Flexibility for Next-Gen Worloads
Enhacnements to aiWare HW and SDK upgrades mean higher performance not only for CNNs but a wider range of workloads including LSTMs and emerging workloads such as Transformer Networks to deliver a future-proof for the most demanding workloads.
Supplementing the INT8 support, aiWare4+ adds the option for FP8 format to take advantage of industry standardization and ecosystem support for FP8 for workloads that can take advantage of this format for better performance and accuracy.
Fine-grained Structured Sparsity
aiWare4+ HW and SW upgrades enable users to take advantage of Fine-grained Structured Sparsity to improve performance and power, building on the growing eco-system support for tools and frameworks.
Improved Multitasking & Scalability
With features such as Interleaved Multitasking, aiWare ensures that multiple workloads can be scheduled concurrently if possible to deliver higher resource utilization to best utilize the available TOPS to enable centralized compute paradigm.
aiWare4+ extends aiMotive’s leadership with up to 98% efficiency over a wider range of NN topologies. This efficiency scales especially for large input sizes such as early fusion of multiple different high data rate sensors, or from one or multiple high resolution camera sensors.