A CPU is optimized to finish individual tasks quickly. Its few fast cores use large caches and sophisticated control logic to support sequential execution. section 1.3.2
In GPU programming, the CPU is the host: it organizes and launches Triton kernels, which execute on the GPU. This contrasts with the GPU’s emphasis on throughput across many simple arithmetic units. section 1.5
Figure 1.5 GPUs dedicate more of their area to compute and less to control and caching.
Compare how CPU and GPU chip area is allocated to computation, control, and caching. section 1.3.2








