Thinking about the odd idea of changing a CPU and its socket.
Thinking about the odd idea of changing a CPU and its socket.
The main issue here is the delay in data transmission when connecting the CPU via cables. Unlike GPUs and hard drives, which use extension cables for flexibility, the CPU is tightly integrated with its socket. Adding a cable would introduce extra latency, making it impractical. The idea of improving cooling by routing data through additional wires isn’t really useful, as the bottleneck remains the physical connection speed. It’s a frustrating situation given the current constraints.
Latency is an issue for all components, but it's not as critical for components that are slow(er) to begin with. Adding 50 nanoseconds to an operation that takes 10 milliseconds doesn't matter much, but adding 50 nanoseconds to an operation that takes 10 nanoseconds is quite relevant (Numbers are just arbitrary examples, but I hope you get the idea). The GPU on its own also keeps paths as short as possible (e.g. memory is clustered around the core). The path between CPU and GPU is as short as it physically can be (usually port closest to CPU). However, latency between CPU and GPU shouldn't be too much of an issue. Most of the time data only goes one way, so it's not like the CPU is forced to wait for the GPU to return data, unlike e.g. an SSD or RAM. I'd rather work than be laid of or get my pay docked. Thankfully I can work from home.
In fact, heat removal onto the motherboard is integrated into chip design, especially for BGA where large power and ground areas can serve as heat sinks. It's also worth noting that 22 AWG wire would be insufficient to handle the current needed by modern CPUs. A 100 W system at 1 V demands 100 A, which would need at least ten conductors. Any wire length will introduce considerable resistance, raising temperatures and necessitating robust LLC solutions.