ZeroPoint Technologies Announces ZeroStream

ZeroPoint Technologies AB, a developer of hardware-accelerated memory compression and optimization for AI, data centers and edge computing, announced ZeroStream, a hardware IP that increases effective memory bandwidth across a variety of use cases, including delivering more tokens per second from AI accelerators.“What sets an AI system apart today is how much performance and context you can pull from the memory you already have,” said Brett Cline, CEO, ZeroPoint. “ZeroStream lets chip designers do more with that same memory, increasing tokens per second, enabling more context held in memory, and fitting larger foundation models in a smaller footprint.”

ZeroStream uses hardware-accelerated compression to move more data across existing memory and streaming interfaces, improving performance for bandwidth-bound applications.

Features include:

• Lossless compression with no quantization or accuracy loss
• Up to 1.5x compression on LLM weights, 2x average improvement on activations and KV cache
• 20-35% effective bandwidth improvement, up to 50% for some workloads
• Deterministic, data-independent latency, fully pipelined to line rate

“Edge AI is fundamentally constrained by memory footprint and data bottlenecks. Our AI Booster software suite independently slashes memory usage by 3x and provides a seamless one-click export for the entire AI vision pipeline,” said Vincent Huard, co-founder and CEO, MountAIn SAS. “Tackling the memory wall from both sides changes the game. Pairing our full-pipeline orchestration with ZeroStream’s hardware-accelerated weight and activation compression creates a powerful hardware-software multiplier, enabling chip designers and OEMs to run cloud-grade vision systems on deeply constrained edge silicon.”

The launch of ZeroStream corresponds with ZeroPoint’s product catalogue update, focused on streamlining, simplifying, and enhancing four core ZeroPoint solution areas, now branded as ZeroStream, ZeroConnect, ZeroStorage, and ZeroAI.