• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Impact of disabling cores for overclocking

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

mackerel

Member
Joined
Mar 7, 2008
There's a challenge currently running on PrimeGrid with their ESP subproject which I'm taking part in. Part of the discussion went on to first to return units, as opposed to doing the most for the challenge. That got me thinking...

Past experience showed you have a balance between CPU demand and ram supply for this type of task. At least when running multiple units at once for max throughput, but if you optimise instead for speed... you can run just one and have effectively unlimited ram. But it doesn't stop there. The units are 960k FFT size, which means it only needs 7.5MB of working data. It would fit inside the L3 cache of an i7 and ram speed becomes irrelevant.

So to return the fastest time, you would simply have to get the CPU core and L3 cache speed up as fast as possible. As you're only running one unit, there's no need to have all the cores. The thought is, is there any overclocking benefit to disabling them? Maybe to two, leaning another spare for background tasks. Would this effectively halve the power consumption? Is there a benefit over having all 4 cores available but leaving the others unused?
 
Yes, disabling cores (either physical or hyper-threaded) does tend to reduce power consumption, and thereby also reduces heat production under heavy or full system loads.

Yes, disabling cores can increase overclocking ability of the CPU as it would likely need less voltage and wattage for the same CPU frequency, since there are less cores/threads to keep stable and less voltage and wattage needed to keep them stable if your using say only 4 threads on an 8-threaded processor. But, the other side of the coin is you don't want to disable too many cores/threads, as if you only have one core/thread active the system can seem slow and unresponsive even to complete the simplest tasks. Even two cores may seem slow compared to four, but it would definitely be better than only one.

Disabling half of the cores/threads though doesn't actually cut the power consumption in half, it's actually less than that, as if I am remembering correctly some websites tested this at some point and found that though it did drop power consumption it didn't cut it in half even when half of the cores on a CPU were disabled. I think it was more along the lines of 1/4th to 1/3rd less power consumption.

I would say testing would be required to find the right balance for your type of workload. Maybe start off by disabling hyper-threading (assuming this is a hyper-threaded CPU) and see if that increases your CPU's overclocking abilities and/or results in faster completion times for the task. Then drop down to two cores with no hyper-threading and overclock them, and maybe try with only two cores enabled but with hyper-threading active so it would be a 2C/4T processor and see how that goes.

I'm afraid I have no experience in the specific task you speak of, but hopefully some of what I said might be helpful to you. And maybe some others might chime in here if they have thoughts on this.
 
HT is already off as past testing has shown it has no benefit at all, and actually can have negative impact unless you counteract it by manually setting affinity.

Think it will be a case of "try it"... it's a 5 day challenge and we're little over half a day in so wont be for a while. This is a pain actually. I have two 6700k's, one in my main rig which I don't want to toast, and another in my VR rig which has cooling issues as I went mini-ITX. Unless I swap the 6600k (in desktop board) and 6700k from VR rig...
 
I gained 100mhz @Same voltage from HT on to HT off if it helps any. Never tried disabling cores on Skylake because even on my old AMD Windows/games seemed to like 4 active cores better then 1/2/3, more responsive. 5/6/7/8 seemed to make zero difference for day to day.
 
In case the point wasn't clear, the goal of this is to run a specific task as fast as possible. It is not threaded so single core performance is all I'm after.
 
I was clear simply that disabling cores managed to get OC a little higher but lost Windows stability/fluidity which might impact on the program's performance. You might get away with just 2 cores active, one for the process and another for Windows (depending on which one you're using).
 
Yes disabling cores and ht will help. I often cut my cpu down to 2 cores when benching a single threaded app like superpi. I assign the program to run on the stronger core and everything else to run on weaker core. You should be able to eek out an extra 1-200mhz by doing this.
 
In case the point wasn't clear, the goal of this is to run a specific task as fast as possible. It is not threaded so single core performance is all I'm after.

Just disable HT and go from there (you already did). If you want, Lord knows you do..., you can find the 'best' core of the bunch and run your apps there. But the reality is IPC will clearly be the same and you won't gain much (100-200 mhz?) in the first place doing so anyway. It's only time wasted... :)
 
I usually use 2 cores and no HT when benching on single core apps (superPi, Aquamark...).

Out of curiosity, I ran RealBench yesterday after reading this thread, with and without HT.

The CPU needs 0.02v less vcore to be [email protected] with HT off, and runs +/- 10c cooler than with HT on.

But it is 30% slower on multi-threaded optimized apps.

Still, to answer your question, as many above already say, if you are looking for pure perf on one core only, disable HT. No need of disabling cores, as I don't think it will help stability significantly (not looking for stability when benching, but it seems that you need it for your app).
 
The app is equivalent to running Prime95 big FFT for however long it takes. My system as it is takes about 9 hours running 4 cores. I'd estimate it might be closer to 7 hours for 2 cores, and I don't have data to suggest how much faster it might be running out of the L3 cache. Combined with a higher overclock, I hope sub 6 hours is possible.

Most likely I'll start exploring OCs from Wednesday evening after the existing challenge ends. Depending on how stability goes, I hope to find the limit by end of next weekend.
 
I think I had a play in the past, and it didn't seem to make any difference. I will try again.
 
Back