Why isn't it possible to clock 16-bit processors higher as opposed to 64-bit processors?

[MadRat]

As I understand the limits of a processor, the trace length of the longest trace of any one stage of a pipeline is the limiting factor when it comes to raw clock speed. Wouldn't a 16-bit processor be much more compact than a 64-bit processor, therefore the longest trace of any one stage should be shorter in comparison? To me that should mean that the raw clock speed should be faster on the 16-bit processor.

It also seems apparent that by adding stages to pipelines the width of the entire bus can somewhat be overcome because of the shortening of each stage in the pipeline as a whole. Introducing more stages per pipeline should offer an even bigger advantage theoretically for the 16-bit processor over the 64-bit processor, would it not?

 

[Sahakiel]

Originally posted by: MadRat
As I understand the limits of a processor, the trace length of the longest trace of any one stage of a pipeline is the limiting factor when it comes to raw clock speed. Wouldn't a 16-bit processor be much more compact than a 64-bit processor, therefore the longest trace of any one stage should be shorter in comparison? To me that should mean that the raw clock speed should be faster on the 16-bit processor.

It also seems apparent that by adding stages to pipelines the width of the entire bus can somewhat be overcome because of the shortening of each stage in the pipeline as a whole. Introducing more stages per pipeline should offer an even bigger advantage theoretically for the 16-bit processor over the 64-bit processor, would it not?

Last I checked, trace length had nothing to do with data width.

Shortening each pipeline stage compounds the problem with wire delay.

 

[Sahakiel]

I came across an article once which introduced wire-level pipelining. If I remember correctly, it dealt with massive die sizes.

 

[blahblah99]

Designers could design a 1Ghz 16-bit processor, but who would pay for one? If you need signal processing, get a DSP. If you want cheap general purpose processor, get an 8-bit or 16-bit one. There simply isn't enough demand for extremely high performance 16-bit processors.

 

[borealiss]

Precision and tracelengths can or cannot effect one another depending on the architecture and how functional units are implemented. When it comes down to extending the precision of a processor, overall it is relatively cheap. Most of the operations that occur on each bit in a piece of data happen in parallel, so most operations would probably take the same amount of time. Critical path timing usually is associated with the complexity added in pipelining a specific architecture, with data forwarding, etc... It has little correspondence to the actuall precision of the data being manipulated, unless the functional units become extremely serial in implementation, in which case they would effect the overall pipeline length and further add to the problem of critical path timing.