StoryHardware
IBM announces NorthPole brain-inspired inference chip with 25x better energy efficiency than GPUs
IBM announced NorthPole, a 22-billion-transistor neuromorphic research processor designed for AI inference. The chip integrates all memory on-die, eliminating the energy penalty of data movement, achieving 25x better efficiency than 12-nm GPUs and 14-nm CPUs on ResNet-50 benchmarks.
- NorthPole contains 256 cores with 13 TB/s on-chip memory bandwidth, enabling rapid, low-power inference.
- Paper published in Science magazine demonstrates the approach with full benchmarks and comparisons.
- Chip is optimized for inference only; not designed for training neural networks.
- The design integrates memory and compute on a single die, eliminating the energy penalty of data movement between processor and memory.