> Since most of the circuitry within each core is dedicated to computation, rather than speculative features meant to enhance single-threaded performance, most of the die area and power consumed by Fermi goes into the application’s actual algorithmic work.
But GPUs are far from easy to code for, even in 2021, and there are a lot of workloads that benefit from speculative branching. Having a combination of GPU cores and speculative CPU cores in an architecture allows you to access the best of both worlds.
We have these - they're called GPUs!
https://www.nvidia.co.uk/content/PDF/fermi_white_papers/P.Gl... provides a good overview (search the word "speculative"):
> Since most of the circuitry within each core is dedicated to computation, rather than speculative features meant to enhance single-threaded performance, most of the die area and power consumed by Fermi goes into the application’s actual algorithmic work.
But GPUs are far from easy to code for, even in 2021, and there are a lot of workloads that benefit from speculative branching. Having a combination of GPU cores and speculative CPU cores in an architecture allows you to access the best of both worlds.