AWS Neuron 2.31.0 now available with NKI 0.5.0 and UltraServer Operator

AWS Neuron 2.31.0 is now available with significant enhancements to the Neuron Kernel Interface and new UltraServer automation capabilities.

• NKI 0.5.0 introduces MX FP8 scale dtype support and tensor indirection for optimized indexed access patterns
• New NkiTensor view APIs enable zero-cost tensor layout transformations
• Neuron UltraServer Operator for Amazon EKS in public beta automates workload allocation and resource management
• Redesigned Neuron Compiler code generation backend improves performance on Trn2 and Trn3
• Neuron Runtime adds contiguous shared scratchpad support for simplified device configuration
• NKI Library adds 14 new experimental kernels for MoE training, deformable attention, and ring attention
• Neuron Explorer adds System Trace Viewer source code linking for improved debugging

AWS Neuron 2.31.0 enhances AI training and inference capabilities across Trn and Inf instance families available globally.