AI software company Nota has secured a follow up software supply contract with semiconductor startup HyperAccel. The agreement focuses on delivering large language model compression and compute kernel optimization for the Bertha Language Processing Unit. This partnership marks the 2nd consecutive year the 2 companies have collaborated to boost inference speeds in enterprise data centers.
HyperAccel designed the Bertha LPU specifically to process heavy language models in server environments. Running enterprise workloads requires close coordination between physical silicon and execution code. The collaboration pairs the HyperAccel software development kit with its Legato domain specific language. Nota applies its NetsPresso platform to adjust neural network weights to the Bertha hardware layout. Engineering teams will optimize core compute kernels and implement quantization to reduce memory overhead during live operation.
Corporate clients evaluate inference processors based on software flexibility rather than raw compute benchmarks alone. Adapting modern language models to the Bertha architecture gives HyperAccel broader deployment options in competitive data center markets.
AI chip competitiveness depends on the software ecosystem that runs models quickly and efficiently, not just superior hardware.
Nota CEO Chae Myung su confirmed that the company plans to expand similar software supply agreements with international semiconductor firms. The business strategy focuses on supplying full stack optimization software, covering everything from initial model reduction to low level computational kernels.
