Qualcomm demos first-gen HBC silicon: 18x LPDDR5X bandwidth, 5x HBM4 per watt
ryanshrout · x · 2026-11-02
Qualcomm showed first-generation High Bandwidth Compute silicon working in its San Diego lab, running the full memory read-write path and near-memory compute at production-representative sizes. Claimed results: 18x the effective bandwidth of LPDDR5X and more than 5x the bandwidth-per-watt of HBM4. These are Qualcomm estimates from ATE testing normalized against published competitor specs. The first accelerator is likely the Dragonfly AI250, expected to sample in H1 2027. Context: memory bandwidth has been the limiting factor for LLM inference, and anyone building inference capacity pays for it in power and cost per token.
More from Infra
- Pinokio Remote lets you launch and control local AI apps from any device — cocktail_peanut · 2026-11-03
- Northwood raises $375M Series C to scale Earth-space data connectivity — oyhsu · 2026-11-03
- Creator to test GB300-class machine with 1TB RAM tomorrow, asks for fine-tuning test ideas — zxtech · 2026-11-03
- Lumentum deal sparks debate: does the sale signal doubts about the AI cycle lasting? — BenBajarin · 2026-11-03
- Rohan Anil's CoreAuto trains MoE across two global clusters over a 25 Gbps link — _arohan_ · 2026-11-03
- AI compute as a planet: three quarters under the US flag, China one seventh — LeviTurk · 2026-11-03