d-Matrix and NVIDIA NVLink Fusion for AI infrastructure

The AI inference chipmaker d-Matrix has announced it will leverage NVIDIA's NVLink Fusion to integrate its new Raptor XPUs with NVIDIA's AI infrastructure. This collaboration positions d-Matrix within a growing network of ecosystem partners, with NVLink Fusion providing them an accelerated and lower-risk path for transitioning from custom silicon to large-scale deployment.

According to Sid Sheth, co-founder and CEO of d-Matrix, the demand for inference has significantly increased, while available resources such as capital and energy remain constrained. By using NVLink Fusion and the MGX platform, d-Matrix can integrate its Raptor XPUs into a broader, liquid-cooled architecture, offering customers a faster and lower-risk route to ultra-low latency inference.

NVIDIA's AI platform is both vertically integrated and horizontally open. NVLink Fusion extends this openness to XPUs and CPUs, enabling silicon companies to focus on process innovations while leveraging NVIDIA's infrastructure for AI factory-scale deployment. The NVLink Fusion technology provides high bandwidth and low latency, making it easier for chipmakers to scale their custom architectures without the need for proprietary racks and infrastructure.

The partnership with NVIDIA allows d-Matrix to take advantage of the established rack and cooling structures of the NVIDIA MGX ecosystem. This ensures that data centers can use one type of rack for various processors, including GPUs, CPUs, and XPUs, leading to cost savings and efficiency improvements. Through NVLink Fusion, d-Matrix can develop its XPUs integrated into a common high-bandwidth, low-latency network with NVIDIA systems, simplifying the deployment of specialized inference within flexible AI factories.

Read the full article from NVIDIA.