Overclocking

N

ng4ever

2[H]4U
Joined
Feb 18, 2016
Messages
3,583
Is it worth it nowadays ?

For lets say a lot more performance ? Like it has been in the past.
 
Depends on what CPU/platform you're starting with, what specific performance area you're intending to increase, and how much power you're willing to use and able to cool.

Not much to gain with desktop CPUs, especially for gaming increases. Everything is already binned close to the limit. HEDT has a ton of headroom if you've got the wattage.
 
The olde days of "CPU go 30-50% faster with OC" are wayyyy behind us, boost works way too well.

Nowadays it's more about "tuning" than specifically OC. One can still get a CPU running better than stock, but it's more about tweaking power consumption and maybe choosing a balance between singlethread/multithread performance. Tuning boost is generally better than going for a static OC.
 
Not really worth it. You can get a little more out modern CPUs but not worth the substantial power/heat increase for a few more percents.
 
It's basically dead in terms of bios tweaking, volt modding etc.

CPU companies are now effectively doing what 3rd party sites like Silicon Lottery were doing. It's no longer the case that you can buy a low end CPU and 'luck out' that it could/should have been a higher binned part but got downclocked to meet demand, especially for AMD. Yields are just too low. I can't even remember what the last 'super clocker' CPU was tbh. 2600K was the last mega one. 4790K was fun and could be tweaked alot for noticable improvement, but after that? All you can do is slap on better cooling and use that to remove the built performance limiters that keep a chip within a marketed power spec, that's it.

And now the e-cores have showed up, game over.
 
That's not all you can do. For example you could spend a few extra bucks for an unlocked Intel "K" series CPU.
 
I wonder if would be worth it to overclock my 2320 even though it's not multiplier unlocked.
 
ZodaEX said:
I wonder if would be worth it to overclock my 2320 even though it's not multiplier unlocked.
Click to expand...
Pretty sure non-K SB chips can do +4 on the multi on Z68, so there's that. At least the 2500 non-K + ASUS Z68 I found in the trash a few years back could do that.
 
Depends on how much voltage you're comfortable running? If I really want to I can do an all core 6 GHz overclock but that means pushing 1.45 volts and higher which I'm just not personally comfortable with. So for me a comfortable overclock is 1.3 volts at 5.7 GHz and 4.4 GHz on the ks version.
 
Last edited:
Grebuloner said:
Depends on what CPU/platform you're starting with, what specific performance area you're intending to increase, and how much power you're willing to use and able to cool.

Not much to gain with desktop CPUs, especially for gaming increases. Everything is already binned close to the limit. HEDT has a ton of headroom if you've got the wattage.
Click to expand...
Is it really better with HEDT? Are you saying that if you have a big enough PSU, you should be able to get a meaningful overclock with non-pro threadrippers?
 
sram said:
Is it really better with HEDT? Are you saying that if you have a big enough PSU, you should be able to get a meaningful overclock with non-pro threadrippers?
Click to expand...
Yes. Both TR and the current Xeon-Ws have tons of headroom available to them.

Unlike the mainstream platforms where TDP is just base clocks and turbo goes to much higher power levels, the HEDT parts actually stop boosting at either TDP (TR) or a bit above it (Xeon). But these are parts stuffed with high performance cores, so it doesn't take a lot of frequency to get to those power levels and they aren't being clocked to the absolute maximum from the get go like regular parts are. Plus, the cores themselves are either the exact same silicon as desktop (TR) or close to it (Xeon), so top frequency potential is largely the same, but essentially power constrained.

Gamer's Nexus and der8auer have recent videos of Threadripper OCing that are worth a watch, but they show the big chips going 800W to 1800W(!) for more than 50% gains in performance (GN has the AMD super nerds with them to explain the whole efficiency thing; it's fascinating). This is why you see the boards with 2x PSU support because "big enough" with a GPU or 3 can mean using more than one electrical circuit.

Cooling is also easier since the dies are all spread out. A Noctua U14 with 2 fans pulled 700W off of a W9 3495X.
 
Grebuloner said:
Yes. Both TR and the current Xeon-Ws have tons of headroom available to them.

Unlike the mainstream platforms where TDP is just base clocks and turbo goes to much higher power levels, the HEDT parts actually stop boosting at either TDP (TR) or a bit above it (Xeon). But these are parts stuffed with high performance cores, so it doesn't take a lot of frequency to get to those power levels and they aren't being clocked to the absolute maximum from the get go like regular parts are. Plus, the cores themselves are either the exact same silicon as desktop (TR) or close to it (Xeon), so top frequency potential is largely the same, but essentially power constrained.

Gamer's Nexus and der8auer have recent videos of Threadripper OCing that are worth a watch, but they show the big chips going 800W to 1800W(!) for more than 50% gains in performance (GN has the AMD super nerds with them to explain the whole efficiency thing; it's fascinating). This is why you see the boards with 2x PSU support because "big enough" with a GPU or 3 can mean using more than one electrical circuit.

Cooling is also easier since the dies are all spread out. A Noctua U14 with 2 fans pulled 700W off of a W9 3495X.
Click to expand...
Very interesting. 800W to 1800W and 2 PSU's, oh WOW. I'll take a look at the videos.
 
sram said:
Very interesting. 800W to 1800W and 2 PSU's, oh WOW. I'll take a look at the videos.
Click to expand...
they suuuuuuck the juice when ocing. just turning on pbo bumped the 7980x(or whatever its called) to 600w+. if its hitting 1800w youll need a 30a circuit to run it, iirc.
 
pendragon1 said:
they suuuuuuck the juice when ocing. just turning on pbo bumped the 7980x(or whatever its called) to 600w+. if its hitting 1800w youll need a 30a circuit to run it, iirc.
Click to expand...
May I know how the calculation is done? In the US, the wall outlet gives 110 volts and 15 amps, right? Which is 1650 in watts so that's why you need a circuit with higher amps rating or two circuits if they pull 1800w.

Where I'm the wall outlet gives 220 volts and 13 amps which is ~2800 watts, do I still need another circuit ?

When you use two PSU's, do you connect one to the motherboard only and the other one to the rest of the system? And each PSU's to its own wall circuit to ensure power is enough? I'm wandering how the whole thing is wired.

But in the end I don't want to beat a world record. I just want a noticeable ~ 15% increase in performance which will make me feel good. How much will a moderate overclock draw?
 
sram said:
May I know how the calculation is done? In the US, the wall outlet gives 110 volts and 15 amps, right? Which is 1650 in watts so that's why you need a circuit with higher amps rating or two circuits if they pull 1800w.

Where I'm the wall outlet gives 220 volts and 13 amps which is ~2800 watts, do I still need another circuit ?

When you use two PSU's, do you connect one to the motherboard only and the other one to the rest of the system? And each PSU's to its own wall circuit to ensure power is enough? I'm wandering how the whole thing is wired.

But in the end I don't want to beat a world record. I just want a noticeable ~ 15% increase in performance which will make me feel good. How much will a moderate overclock draw?
Click to expand...
no, i was basically talking out my ass. i know a 110v 15a you want to limit to about 1500w.
probably not.
usually one power the board, one for gpu(s).
then just turn on pbo and call it a day.
 
sram said:
May I know how the calculation is done? In the US, the wall outlet gives 110 volts and 15 amps, right? Which is 1650 in watts so that's why you need a circuit with higher amps rating or two circuits if they pull 1800w.

Where I'm the wall outlet gives 220 volts and 13 amps which is ~2800 watts, do I still need another circuit ?

When you use two PSU's, do you connect one to the motherboard only and the other one to the rest of the system? And each PSU's to its own wall circuit to ensure power is enough? I'm wandering how the whole thing is wired.

But in the end I don't want to beat a world record. I just want a noticeable ~ 15% increase in performance which will make me feel good. How much will a moderate overclock draw?
Click to expand...
Running the CPU at 1800W was all about extreme LN2 cooling. If you were to run regular OC'ing on a water system, 800-1000W is a more "reasonable" (*rolls eyes*) amount of power, plus say, a 4090/RTX 6000 Ada or two for rendering/AI duty would fit on one 220V circuit, but you would still need 2 power supplies. The 2 PSUs would be split between CPU and rest of the system powering merely because of the number of required connectors. 4-5 EPS12V cables are needed to run that kind of CPU power, and that's more than any one PSU can provide without buying extra cables. The boards themselves actually have 2 24 pin motherboard connectors to facilitate dual PSUs without fancy tricks or additional hardware.

15% increase (booooorrrrrring) would be say 500-600W depending on ram. That, plus a single 4090, would fit in a single high-wattage PSU using just the minimum 2 required EPS12V plugs. The efficiency, especially of the 7980X, falls off fast. AMD really nailed the sweet spot of the voltage/frequency curve with that CPU.
 
pendragon1 said:
no, i was basically talking out my ass. i know a 110v 15a you want to limit to about 1500w.
probably not.
usually one power the board, one for gpu(s).
then just turn on pbo and call it a day.
Click to expand...
No to PBO, I can now see that there is more fun in overclocking threadrippers.
Grebuloner said:
Running the CPU at 1800W was all about extreme LN2 cooling. If you were to run regular OC'ing on a water system, 800-1000W is a more "reasonable" (*rolls eyes*) amount of power, plus say, a 4090/RTX 6000 Ada or two for rendering/AI duty would fit on one 220V circuit, but you would still need 2 power supplies. The 2 PSUs would be split between CPU and rest of the system powering merely because of the number of required connectors. 4-5 EPS12V cables are needed to run that kind of CPU power, and that's more than any one PSU can provide without buying extra cables. The boards themselves actually have 2 24 pin motherboard connectors to facilitate dual PSUs without fancy tricks or additional hardware.

15% increase (booooorrrrrring) would be say 500-600W depending on ram. That, plus a single 4090, would fit in a single high-wattage PSU using just the minimum 2 required EPS12V plugs. The efficiency, especially of the 7980X, falls off fast. AMD really nailed the sweet spot of the voltage/frequency curve with that CPU.
Click to expand...
Yeah......I guess you are right. 15% is boring. I've seen the 7995WX overclocked to about 92% more with only air (That's almost double the base):

https://www.google.com/amp/s/wccfte...ps-cinebench-record-breaking-performance/amp/

I didn't know threadrippers are much more overclockable than their mainstream counterparts. Although I think this 7000 generation of threadrippers launch isn't as impressive as before, this overclock-ability makes up for it for me at least. I'm targeting the 7970X threadripper. I wonder how much the gaming performance gap will shrink between this threadripper and Ryzen 7950x or 7950x3d when it is overclocked to about ~ 40-50 %.

So 2 power supplies are a must? I will only use one video card. I would like to avoid two power supplies if possible. If not, then so be it.

Edit: This guy got the 7970X to 5.4 GHz as I can see:

https://en.overclocking.com/review-threadripper-7980x-7970x-and-asus-pro-ws-trx50-sage/8/

So that's what? About 70% increase in clock speed.
 
Last edited:
sram said:
No to PBO, I can now see that there is more fun in overclocking threadrippers.

Yeah......I guess you are right. 15% is boring. I've seen the 7995WX overclocked to about 92% more with only air (That's almost double the base):

https://www.google.com/amp/s/wccfte...ps-cinebench-record-breaking-performance/amp/

I didn't know threadrippers are much more overclockable than their mainstream counterparts. Although I think this 7000 generation of threadrippers launch isn't as impressive as before, this overclock-ability makes up for it for me at least. I'm targeting the 7970X threadripper. I wonder how much the gaming performance gap will shrink between this threadripper and Ryzen 7950x or 7950x3d when it is overclocked to about ~ 40-50 %.

So 2 power supplies are a must? I will only use one video card. I would like to avoid two power supplies if possible. If not, then so be it.

Edit: This guy got the 7970X to 5.4 GHz as I can see:

https://en.overclocking.com/review-threadripper-7980x-7970x-and-asus-pro-ws-trx50-sage/8/

So that's what? About 70% increase in clock speed.
Click to expand...
make up you mind then...
and that math is off. nobody has gotten a 92% oc on a 7995wx. its base speed is 2.5, its boost is 5.1, they managed 4.8 all core. that is not a 92% oc....
7970x boosts to 5.3...
 
pendragon1 said:
make up you mind then...
and that math is off. nobody has gotten a 92% oc on a 7995wx. its base speed is 2.5, its boost is 5.1, they managed 4.8 all core. that is not a 92% oc....
7970x boosts to 5.3...
Click to expand...
I calculated the percentage comparing the all core overclock speed (4.8) to the base speed (2.5) . That's how I got 92%. (4.8 - 2.5) / 2.5 = 0.92
I mean when the cpu boosts not all cores go to the turbo/boost frequency. When you do all core overclock, you force the cores to operate at the overclocked speed, and not the base frequency. That's my understanding.
 
sram said:
I calculated the percentage comparing the all core overclock speed (4.8) to the base speed (2.5) . That's how I got 92%. (4.8 - 2.5) / 2.5 = 0.92
I mean when the cpu boosts not all cores go to the turbo/boost frequency. When you do all core overclock, you force the cores to operate at the overclocked speed, and not the base frequency. That's my understanding.
Click to expand...
thats not right. if anything do the math based on its actually all core speed not the base speed.
 
pendragon1 said:
thats not right. if anything do the math based on its actually all core speed not the base speed.
Click to expand...
What's the all core speed? Doesn't all cores operate at the base clock speed, and when it is under load it boosts (not all cores) to the turbo frequency? From what you are saying I can see that it is a bit more tricky. Cores can run at different speeds depending on the situation. Maybe you need to analyze how the cpu deals with different loads and then come up with an average all core speed?


Edit: I mean if you compare the 4.8 all core overclock to the boost frequency of 5.1, it will be a decrease . That's why I compared to the other known frequency spec which is the base clock. I haven't heard of all core speed or how it is calculated. Thanks.
 
Last edited:
sram said:
What's the all core speed? Doesn't all cores operate at the base clock speed, and when it is under load it boosts (not all cores) to the turbo frequency? From what you are saying I can see that it is a bit more tricky. Cores can run at different speeds depending on the situation. Maybe you need to analyze how the cpu deals with different loads and then come up with an average all core speed?
Click to expand...
dont know. but your math, and the tom's article you may have read with the exact same math and wording, is off. if it boost to 4g all core then youd base the math on that, not the base clock.
 
sram said:
didn't know threadrippers are much more overclockable than their mainstream counterparts. Although I think this 7000 generation of threadrippers launch isn't as impressive as before, this overclock-ability makes up for it for me at least. I'm targeting the 7970X threadripper. I wonder how much the gaming performance gap will shrink between this threadripper and Ryzen 7950x or 7950x3d when it is overclocked to about ~ 40-50 %.

So 2 power supplies are a must? I will only use one video card. I would like to avoid two power supplies if possible. If not, then so be it.
Click to expand...
For gaming, the gap will really depend on a lot of factors, like the game itself, screen resolution, GPU, etc., and how you OC it. Most games aren't going to use 32 cores, and considering the 7800X3D is basically king of gaming at 8 cores, if you were to tweak things so that, say, 8-16 cores are at a very high frequency and then it drops to a lower all core from there, you might see some good results.

2 PSUs are not required so long as everything you are running is under 1600W. I don't know what your primary use case and setup is intended to be, but a moderate OC+high end GPU and some accessories will easily stay under that limit. Add some serious power sucking stuff (2nd, 3rd GPU or certain accelerators), and now you're looking at needing another PSU. My own system running full tilt (up to 350W on CPU) with AICs that use some juice stays under 1000W.
 
pendragon1 said:
dont know. but your math, and the tom's article you may have read with the exact same math and wording, is off. if it boost to 4g all core then youd base the math on that, not the base clock.
Click to expand...
What tom's article? I did the math myself and didn't read anything. I can see why it could be wrong though. A 100% increase is too good to be true. Like I said, it may not be easily calculated because cores boost differently depending on the load I think. With one load you may have 1 core spike to the boost frequency (a game), and with another load you may have 4 cores spiking up to the boost frequency. I don't know I'm just guessing. If there is a right way to calculate the performance increase, please enlighten me.
 
Grebuloner said:
For gaming, the gap will really depend on a lot of factors, like the game itself, screen resolution, GPU, etc., and how you OC it. Most games aren't going to use 32 cores, and considering the 7800X3D is basically king of gaming at 8 cores, if you were to tweak things so that, say, 8-16 cores are at a very high frequency and then it drops to a lower all core from there, you might see some good results.

2 PSUs are not required so long as everything you are running is under 1600W. I don't know what your primary use case and setup is intended to be, but a moderate OC+high end GPU and some accessories will easily stay under that limit. Add some serious power sucking stuff (2nd, 3rd GPU or certain accelerators), and now you're looking at needing another PSU. My own system running full tilt (up to 350W on CPU) with AICs that use some juice stays under 1000W.
Click to expand...
Well, if I end up buying a threadripper system, I will get the 32-core 7970X non-pro threadripper with a compatible motherboard (Most likely Asus), and a 4090, one or two SSD's for OS, and one large capacity spinner HDD for data, 128 RAM kit, and a water cooling kit. That's about it. I may add more drives later but not necessarilly. I use my pc for general home/office use, video playback, occasional gaming, browsing of course (with zillions of tabs open), and the main use is running many VMs with 4 cores and 16 GB of RAM for each. (I tend to run like 5 to 7 VMs for a pentesting lab). So I think my system will not need a 2nd PSU. Thanks for your help and input.
 
sram said:
What tom's article? I did the math myself and didn't read anything. I can see why it could be wrong though. A 100% increase is too good to be true. Like I said, it may not be easily calculated because cores boost differently depending on the load I think. With one load you may have 1 core spike to the boost frequency (a game), and with another load you may have 4 cores spiking up to the boost frequency. I don't know I'm just guessing. If there is a right way to calculate the performance increase, please enlighten me.
Click to expand...
the one thats has the same math and wording as your post. i gave you the numbers in the post you replied to but you have to wait and see what the actual all core clock is....
 
pendragon1 said:
the one thats has the same math and wording as your post. i gave you the numbers in the post you replied to but you have to wait and see what the actual all core clock is....
Click to expand...
I really hate it when people don't believe me. Provide a link for this article if you insist that I have read some tom's article. I would be surprised if there is an article with the exact same wording and math as mine. I'm old school and don't like current social media apps. I like forums and I come here to learn and get advises from people more knowledgeable than me. There is no reason for me to lie in a PC hardware forum.
 
Grebuloner said:
Running the CPU at 1800W was all about extreme LN2 cooling. If you were to run regular OC'ing on a water system, 800-1000W is a more "reasonable" (*rolls eyes*) amount of power, plus say, a 4090/RTX 6000 Ada or two for rendering/AI duty would fit on one 220V circuit, but you would still need 2 power supplies. The 2 PSUs would be split between CPU and rest of the system powering merely because of the number of required connectors. 4-5 EPS12V cables are needed to run that kind of CPU power, and that's more than any one PSU can provide without buying extra cables. The boards themselves actually have 2 24 pin motherboard connectors to facilitate dual PSUs without fancy tricks or additional hardware.

15% increase (booooorrrrrring) would be say 500-600W depending on ram. That, plus a single 4090, would fit in a single high-wattage PSU using just the minimum 2 required EPS12V plugs. The efficiency, especially of the 7980X, falls off fast. AMD really nailed the sweet spot of the voltage/frequency curve with that CPU.
Click to expand...
I can see in der8auer's video that in the most demanding moment (Extreme OC with LN), the 7980X drew ~900 watts itself. The 7970X should draw considerably less specially that I'm only going to use water cooling and perform less extreme OC. A 2000 watt PSU should be more than enough to power a 7970X system intended to be overclocked. My modest conclusion.


View: https://www.youtube.com/watch?v=jnUBYgieQYU&embeds_referring_euri=https%3A%2F%2Fforums.anandtech.com%2F&feature=emb_imp_woyt

One of these beasts should do:

https://www.coolermaster.com/catalog/power-supplies/m-series/m2000-platinum/#image-Item9

https://www.silverstonetek.com/en/product/info/power-supplies/HELA2050Platinum/

https://www.techpowerup.com/review/silverstone-hela-2050-w/12.html

https://www.techpowerup.com/review/cooler-master-m2000-platinum-2000-w/

They are humongous!!!
 
Have you guys seen this:
https://skatterbencher.com/2023/11/...7000 processors,implicit agreement is no more.

Have no idea if it is true, but it is bad news. Overclocking a 7000 series threadripper will leave a permanent mark on the cpu and AMD can know and may not offer you the warranty you should have. Isn't it weird advertising a chip as being overclockable and then preventing you from using its warranty if you used said advertised feature? I know it has always been like this but going to the extent of leaving a permanent mark is too much I think. I just hope it is not true and if yes there will be a way around it.
 
There a bit of a tension to have overclockable so big and up front in the marketing, with little note at the bottom that say:
5 Overclocking and/or Undervolting AMD processors and memory, including without limitation, altering clock frequencies / multipliers or memory timing / voltage, to operate outside of AMD’s published specifications will void any applicable AMD product warranty, even when enabled by AMD hardware and/or software. This may also void warranties offered by the system manufacturer or retailer. Users assume all risks and liabilities that may arise out of overclocking / undervolting AMD processors, including, without limitation, failure of or damage to hardware, reduced system performance and/or data loss, corruption, or vulnerability. GD-106

Intel notice seem much softer:
Overclocking Disclaimer: Unlocked features are present with select chipsets and processor combinations. Altering clock frequency or voltage may void any product warranties and reduce stability, security, performance, and life of the processor and other components. Check with system and component manufacturers for details.


Like maybe, it will remove Dell, Hp, etc... warranty, look with them.

On the other way around, it would be strange for the warranty to not be voided by some form of overclocking, but just setting AMD PBO on the AMD chipset to on and the EPO ddr-5 setting to on should not void the warranty imo.

If PBO on type of overclocking exist on threadripper and does not count as operating outside of AMD specs because it make sure of it, it is just the ability to go outside some ranges that void the warranty, that can OK to me.
 
sram said:
Have you guys seen this:
https://skatterbencher.com/2023/11/21/9-new-things-about-overclocking-ryzen-threadripper-7000/#:~:text=All Ryzen Threadripper 7000 processors,implicit agreement is no more.

Have no idea if it is true, but it is bad news. Overclocking a 7000 series threadripper will leave a permanent mark on the cpu and AMD can know and may not offer you the warranty you should have. Isn't it weird advertising a chip as being overclockable and then preventing you from using its warranty if you used said advertised feature? I know it has always been like this but going to the extent of leaving a permanent mark is too much I think. I just hope it is not true and if yes there will be a way around it.
Click to expand...
pretty sure its the same warning we already get
 
pendragon1 said:
what happened to the 92% OC? ;)
30-40% is probably about right and yeah, you NEED a good water setup to reach it, air and aios will not cut it.
Click to expand...
Yeah I was assuming he meant air cooling. 30% is pretty good, although "the best water setup" sounds pretty expensive. :(
 
You must log in or register to reply here.
Back
Top