Ditching HBM: d-Matrix Unveils Raptor, the 3D-DRAM Powerhouse for GenAI Inference
d-Matrix is reimagining AI hardware by stacking DRAM directly on logic to bypass the limitations of High Bandwidth Memory.
By Blip Tech
1 min read
At Hot Chips 2026, d-Matrix revealed its latest innovation, the Raptor 3D-DRAM accelerator, designed specifically for generative AI inference. By moving away from traditional High Bandwidth Memory (HBM) and instead stacking DRAM directly onto the logic die, the Raptor architecture aims to significantly improve memory bandwidth and power efficiency. This 3D packaging approach addresses the memory wall bottlenecks that often hinder large language model performance, offering a high-performance alternative for data centers focusing on generative workloads.
