Dnotitia Brings Dedicated Vector Silicon to Server Scale at AI Infra Summit 2026
AI-summarised brief · reviewed before publication
Dnotitia Inc. announced that the first ASIC samples of its Vector Data Processing Unit (VDPU) have returned from fabrication, and chip‑level characterization is underway. At the AI Infra Summit 2026, the company demonstrated a server‑scale VDPU architecture for retrieval‑augmented generation and agentic AI, following its Best of Show award at the Future of Memory and Storage event. On an FPGA platform, a four‑card VDPU configuration achieved 5.77× higher vector‑search throughput than a dual‑socket CPU server while matching recall, and cut host CPU usage by 92% and memory by 73% during index building. Dnotitia plans ASIC‑based evaluations in Q4 2026, targeting up to 10× performance over CPU servers, and is engaging infrastructure partners for integration.
💡 Why It Matters
- · The VDPU’s focus on retrieval workloads addresses a growing bottleneck in agentic AI, freeing CPU and GPU resources for core model execution.
- · By delivering significant throughput gains and resource savings at server scale, it could reshape how AI infrastructure balances compute and memory demands.
