The new HiFi and Vision DSPs have been designed to address increasing system-level performance and AI requirements for an expanded range of applications – combining enhanced performance across multiple algorithms with increased energy efficiency.
The expanded portfolio includes the Cadence Tensilica HiFi 1s DSP and HiFi 5s DSP for audio/voice, lightweight imaging and AI applications and the Tensilica Vision 110 DSP and Vision 130 DSP for image sensor, camera, radar and lidar applications.
“With lengthy SoC design cycles and rapidly changing AI algorithms, our SoC customers require flexibility. On-device and edge AI systems feature numerous diverse sensors, so any AI inference solution must be adaptable to the end application requirements,” explained David Glasco, vice president of research and development for Tensilica IP at Cadence.
“Designing with a system-level perspective is crucial, and these SoCs must be energy-efficient, flexible and future proof to accommodate new neural networks. Cadence continues to make significant investments in our Tensilica product line, and our latest HiFi and Vision DSPs reflect that commitment.”
The new HiFi and Vision DSPs benefit from the advantages associated with the Xtensa LX8 platform, including native AMBA AXI support for lower latency and L2 cache for increased system performance. Branch prediction enables a 5% to 20% reduction in cycles for both the HiFi and Vision DSPs.
Supported by Cadence’s NeuroWeave Software Development Kit (SDK), the DSPs can be paired with the Cadence Neo Neural Processing Units (NPUs) to offload AI inference workloads.
These latest Tensilica HiFi DSPs add more capabilities for lightweight imaging and AI to meet market demands, enabling them to serve as additional compute resources beyond just audio, either as a standalone solution or complementing the Vision DSPs and Neo NPUs.
Individually and collectively, the following features will, according to Cadence, significantly improve out-of-box performance and reduce time to market for OEMs and software developers:
- Double-precision floating-point acceleration provides up to 30X better out-of-box performance of popular codecs, enabling easier handling for computation requiring expanded dynamic range and precision
- Hardware/software co-design enhances auto-vectorization, greatly reducing the need for hand-optimized code and providing source code compatibility across HiFi DSPs
- New 8-bit operations in the HiFi 1s DSP boost imaging and AI performance by up to 2X at the kernel level, compared to the HiFi 1 DSP
- L2 cache increases system performance by up to 50%
