D-Matrix introduces AI network accelerator card for ultra-low-latency inference (siliconangle.com)

D-Matrix has launched JetStream, a custom AI network accelerator card designed to deliver ultra-low-latency inference for data centers. As generative AI models grow larger and more distributed across multiple machines, high-speed networking becomes critical. JetStream aims to address this by providing a PCIe Gen5 card capable of up to 400 Gbps bandwidth, compatible with standard Ethernet switches and without requiring specialized infrastructure upgrades. This plug-and-play approach allows data centers to scale AI workloads efficiently while maintaining compatibility with existing hardware. The significance of JetStream lies in its integration with D-Matrix’s Corsair compute accelerators and Aviator software, forming a holistic ecosystem that tackles memory, compute, and networking bottlenecks in AI inference. Compared to traditional GPU-based deployments, the combined solution promises up to 10 times faster inference speeds, triple the cost performance, and threefold energy efficiency gains. This innovation is particularly important for cloud providers and enterprises seeking to deliver hyper-fast, multimodal AI applications at scale without incurring prohibitive costs or latency issues. Sampling of JetStream is available now, with full production expected by year-end, positioning D-Matrix as a notable player in AI infrastructure optimization.