The CEVA-X2 is an advanced DSP optimized for intensive sound processing applications. It has been specifically designed to tackle multi-microphone speech processing use-cases, high quality audio playback and post-processing, and embedded sound neural network implementations.
The CEVA-X2 is a combined DSP/controller based on a VLIW/SIMD architecture with a 10-stage pipeline, operating at 2GHz in typical conditions of a 16nm process.
Other key features include:
- 4.5 CoreMark/MHz score,
- dual 32×32 MAC operations per cycle,
- quad 16×16 MAC operations per cycle,
- 64-bit SIMD fixed-point operations,
- Up to two IEEE single-precision floating-point units operating in parallel.
The processor includes two Scalar Processing Units (SPUs), which support 8/16/32/64-bit data types for arithmetic and logic operations. It supports both static branch prediction and optional dynamic branch prediction, and offers both supervisor and user modes.
At a system level, the CEVA-X2 includes a 2-way or 4-way set-associative data cache with write-through and write-back policies, as well as hardware and software pre-fetch capabilities.
Data traffic is managed by the CEVA-X2 optional CEVA-Connect mechanism, which ensures intelligent scheduling for high Quality of Service (QoS) and low latency data transfer tasks and minimizes CEVA-X2 processor load.
CEVA-X2 supports your audio/voice/speech software needs using the extensive CEVA Sound Library, CEVA Neural Network Library, CEVA’s ClearVox Voice processing software package, and a vast ecosystem partner offerings of software solutions for any sound application.
A fully featured VLIW/SIMD architecture optimized for the most demanding DSP algorithms
- Parallel DSP with two SPUs (Scaler Processing Units)
- Flexible data plane processing with 8/16/32/64-bit native C data types support
- Highly efficient controller with a 4.5 CoreMark/MHz score
- Advanced Instruction and Data cache architecture with ACE coherency support