[SOLVED] Expected performance gain from overclocking CPU by 10%

Mawla

Reputable
May 21, 2021
59
0
4,530
The last time I overclocked my computer was over 20 years ago. It was a Pentium 200 system put together from old parts and overclocked to 266 MHz, the maximum that the motherboard jumpers could do. So please consider my question as coming from a total newbie.

With a modern low-midrange setup, what kind of performance gain could be expected by overclocking the CPU by 10% in terms of game fps, CPU rendering of 3D images, file compression/decompression, etc. etc. At least I'm aware that this is a simplistic question and I expect the results to vary with the type of work and the software used. I just want to have a broad general idea. Thanks in advance for any helpful input.
 
Solution
....
Anyway, suppose I have a Ryzen 5 3600X, 8GB RAM, GTX 1650 and a mid-range NVMe SSD.
....

If you want to 'overclock' a 3600X by 10% you have to define the starting point first and methods.

You might could get 10% fixed overclock on the base clock: so 4.2G instead of 3.8G. But that will result in a clear loss of performance for gaming since you could be getting 4.4Ghz single core boosts if you use PBO instead. And extremely heavy workloads (3d rendering, for instance) might be unstable at any higher of an overclock since it will need high fixed voltage that degrades the CPU early even with extremely good cooling.

But using PBO you should be able to get close to the same 4.2Ghz in heavy loads while letting the processor...
The issue with your question is that it's vaguely stylized with generic information. You're advised to include parts/specs to the system you intend to perform a 10% overclock on. Also, you should include the names of the app's titles without stating an etc to help us understand if you're wasting resources with an overclock(which can easily be done via Turbo Boost or PBO) or a simple ram overclock(with regards to AMD's Ryzen platform.
 
I couldn't be specific because I don't intend to overclock my (or any other) computer in the near future, and I have no specific application in mind. And I'm not really looking for help regarding the best way to overclock or whether it's worth it.

Anyway, suppose I have a Ryzen 5 3600X, 8GB RAM, GTX 1650 and a mid-range NVMe SSD. Further suppose that I play older FPS games, render architectural 3D images with the CPU (not GPU) using Lumion and V-Ray, and compress/decompress with WinRAR. (This does NOT describe my system or the type of work I do). Does that make it easier to answer my question?
 
Tough question, but it begs others.

If you made a pc specific for CPU rendering, you would (hopefully) not put yourself in a position where you would need to overclock instead of simply starting out with more cores.

BUT, to eek out a few minutes saving on a half hour render, sure it would be possible. Whether you would need 10/15/20% OC to shave off 5% time saving, idk. OC is not linear
 
....
Anyway, suppose I have a Ryzen 5 3600X, 8GB RAM, GTX 1650 and a mid-range NVMe SSD.
....

If you want to 'overclock' a 3600X by 10% you have to define the starting point first and methods.

You might could get 10% fixed overclock on the base clock: so 4.2G instead of 3.8G. But that will result in a clear loss of performance for gaming since you could be getting 4.4Ghz single core boosts if you use PBO instead. And extremely heavy workloads (3d rendering, for instance) might be unstable at any higher of an overclock since it will need high fixed voltage that degrades the CPU early even with extremely good cooling.

But using PBO you should be able to get close to the same 4.2Ghz in heavy loads while letting the processor manage it's voltage to keep itself safe with the same really good cooling. At the same time, it will be free to boost to 4.4Ghz or higher for single thread or light multi thread workloads like gaming. But it's hard to say what 'percent' overclock that is as it's so variable.

Modern CPU's that use dynamic boost clocks and power management don't work the same as older ones. You can't use the same approach to overclocks. In fact, just upgrading the stock cooler can be as effective as overclocking since Ryzen is particularly effective at pushing the limits of it's silicon when given sufficient temperature margin.
 
Last edited:
Solution
If you are bulding from scratch get a 12th gen Intel, a 12400f is going to destroy a 3600x for half the price.

Everyone is correct. CPUs are heavily binned and use all the cooling they can get. Meaning they overclock themselves for the most part with a good cooler and enough power.

Making sure the power limits on the motherboard are set as high as the cooling allows will get a lot out of them: example: I set my 11400 to 95w continuous, I was seeing temps of mid 50C, with a cheap 120mm Water cooler and my performance was near a stock 10600K.

A budget Z690 and 12600K is really good at gaming and crunching numbers, if you shop around, mine were ~ $400 for both. The extra 4 cores help with rendering etc, but may hurt with games.
 
12400F is the budget king of gaming, currently; that being said, I'd shoot for the 12600K, if possible, as I suspect those few extra E-cores will give it a bit more 'gaming longevity' as time marches on....

(The 'golden days' of overclocking seem done for; never to return are days of 4-5 clicks in the BIOS and running the Celeron 300A at 464 MHz on a 440BX, the P3/600A at 900 MHz, etc...)
 
12400F is the budget king of gaming, currently; that being said ...

Not really so much so with recent price drops of Ryzen CPU's. Also shop carefully as many locales can have different pricing pricing and sales at any time.

Also, just looking at CPU price in isolation is a bad way to assess price-performance: you have to include motherboards with an apples-apples comparison of capabilities and features to make a fair comparison of price in that equation. Especially important to consider is that if you don't have to buy a motherboard (many people have perfectly good AM4 boards right now) then changing platform makes very little sense for any sort of price/performance comparison.
 
Last edited:
I built a new system about a year ago using an i7-11900kf and an Asus Maximus XIII Apex MB, and just using the built-in AI overclock feature boosted my speed by 56%, so 10% is probably well within reach. The best wat to determine your benefit is to just do it and run everything you normally run.
 
It seems like overclocking processors becomes less of a thing each year we go onwards, im currently running 51% overclock on first generation intel processor, which makes it fine for modern gaming on 1080p. Margins have gone smaller from this point. I wouldnt expect more than 10-15% overclock from new ryzen system and even with intel something like 20% is best you can do.
 
I built a new system about a year ago using an i7-11900kf and an Asus Maximus XIII Apex MB, and just using the built-in AI overclock feature boosted my speed by 56%, so 10% is probably well within reach. The best wat to determine your benefit is to just do it and run everything you normally run.

An 11900K/KF would hit 5 GHz all-core turbo under load...

Exactly where/how did someone determine /measure/claim a 56% 'boost in speed'? (Hopefully, this is not someone thinking that boosting to 5.2 GHz from 3.5 GHz 'base clock' constitutes any sort of miraculous 'AI overclock'? As it is normal behavior for that CPU anyway)
:)
 
  • Like
Reactions: lga1156_ftw
An 11900K/KF would hit 5 GHz all-core turbo under load...

Exactly where/how did someone determine /measure/claim a 56% 'boost in speed'? (Hopefully, this is not someone thinking that boosting to 5.2 GHz from 3.5 GHz 'base clock' constitutes any sort of miraculous 'AI overclock'? As it is normal behavior for that CPU anyway)
:)

It actually shows up on the Asus flash screen when booting up. Of course they want it to look its best, after all, this is after applying their AI Overclock feature. All cores will run 5.0Ghz+, but I currently have it setup for 2 cores to run at 5.4Ghz and the rest around 4.8Ghz.
I remember overclocking my CPU about 8 years ago, which took a lot of research and tweaking before getting it to post. Why don't they just make the CPUs to maximize their performance capability out of the box instead of toning it down and asking for everyone OC their CPU to capture this "increase" in performance?
 
Simple answer, nothing.
You could squeeze out an extra 5-10fps from a 12600k, which basically does nothing, the 12600k being strong enough to get beyond the refresh of most monitors, and even in the highest demanding, lowest fps titles you couldn't see the difference physically anyway. Below the 12600k there's isnt any OC worth mentioning and above the 12600k the cpus are strong enough stock that OC is pointless.

Which leaves Ryzen which benefits more from undervolting than it does overvolting with OC.

Point is, unless you really just have to push the envelope, the need to OC to get tangible differences has about joined the Dodo and SLI in the graveyard.
 
7 years ago I took a 3770k from 3.9GHz turbo and OC'd the snot out of it, hitting 5.0GHz at 1.4v, all core.

Nowadays, cpus are already doing 5.0GHz or better on at least 4 cores/8 threads right out of the box. About the only OC option is maybe adding 100MHz - 200MHz If you are lucky, and changing nothing else but going from turbo speed limits to all core limits. Intel even has that option, it's called MCE.

But you will not get a 1.1GHz OC out of anything, not even close, not without LN2 cooling, and thats useless for a daily driver.

OC, Dodo, Sli, same boat....

If you look at the fps ratings, 12400-12900k gets almost the same, within spitting distance. The difference being the L3 Cache amount playing a much more intensive role in the bigger cpus, giving them that edge. To a point. Which makes the 12700k best value big Intel. It has enough L3 to compete with the i9's in anything except benchmarks. In a side by side game comparison irl, there's not a person on the planet who could tell the difference. Be hard pressed to see any difference in a 12600k or 12900ks.

OC can give a couple of fps, but honestly it's so close and not enough fps to be really viable as an option. It's pure luxury now, very, very far from being a somewhat advantage.
 
Last edited:
  • Like
Reactions: lga1156_ftw
7 years ago I took a 3770k from 3.9GHz turbo and OC'd the snot out of it, hitting 5.0GHz at 1.4v, all core.

Nowadays, cpus are already doing 5.0GHz or better on at least 4 cores/8 threads right out of the box. About the only OC option is maybe adding 100MHz - 200MHz If you are lucky, and changing nothing else but going from turbo speed limits to all core limits. Intel even has that option, it's called MCE.

But you will not get a 1.1GHz OC out of anything, not even close, not without LN2 cooling, and thats useless for a daily driver.

OC, Dodo, Sli, same boat....

If you look at the fps ratings, 12400-12900k gets almost the same, within spitting distance. The difference being the L3 Cache amount playing a much more intensive role in the bigger cpus, giving them that edge. To a point. Which makes the 12700k best value big Intel. It has enough L3 to compete with the i9's in anything except benchmarks. In a side by side game comparison irl, there's not a person on the planet who could tell the difference. Be hard pressed to see any difference in a 12600k or 12900ks.

OC can give a couple of fps, but honestly it's so close and not enough fps to be really viable as an option. It's pure luxury now, very, very far from being a somewhat advantage.


This is why i like to use ancient tech and overclock it just for the fun of it. That is really awesome result for 3770k didnt know it can go that high, this means intel did not have much competition before amd 3000 series, so little IPC gains past 14 years.

First generation intel is one of the best when it comes to overclocking, expecially westmere and lynnfield. These times intel turbo boost was best to be turned off even for a small overclock since it was almost useless. Some results for example:

i7-870 4c8t 2.93 => 4.1ghz 39% increase
i7-875k 4c8t 2.93 => 4.2ghz 43% increase (i7 have slightly weaker memory controller, running 2133mhz cl10 ddr3 on these)
x3440 4c8t 2.53 => 4.0ghz 57% increase
x3470 4c8t 2.93 => 4.3ghz 46% increase (both of these can run ddr3 at cl10 2400-2600mhz, beating modern cheap ddr4)
x5650 6c12t 2.66ghz => 4.4 ghz 67% increase
x5675 6c12t 3.06ghz => 4.55ghz 48% increase (triple channel setup, harder to run ram over 1866 - 2133mhz, sweet spot usually around 2000mhz)

No degration or problems yet with any of these, i been more conservative with voltages on the 32nm westmere (x5650 / x5675) under 1.4v on the vcore and 1.325v on the imc. With lynnfield (i7 + x3440 x3470) 45nanometer you can bump up the vcore even to 1.45v without any issues if cooling is fine, these can also handle more imc voltage but still stay below 1.35v with that. Only time i had degration was with i5-760 which i was running at 1.45v IMC/vtt for two years, that can degrade the memory controller.
 
My 3770k was stable at 5.0GHz, there were a few batches that could hit 5.1GHz at 1.408v but the heat output at that level was intense, it was taking a 280mm kraken x61 up to 90°C in p95 small fft, so I dropped it for daily use to 4.9GHz at 1.32v and got 72°C instead. The 280mm kraken is roughly 300w+ capable, so hitting 90° (22°C ambient) is pushing 250-270w OC, on a 77w TDP cpu.
 
I remember overclocking my CPU about 8 years ago, which took a lot of research and tweaking before getting it to post. Why don't they just make the CPUs to maximize their performance capability out of the box instead of toning it down and asking for everyone OC their CPU to capture this "increase" in performance?
If we take say an i7-3770K or an i7-4770K, the turbo boost provided was 400 MHz above base clock speeds. Compared to say an i7-12700K, where you get 1,400 MHz from its base clock speeds. Even on AMD's side you're still looking at nearly a 1,000 MHz turb boost difference, though you could certainly push it above that.

And even then, Intel used to be conservative about the turbo boost limit. They used to cap it to the maximum speed guaranteed by any core. AMD came along and capped it to the best core and told Microsoft to stop floating threads around on their processors. Now Intel's likely doing the same thing.

So yeah, they do all the hard work for you now.
 
I couldn't be specific because I don't intend to overclock my (or any other) computer in the near future, and I have no specific application in mind. And I'm not really looking for help regarding the best way to overclock or whether it's worth it.

Anyway, suppose I have a Ryzen 5 3600X, 8GB RAM, GTX 1650 and a mid-range NVMe SSD. Further suppose that I play older FPS games, render architectural 3D images with the CPU (not GPU) using Lumion and V-Ray, and compress/decompress with WinRAR. (This does NOT describe my system or the type of work I do). Does that make it easier to answer my question?

3600X has a boost clock of 4.4 Ghz. 10% of that is 440 Mhz. So that is clocks at 4840 Mhz. Even that is questionnable if you can do it. If the chip can do it. And then, if you can cool it. https://hwbot.org/benchmark/cpu_fre...Id=processor_5872&cores=6#start=0#interval=20 After 4.75 Ghz, there is nothing but LN2. Sure, at 5 Ghz+ but the fact there isn't any in the 4.8-4.9 Ghz range on water or air...We have a problem here.

Generally, 10% OC, I would expect 2-5% more perf. Depends on workload, if it is even sped up by your overclock. Let's say I have a memory bandwidth-intensive workload. I can OC my RAM. But overclocking my GPU will do absolutely nothing.
Performance doesn't scale linearly, powerdraw scales pretty much exponentially. An Alderlake 12900k draws around 300 watts stock. If you OC it, I bet you can make it draw around 1000 watts. With LN2 of course.