerated x86 SoC achieves impressive latency (0.976 ms) on ResNet-50 and 1.2 ms latency for SSD MobileNet-V1
- Inference throughput equivalent to 24 high-end x86 cores from other vendors (1,281 fps on ResNet-50 and 1,988 fps SSD MobileNet-V1)
The ISC Trade Show Demo
Attendees at the ISC East trade show in NYC saw Centaur’s new technology up close for the first time. The demo showcased video analytics using Centaur’s reference system with x86-based network-video-recording (NVR) software from Qvis Labs. In addition to conventional, real-time object detection/classification, Centaur was the only vendor at the show to highlight leading-edge applications such as semantic segmentation (pixel level image classification) and a new technique for human pose estimation (“stick figures”).