PCB frequency limit.

Thread Starter

paulktreg

Joined Jun 2, 2008
833
Is it just me or have the leaps and bounds, relatively speaking, in processor speeds slowed down recently in favor of higher core counts?

I'm specifically talking about the desktop PC. Have the manufacturers started to reach the limits in terms of frequency that an ATX motherboard can reliably operate?

Printed circuit board design in the communications and microwave areas for example operate at very high frequencies but the printed circuit boards can be designed that way. The ATX motherboard as to meet certain design criteria and space after all is limited but can the clocks go significantly higher based on the ATX format or is a change needed? A frequency of 5GHz can be realized with some processors with the right motherboard but how much higher can thing go?

Can anyone point me in the direction of any decent articles on this subject?
 

kubeek

Joined Sep 20, 2005
5,794
I think you´re missing the point here. The processor speeds don´t go up because the silicon processes limit the speed, it really has nothing to do with the mother board (in the sense of the PCB material etc.).
For example, the core of my CPU runs at 2.8GHz, but the frontside bus is only at 1333MHz. This has nothing to do with the PCB itself, the problem is in the components that the CPU is connected to. The main bottleneck is the memory, and because the absolute bulk of data transactions is between memory and cpu, there is no point making the fsb faster when the memory can´t use it.

Another thing, as far as I know the ATX standard only defines the physical size of the board, mounting holes, I/O card positions and power supply. In no way it says that memory needs to be this far from the cpu, or that the board needs to be made of FR4 or any other particular pcb material. Thus the ATX standard has no impact on maximum freqency what so ever.
 

Thread Starter

paulktreg

Joined Jun 2, 2008
833
I'm not talking about "Silicon Processes" I'm talking about stray capacitance/inductance that must exist between printed circuit board tracks.
 

kubeek

Joined Sep 20, 2005
5,794
Yes there is stray capacitance and inductance on any pcb, but this has little to do with ATX standard.
Modern data paths on PCBs use controlled impedance traces like microstrips and striplines, which basically are transmission lines similar to a coax cable, so there is not much trouble with signal integrity. The main limiting factor here is the length of the trace, because the singal has finite propagation delay of roughly 1ft per ns, so to get good performance you need to keep the traces short (and of matched length in parallel buses).
You see that to get things faster, you generally need them to be spread closer together, so I don´t think the huge size of an ATX board poses a limit to increasing the speed.
 

WBahn

Joined Mar 31, 2012
29,978
The PCB issues place limits on how fast things that communicated via signals the PCB can run. But that is not what limits the speed of the processor -- they are already beyond that (the PCB design and manufacturing effort that would be needed to directly clock the CPU at full rate would be too expensive). So a long long time ago they started taking a pretty tame oscillator and using a PLL (or a DLL) on the processor to derive a much high CPU clock that never leaves the CPU package. About the time they got to the ~3GHz range they started running into issues that would start costing lots of design and manufacturing effort to go faster, so they hung a right and started focusing on getting more computation out of each clock cycle (and generally at a slower clock rate of 1.somthing to 2.4ish GHz). So more cores, SIMD, MIMD, more and more "magic" blocks to perform speculative execution, out-of-order execution, adaptive branch prediction, and so and and so on. Those are all issues that play out within the CPU package and have nothing to do with PCB speed limits (which exist, but which impact a different set of blocks).
 
Top