Overclocking question that I have found no info about.

[coder543]

I have an Intel i5-2500K and I've had it stably overclocked for months at 4.2GHz quad core, but I had an interesting idea: What if I ran the processor with two cores disabled? The CPU temperatures dropped by 20 to 30 degrees Celsius under load with that configuration, so I bumped it up to 4.4GHz dual core and the temps were still insanely low. I upped it to 4.8GHz dual core to see what would happen, but my computer refused to boot so I had to do a CMOS reset.

I don't know enough about overclocking to get very high it seems, but I would love to know just how high one of you more experienced overclockers can get one of the K-series chips using just two (or even one!) cores and the stock low profile cooler preferably, or whatever other cooler. Obviously cutting out cores makes you lose performance, but not everything is highly multithreaded, so jumping single threaded performance can have interesting results.

If there is enough community interest, I would love to see AnandTech review the K-series overclocking scalability under varying numbers of cores, with and without hyperthreading, because I googled and could find NO information related to this anywhere. At all. So, I signed up for an account here to ask about this.

 

[Ben90]

Welcome to the forums, and great first few questions. I'm not very articulate when it comes to things like this, but I'll try my best.

There are a lot of physics working together at the same time that limit how high we can overclock our processors. The main things that are different in your question versus general overclocking is reduced operating temperature/power consumption and the possibility of eliminating gimp cores.

Getting rid of gimp cores is pretty easy to explain. When running using only one core, the chance of a slow transistor holding back the overclock is reduced. It is possible however, that the transistors with the tightest tolerances are not in the core at all. Basically it can get you a few more Mhz out of the overclock or it could do nothing at all.

Temperature is a little less cut and dried. As temperatures increase, the minimum stable voltage for a specific clockspeed also increases. Unfortunately, I don't think there are any graphs out there that show us just how pronounced this effect is. I would estimate for the temperature deltas we are talking about for a possible 200Mhz increase in max overclock, if that.

Lastly, we have reduced power consumption. How much this affects your overclock really depends on your motherboard. I wouldn't imagine a $400 enthusiast board would have a problem with an increased load on the CPU rail. Once you start pushing the limits of the power delivery system such as cheaper boards or mATX, the power delivered has more noise in it which can cause instability.

If I have time, I might test my older i7 920 with fewer cores and share the results.

 

[Ferzerp]

That and what people seem to have forgotten over time is that not all processors are created equal. No overclock is ever a certainty.

 

[Billb2]

Hmmmm...

Did the CPU fail the overclock or something else failed? memory?

 

[coder543]

Yeah, I understand that overclocking is never a certainty, and what I'm proposing is hardly revolutionary, but these newer chips (such as Sandy Bridge) are able to overclock a tremendous amount just with the stock cooler and all cores enabled. I wonder how high those overclocks could go with several cores disabled, just for the sake of improving single threaded performance by having say, two really fast cores instead of four semi-fast cores. These new chips can probably achieve dual core overclocks that are crazy on the stock cooler.

Mine failed probably due to a wrong voltage or other thing, I really don't know enough about overclocking to do much with it. I was just adjusting the multiplier, not really anything else since I don't understand the other stuff. Mine failed to boot when I did 4.8GHz dual core, but the hardware could probably handle that more than fine if I knew the proper voltages and power values and the like. My RAM is rated at 1600MHz and I've tuned the BIOS to use it at 1600MHz, so I don't think it was RAM-being-too-slow issue, but I honestly don't know.