What limits overclocking?

Associate
Joined
28 Jun 2005
Posts
895
I was once told that the thing that limited overclocking was the fact that when the logic gates got hot they didnt work as fast, so this gave you a max overclock

Though I was under the impression that if you gve it enough volts you can just keep cranking it higher and higher, and when you gave it too many volts you fried it. Obviosuly increasing the volts means more heat and if you can't control this you can't keep whacking up the volts as your pass the maximum operating temperature of the core (which I assume permanently damages the chip)

Perhaps someone could offer a more correct view?

Thanks
 

daz

daz

Soldato
Joined
18 Oct 2002
Posts
24,073
Location
Bucks
Also since we went down to 0.13 processes with the P4 Northwood core a few years ago, people have witnessed a new phenonemon known as electromigration where giving a CPU too much voltage causes the electrons to tunnel out of the internal CPU pathways, causing serious permanent damage over time.

What limits overclocking is just what limits the CPU manufacturers from simply upping the default speeds to 5ghz+ though... and you seem to have described them pretty well in your first post. :)
 
Associate
OP
Joined
28 Jun 2005
Posts
895
so does the gate delay increasing actually provide a real upper bound or is it theoretical? (i.e. the actual problem is heat dissipation and the CPU would burn out before it mattered)
 

daz

daz

Soldato
Joined
18 Oct 2002
Posts
24,073
Location
Bucks
Well there's a physical limit to the switching frequency of the transistors as well as the circuits used to tell them what to do.

Heat output is directly proportional to the increase in speed. Also remember, that heat is proportional to V^2, so a small increase in V equals a large increase in power.
 

daz

daz

Soldato
Joined
18 Oct 2002
Posts
24,073
Location
Bucks
joeyjojo said:
Isn't it more likely to be proportional to (v^2)/R or similar? Doesnt the R of a semiconductor decrease with greater temps?

Correct on the first point, I'm not sure on the second. It's obviously a very complicated business, and I'm not an electronic engineer. :o
 
Soldato
Joined
19 Jun 2009
Posts
3,852
The limitation is the electrons, as when the speed increases they will eventually jump circuits inside the CPU, causing a crash.
 
Last edited:
Associate
Joined
3 May 2018
Posts
604
Grey beard here.

Initially PCs had fixed clocks or fixed multipliers of the Front Side bus (memory bus).

Manufacturers couldn't predict what cooler you were running, what power supply or motherboard or what work loads, so they clocked the chip to be bullet proof and avoid RMIs.

With the right process and knowledge you could significantly increase those clocks. Things were simple. Single clock (FSB) and for the most part a minimum of a single instruction per cycle.

Things started to get complex when DX chips started running internal CPU instructions on the rising and falling edge of the 33Mhz clock of the day.

By the time we got to Pentium/Pentium2 things had started to change, with processor pipelining coming of age they could sometimes do 4 different instructions simultaneously.

While there were other bus clocks added so that memory could run at different rates to the main CPU, it was still fairly easy adjust clocks and voltage until it didn't crash.

Intel introduced dynamic clocking around the Celeron days. Which is a rather tame way would drop and raise clocks between idle and intense workloads. Still they could be overclocked.

Today however advancing up to current day the likes of the 5800X is already overclocking itself hundreds of times a second, running those same processes you used to do taking hours of balancing frequencies to voltage to get stability. It's already doing that at light speed. It's all about boost clocks. The chips, CPU and GPUs are self-overclocking using the concept of a boost clock. The chip senses within itself what frequency to set cores to above their base clock based on things like power, temperature, voltage and load type.

Mostly the overclocking on these is rather sending limitations and "hints" to these automatic routines to make them more or less aggreesive.

The real fiddling has moved to undervolting. To get more 'clocks' for your Watt and get more performance under the thermal and/or power limiters.

The only advice I'd give if you are just starting out with overclocking usig 2020+ era chips. I'd say, Dont. Start with upgrading your cooling. When you upgrade the cooling the chip WILL clock itself up without you doing anything. Only when you have spent all you want to spend on cooling should you consider anything but a minimal tune (like enabling Precision boost) which you should kinda do anyway, It's just a toggle switch... and a memory profile that matches what you bought!
 
Last edited:
Associate
Joined
3 May 2018
Posts
604
What is definitely old-school and no longer effective, certainly in Ryzen is all core fixed clock overclocks. I mean I can probably get my 5800X stable with a fixed clock of 4.3Ghz. But with the standard dynamic boost clocking it will hit 4.6Ghz or higher, just not on all cores or for long. If I pin it 100% on all cores with prime95 it might drop clocks to 4.2 or 4.1Ghz to save thermals and power. Still over all under normal daily load, the dynamic clock is more performant.

I have seen people using BIOS tricks to switch between the two. So under a certain total die power, it is allowed to boost as it wishes. Above a certain die power, like 75% it switches to fixed clocks.
 
Back
Top Bottom