DSPs, media engines & fixed-function blocks
Dedicated hardware handles repetitive signals and media with less general-purpose overhead.
Your camera preview may pass through several engines without using the CPU for each pixel.
Programmable signal work.
Find the bottleneck
Low arithmetic intensity hits the memory ceiling. More data reuse can move the workload toward the compute ceiling.
What this model includes
An ideal upper bound with fixed peak compute and bandwidth. Ignores latency, overhead, cache-level traffic, and instruction mix.
What happens inside
Process signal streams
DSP-oriented designs favor repeated multiply-accumulate work, vector operations, and predictable streaming. Filters, transforms, audio, and communications are common workloads. A DSP is not necessarily fixed-function; many are programmable, with instruction and memory features chosen for their workload.
Recognize specialized pipelines
An ISP transforms sensor data, a video engine encodes or decodes supported formats, and a display engine scans out buffers. Crypto and compression blocks offload specific operations. Supported formats, profiles, bit depths, and throughput vary by generation; software fallback has different power and latency.
What this means for your code
Low-level engineer
Check buffer formats, synchronization, DMA, and real-time deadlines. A hardware block may require exact alignment and frame layout.
Software developer
Use platform media APIs rather than manually processing every frame when supported acceleration fits. Check whether your exact codec profile is accelerated.
Read the actual specifications
These references supply the underlying contracts and implementation details. The diagrams here are simplified teaching models.