d-Matrix

d-Matrix is a semiconductor company that develops hardware and software for AI inference based on a memory-centric architecture, in which compute is placed within memory to reduce data movement. Its Corsair accelerator is a PCIe card that combines high-bandwidth memory and an inference-optimised compute engine with chip-to-chip connectivity, and it can scale from single servers to rack-level systems. Corsair can run inference on its own or alongside GPUs. The Aviator software development kit supports model deployment from high-level, PyTorch-like interfaces down to custom kernels.

• Corsair PCIe accelerator for AI inference with integrated high-bandwidth memory
• Memory-centric (compute-in-memory) architecture to reduce data movement
• Scales from single cards to rack-level systems and compute clusters
• Can be deployed standalone or alongside GPUs
• Aviator SDK with PyTorch-like interfaces and custom kernel support