Does locking a CPU frequency to a higher state (even OC) have any downsides?

[IR Baboo]

I got into an argument with a friend that thinks generally is not a good idea to manually lock a CPU to a high frequency, or even overclocking by locking the frequency, in a stable scenario. His main argument is: 'why would I want it to run like that, because most of the time my computer sits idle, as if a car engine is revved without moving'. But from my experience, the analogy is not quite so because a CPU locked at a high frequency and locked voltage, will barely consume anymore power in idle states than a CPU that behaves in default configuration.

 

But what about the clock cycles? Are there any advantages of having the CPU locked and ready for when the load demands high performance? Are there any disadvantages?

Wikipedia in CPU clocks states:

 

Quote

 

After each clock pulse, the signal lines inside the CPU need time to settle to their new state. That is, every signal line must finish transitioning from 0 to 1, or from 1 to 0. If the next clock pulse comes before that, the results will be incorrect. In the process of transitioning, some energy is wasted as heat (mostly inside the driving transistors).

 

Can this count as an advantage towards locking the frequency?

 

Also from Wiki that kinda answers this:

Quote

 as clock rates increase dramatically, is the amount of heat that is dissipated by the CPU. The constantly changing clock causes many components to switch regardless of whether they are being used at that time. In general, a component that is switching uses more energy than an element in a static state. Therefore, as clock rate increases, so does energy consumption, causing the CPU to require more heat dissipation in the form of CPU cooling solutions.

 

In my personal observation, with my my TR3960x with locked voltages and locked clocks, first at 1.8Ghz all core and the second at 3.6Ghz all core. So essentially double the frequency in idle, with like just 10Ws more in idle power. Bare in mind however that this is a rather beastly CPU with 24 cores and the frequency is all core. I bet the delta in idle power gets smaller as CPU core count gets smaller, meaning it is directly correlated to how powerful the CPU is overall.

 

 

I would like to know more about this from an engineering / electrical POV.

[KnoT]

Currently apart from stability testing or A/B cooler testing there are not advantages, Changing clock states takes nano seconds since (Intel Speed Shift), generally you might be loosing Single core performance tho, as most modern CPU will clock higher while doing 1core work load.

[Mark Kaine]

no, its not 1990 anymore,  just leave your cpu at stock and don't use sketchy "overclocking" software.

[Mark Kaine]

Quote

Does locking a CPU frequency to a higher state (even OC) have any downsides?

TLDR; it has ALL the downsides, heat, shortened longevity, performance / stability issues, clock stretching... 

 

 

 

[Le_Taken]

2 minutes ago, Mark Kaine said:

no, its not 1990 anymore,  just leave your cpu at stock and don't use sketchy "overclocking" software.

You say "sketchy...software" but overclocking can just be done via the bios ( supported Chipset obv ) which I assume is the case here given I don't say many oc softwares orientated to the cpu and given those that do don't tend to do anything unless they were integrated to a firmware level ( with that kind of access you'd hope it be coming from a reputable source anyway) 

[Mark Kaine]

1 minute ago, Le_Taken said:

You say "sketchy...software" but overclocking can just be done via the bios ( supported Chipset obv ) which I assume is the case here given I don't say many oc softwares orientated to the cpu and given those that do don't tend to do anything unless they were integrated to a firmware level ( with that kind of access you'd hope it be coming from a reputable source anyway) 

of course it can, but there aren't really many benefits outside of benchmarking (which often is irrelevant to real life performance because clock stretching etc is a thing)

[HorseBattery]

I'm underclocking my CPU (5800X3D) with 4GHz all core set in the BIOS. It doesn't get to boost so it always runs very conservative voltages and never gets hot or loud. When cores aren't being loaded they use minimal power.

[Applefreak]

32 minutes ago, IR Baboo said:

So essentially double the frequency in idle, with like just 10Ws more in idle power

Modern CPUs have dozens of power saving methods only one of them is dropping clock speeds. So unless you compare both clock speeds at the same workload, you won't see much difference in power drawn. An idle CPU may clock down the internal bus speed, disable cores or even clusters (CCX on AMD) to save on power. 

 

The benefits would be that you are forcing your CPU to stay at a certain clockspeed that you can control, may that to achieve a higher score or to compare silicon quality or for some other clock related issue like timing other controllers etc. 

The disadvantages are as follows: Higher thermals, more stress on the system, little higher power drawn, a lot higher power drawn under load, possible instabilities long term, degradation of the CPU (couple of years instead of a decade or more), of course you loose your warranty if something goes bad. Also keep in mind the extra stress of messing with voltages effects the motherboard and even the ram if you mess around with that too. 

As for data integrity, you do introduce a lot more errors, even with ECC you cannot compensate for them all so not something you want to consider on a mission critical machine. 

As mentioned before and time and time again, there is no real world benefit in OC anymore, maybe a little but just that much. 

If you really have the expertise and can easily replace a broken CPU and or motherboard, go for it (meaning you have money to spend as OC is your only hobby). The only thing I can recommend, if you have a decent cooling setup (water) is to set your board to an all core boost if you are using all cores at the same time. The CPU will then still be able to clock down and adjust to avoid damaging components and cause system stabilities. Not all board however support that function and not all CPUs are suited for it in the same way. Your workstation CPU is probably not one of them. 

[IR Baboo]

11 minutes ago, HorseBattery said:

I'm underclocking my CPU (5800X3D) with 4GHz all core set in the BIOS. It doesn't get to boost so it always runs very conservative voltages and never gets hot or loud. When cores aren't being loaded they use minimal power.

I did the same thing with my TR. Locked all core frequencies, locked voltage for stability and it runs rock solid and way cooler for less power on load. The only drawback is the amount of time I spent testing different voltages for the select frequencies, under various scenarios and workloads.

 

13 minutes ago, Applefreak said:

An idle CPU may clock down the internal bus speed, disable cores or even clusters (CCX on AMD) to save on power. 

 

The benefits would be that you are forcing your CPU to stay at a certain clockspeed that you can control, may that to achieve a higher score or to compare silicon quality or for some other clock related issue like timing other controllers etc. 

The disadvantages are as follows: Higher thermals, more stress on the system, little higher power drawn, a lot higher power drawn under load, possible instabilities long term, degradation of the CPU (couple of years instead of a decade or more), of course you loose your warranty if something goes bad. Also keep in mind the extra stress of messing with voltages effects the motherboard and even the ram if you mess around with that too. 

As for data integrity, you do introduce a lot more errors, even with ECC you cannot compensate for them all so not something you want to consider on a mission critical machine. 

As mentioned before and time and time again, there is no real world benefit in OC anymore, maybe a little but just that much. 

If you really have the expertise and can easily replace a broken CPU and or motherboard, go for it (meaning you have money to spend as OC is your only hobby). The only thing I can recommend, if you have a decent cooling setup (water) is to set your board to an all core boost if you are using all cores at the same time. The CPU will then still be able to clock down and adjust to avoid damaging components and cause system stabilities. Not all board however support that function and not all CPUs are suited for it in the same way. Your workstation CPU is probably not one of them. 

For all my day to day workstations I actually under clock the CPU and lock the frequency and voltage to that sweet spot of best efficiency. This returns best power consumption for the relative performance desired, rock solid stability (tests are done in various workloads, with and without AVX and stress tested for hours at a time), that results also in good thermals, less strain on the cooling solution, less strain on components overall etc. So yeah, instead of letting it default turbo to 4.5Ghz on some cores and hit that 280W TDP, I leave on the table some marginal performance by locking all cores to 3.8Ghz (since my use-case usually lights up all cores possible), and settle for 215Ws power consumption, for just a few Ws increase in power consumption when idle.

 

For OC, as a hobbyist, I will for sure lock my 13900KS well above 6GHz once I finish the build.

[IR Baboo]

1 hour ago, Mark Kaine said:

TLDR; it has ALL the downsides, heat, shortened longevity, performance / stability issues, clock stretching... 

 

 

 

What? I beg to differ on all counts, except for clock stretching, a term I'm unfamiliar with...

 

By this very definition of usage, 13900K / 13900KS and even AMD high end users should have no party in the OC/tuning world!

 

There are many shades of grey in the computing world when it comes to unlocked CPUs. I'd divide them in 3 categories, based on manufacturer's base, turbo designs and OC limits. I'd say first category is under-clock to under-base frequency, second between base and turbo, and third (OC) over turbo. In at least the first two categories I'm willing to bet that the aspects you pointed out are nonissues, because essentially the CPU runs within specs but in a locked state. For the last category...that is also not so clear cut. If it's a mild OC I'm willing to also bet that locked frequency and voltages will not degrade the CPU significantly over the typical 5 year lifespan or so of a computer part that eventually gets replaced anyway. The degradation gets probably higher and higher with how extreme the OC gets and how efficient the cooling solution is.


For example my 3960x runs @3.6Ghz under the base frequency of 3.8Ghz but locked on all cores at the lowest voltage possible for super stability (my case 1.0v). In idle, it sits at just a few more Watts (80ish) compared to if the CPU would have been locked at 1.8Ghz (70ish). So that is negligible in all aspects (heat, shortened longevity, performance / stability issues). 

[Somerandomtechyboi]

1 hour ago, IR Baboo said:

In my personal observation, with my my TR3960x with locked voltages and locked clocks, first at 1.8Ghz all core and the second at 3.6Ghz all core. So essentially double the frequency in idle, with like just 10Ws more in idle power. Bare in mind however that this is a rather beastly CPU with 24 cores and the frequency is all core. I bet the delta in idle power gets smaller as CPU core count gets smaller, meaning it is directly correlated to how powerful the CPU is overall.

Only thing i can think of would be having more consistent performance as cores wont be jumping around in speed and also being able to get higher multicore clocks when doing all core oc as pbo dont think it can run 4.7-4.8ghz allcore but no real experience with it since i still static oc

 

Idle draw, idle thermals are definitely gonna be sacked but kinda irrelevant as their impacts are minimal at best, were talking 5 vs 10w with all power saving stuff still enabled and maybe 30 vs 40c, not a big deal

 

3 minutes ago, IR Baboo said:

I did the same thing with my TR. Locked all core frequencies, locked voltage for stability and it runs rock solid and way cooler for less power on load

^^^

Exactly what i was about to say

 

Under load youll hit higher freqs with the same power draw or the same freqs with lower power draw compared to stuff like pbo, and under load power draw is definitely important unlike the minor increase in idle draw (sure 2x but when we look at the number itself its pretty damn small), caveat obviously being singlecore performance

 

So to sum it up

Pros : maybe more consistent performance, higher allcore clocks/lower power draw under load (depending on tune)

 

Irrelevant cons : idle draw, idle temps

Actual cons : lower single core perf cause no single core boost, takes time to actually tune the thing

 

58 minutes ago, Applefreak said:

degradation of the CPU (couple of years instead of a decade or more)

And ill just call bullshit here like wtf if this is the case why the hell are all those ancient core2 systems still alive and well after 15-18 years? Heck theyll still oc a fair amount depending on binning, not to mention cedar mill pentium4 and celeron d still around and clocking 5ghz+ with ease both bios and windows (if you got the cooling) even today and those things actually ran hot

 

were still talking decades if its one of those conservative ocs that dont push past 1.4v or whatevers the efficiency zone aka best v/f of nowadays chips, maybe in my hands where i crank the shit out of the voltages (particularly imc related ones) thatd be an actual point but for most ppl that arent insane they wont be running high volts simply because they either cant cool it or they dont deem the minor clock increase going into diminishing returns zone is worth it

 

If youve never overclocked before this is a pretty common misconception

 

 

12 minutes ago, IR Baboo said:

For OC, as a hobbyist, I will for sure lock my 13900KS well above 6GHz once I finish the build.

Sorry to say but i dont think youll be able to hit 6ghz on all cores unless you run the thing under phase change cooling and disable all the ecores

 

imo ryzens the best bet for 6g allcore since theyll be alot easier to cool to subzero temps on a phase change cooler and hence crank the clocks up a ton while still running cool cause the cooler you can get the chip the less power it draws and the more it ocs so pretty much a positive feedback loop

 

i would totally be crazy enough to daily a cascade phase change or JT phase change cooler but i kinda like the mobility of a laptop so id rather opt for a diy behemoth of a laptop instead of a fixed desktop, and i currently have no money =p  but used ac units are pretty cheap so when i have 100$ to burn guess i can just buy a used outdoor ac and try cobbling together a cpu block just for shits and giggles and for overclocking ofc

 

2900 ddr3 on nehalem with an x58a ud3r, just to give an idea of the kind of oc nut i am

[Applefreak]

5 minutes ago, Somerandomtechyboi said:

And ill just call bullshit here like wtf if this is the case why the hell are all those ancient core2 systems still alive and well after 15-18 years? Heck theyll still oc a fair amount depending on binning, not to mention cedar mill pentium4 and celeron d still around and clocking 5ghz+ with ease both bios and windows (if you got the cooling) even today and those things actually ran hot

 

were still talking decades if its one of those conservative ocs that dont push past 1.4v or whatevers the efficiency zone aka best v/f of nowadays chips, maybe in my hands where i crank the shit out of the voltages (particularly imc related ones) thatd be an actual point but for most ppl that arent insane they wont be running high volts simply because they either cant cool it or they dont deem the minor clock increase going into diminishing returns zone is worth it

 

If youve never overclocked before this is a pretty common misconception

Try OC on a real processor that has everything linked to the Voltage on the Northbridge. Those Core 2 machines that are still around are just regular ones. Except for a few series OC was bad and got worse over time. I've destroyed a couple of them as well as Athlons at the time. 1.4V is nothing, in the 90s we ran 3.3V stock, those were hard chips to control, there was only FSB, voltage and if you were lucky you could unlock the multiplier. breaking them was easy. Yes modern chips are more resistent to user error but still extreme OC on air with high temps all the time won't do you any good. I was only lucky once with a K6-2 I was able to push to 433.25 MHz, still runs today, good for XP and older games from the late 90s. Had a pretty decent Q6700 stable at 3.6 on water that I had to lower clocks about 2 years ago, a couple months late it died. That processor was working all it's life though not sitting idle processing youtube and twitch streams. You may have had better luck in the silicon lottery here. Btw. those Cedar Mill processors were not available to me, not even for a lot of money, only could get Prescott ones and those did not like higher clocks much.

[IR Baboo]

6 minutes ago, Somerandomtechyboi said:

...

 

Irrelevant cons : idle draw, idle temps

Actual cons : lower single core perf cause no single core boost, takes time to actually tune the thing

I think the same too but alas I'm no expert and thought I'd ask others too about this because the thing is, my friend I think didn't grasp that locked core w/ locked voltage does not = high power draw under idle like it would mean under load. Yeah voltage is locked but load counts like a shit ton and that's what fires up the current (IOUT) measured in amps. And frankly few people actually realize this! I myself realized it relative recently after many years of tooling with the BIOS with unlocked CPUs. 

6 minutes ago, Somerandomtechyboi said:

....

 

 

Sorry to say but i dont think youll be able to hit 6ghz on all cores unless you run the thing under phase change cooling and disable all the ecores

...

Frankly you may be right. I build recently for a kid, a custom water cooled PC and was shocked on how hard was to cool a 13900K in a 2 x 360 loop. As an AMD user myself I was shocked really. The thing would thermal throttle like non-stop in stock scenarios in stress tests. But it was his choice.

 

As an enthusiast, however, I got the chance to acquire a golden sample 13900KS, that's been professionally dellided and resealed with supercool directdie with LM inbetween. Will attempt to hook it to a Mora 420 and see if that can keep it in check at 6Ghz+ on P cores and 5GHz on E cores (haha yes aiming high...landing low probably  )

 

[Somerandomtechyboi]

3 minutes ago, Applefreak said:

Try OC on a real processor that has everything linked to the Voltage on the Northbridge. Those Core 2 machines that are still around are just regular ones. Except for a few series OC was bad and got worse over time. I've destroyed a couple of them as well as Athlons at the time. 1.4V is nothing, in the 90s we ran 3.3V stock, those were hard chips to control, there was only FSB, voltage and if you were lucky you could unlock the multiplier. breaking them was easy. Yes modern chips are more resistent to user error but still extreme OC on air with high temps all the time won't do you any good. I was only lucky once with a K6-2 I was able to push to 433.25 MHz, still runs today, good for XP and older games from the late 90s. Had a pretty decent Q6700 stable at 3.6 on water that I had to lower clocks about 2 years ago, a couple months late it died. That processor was working all it's life though not sitting idle processing youtube and twitch streams. You may have had better luck in the silicon lottery here. Btw. those Cedar Mill processors were not available to me, not even for a lot of money, only could get Prescott ones and those did not like higher clocks much.

Pre 775 and am2 are now pretty damn rare atleast here in indo so dont think ill be able to oc one of those ancient chips, but still 20 years ago isnt gonna be the same as nowadays cpus in terms of behaviour, not old enough to experience the days of physical modding and not technically informed enough and dont have the tools to do a basic voltmod on todays mobos

 

kinda surprised those old chips werent as resistant considering ive run 1.8v through some wolfdales before on air just screwing around with 5ghz bios boot and running like 70c in bios even 1.85v through an e5800 and like 85c idle in bios iirc for 5ghz and none ever died or even degraded, only fatality so far has been an i3 540 but oh well rip a chip worth 2$ and a mid sample only capable of 4.5g at 1.5v, kinda surprising since ive also bios booted 5g on an x5660 and that didnt die but maybe 1156 westmere is just fragile, i mean it has 2 dies not a monolithic chip

 

Thing is were talking about new chips so theyre quite abit more resistant to high temps and dont volt rollover as much as older chips, literally cant go much over 85c on wolfdale >1.4v otherwise itll start destabilizing, where ive seen ryzen 3100 and 3300x oc results being able to push 4.5ghz at 1.45v whilst running at 95c so clearly times have changed

 

Im in a pretty goofy middleground where i dont have access to new hardware but i basically ignore the old standards of "safety" so bye bye bullshit 1.35v vtt max hello 1.6v max for 32/45nm and 1.7v for 65nm. Hence the oc results you see above, maybe 1.65v vtt is overkill when uncore isnt even running at 3ghz in that pic but ill continue testing desync uncore memclk on nehalem once midterms are finished, pretty insane imc, sandy bridge imc looks like a complete joke compared to nehalem

 

So far no real fatalities aside from that poor i3 which seems to have just been bad luck or 1156 westmere simply being fragile even with the liberal volts i use, i simply show no fear when using voltage simply because i know the cpu cant die from such low volts, only 2.1v+ maybe theyll start dropping dead, pushing 45nm to 1.8v, pushing vtt on 45nm and 32nm to 1.75v on some instances, and vdimm is where i pull no punches unless the ics simply dont scale with that much vdimm, i wanted to test my 1gbit f die on this board and immedeatly the first voltage that comes to mind is 2.6v but toned that down to 2.4v and hey it booted 2400, unfortunately 2800 didnt work even with 2.6v set, and unfortunately mobo doesnt go higher than 2.6v, keep in mind dummies back in the day thought 1.65v would "damage" their chips with no proof to back it up, well heres me running almost 1v over that and nothing happened, ram is fine, cpu is fine, all is fine, literally just debunking this old bullshit

 

 

Im curious as to that q6700 though, what vcore and how long has it been running at 3.6ghz? Obviously 100% load constantly will degrade faster so degradation will be noticable alot sooner than normal use, but if its been over a decade id still consider it pretty damn tough, iirc 3.6g usually needs around 1.4-1.5v on 65nm

 

33 minutes ago, IR Baboo said:

As an enthusiast, however, I got the chance to acquire a golden sample 13900KS, that's been professionally dellided and resealed with supercool directdie with LM inbetween. Will attempt to hook it to a Mora 420 and see if that can keep it in check at 6Ghz+ on P cores and 5GHz on E cores (haha yes aiming high...landing low probably  )

Welp that explains the 13900ks, but hey its better landing low than landing just before your target

 

That 2900 seems to be either imc or mobo hardwall, my goal was 3000 ddr3, fucks sake so damn close ;-;

[Applefreak]

3 minutes ago, Somerandomtechyboi said:

iirc 3.6g usually needs around 1.4-1.5v

Yup, 1.41 got it stable, my sample was not that great, needed more voltage to be be stable. Tried for more but couldn't get it to post. It used to be a file server and last was used as a steam cache / game server. After the CPU went up in smoke (SMDs shorted out on the CPU), I've replaced it with a conservative Q9550 with a mild 3.166 GHz OC on it, just because it would be a waste of a perfectly good motherboard otherwise. My best results were on a P4 2.4 that went to almost 3.0 (2.9 something with the multiplier) stable with the Thermaltake Aquarius external watercooler. Still have the unit in storage but the pump was leaking and it has absolutely not head pressure whatsoever compared to a DDC or D5. Modern CPUs are not as much fun anymore to OC, too much safety like modern cars, maybe I should get my hands on some LNG.... that would be fun. Anyway good luck to all your OC adventures out there. 

[Mark Kaine]

1 hour ago, IR Baboo said:

What? I beg to differ on all counts, except for clock stretching, a term I'm unfamiliar with...

 

By this very definition of usage, 13900K / 13900KS and even AMD high end users should have no party in the OC/tuning world!

 

There are many shades of grey in the computing world when it comes to unlocked CPUs. I'd divide them in 3 categories, based on manufacturer's base, turbo designs and OC limits. I'd say first category is under-clock to under-base frequency, second between base and turbo, and third (OC) over turbo. In at least the first two categories I'm willing to bet that the aspects you pointed out are nonissues, because essentially the CPU runs within specs but in a locked state. For the last category...that is also not so clear cut. If it's a mild OC I'm willing to also bet that locked frequency and voltages will not degrade the CPU significantly over the typical 5 year lifespan or so of a computer part that eventually gets replaced anyway. The degradation gets probably higher and higher with how extreme the OC gets and how efficient the cooling solution is.


For example my 3960x runs @3.6Ghz under the base frequency of 3.8Ghz but locked on all cores at the lowest voltage possible for super stability (my case 1.0v). In idle, it sits at just a few more Watts (80ish) compared to if the CPU would have been locked at 1.8Ghz (70ish). So that is negligible in all aspects (heat, shortened longevity, performance / stability issues). 

again, no. its not 1990 anymore, cpu overclocking is in almost all case useless and often detrimental. 

 

 

 

1 hour ago, IR Baboo said:

except for clock stretching, a term I'm unfamiliar with...

that's your number one problem,  you think higher frequencies,  more performance,  when its not (necessarily) 

[Mark Kaine]

30 minutes ago, Somerandomtechyboi said:

degradation

that's the thing, the chips will degrade, just varies by how conservative/ knowledgeable someone is.

 

in almost all cases you're better off undervolting than overclocking nowadays,  that goes for gpus too ofc.

 

 

^which btw is more a software issue than hardware,  its just how modern boot algorithms work, and rightly so!

[Somerandomtechyboi]

9 minutes ago, Applefreak said:

Yup, 1.41 got it stable, my sample was not that great, needed more voltage to be be stable. Tried for more but couldn't get it to post. It used to be a file server and last was used as a steam cache / game server. After the CPU went up in smoke (SMDs shorted out on the CPU), I've replaced it with a conservative Q9550 with a mild 3.166 GHz OC on it, just because it would be a waste of a perfectly good motherboard otherwise. My best results were on a P4 2.4 that went to almost 3.0 (2.9 something with the multiplier) stable with the Thermaltake Aquarius external watercooler. Still have the unit in storage but the pump was leaking and it has absolutely not head pressure whatsoever compared to a DDC or D5. Modern CPUs are not as much fun anymore to OC, too much safety like modern cars, maybe I should get my hands on some LNG.... that would be fun. Anyway good luck to all your OC adventures out there. 

How do the smds just randomly short themselves? Kinda doubt thats anything to do with the oc since i also have a dead e5645 which seems to be shorted out but it just randomly happened while being unused and sitting ontop a table, just a guess since when i power the thing on it heats up ludicrously fast, like burning hot in 1 second whereas a normal working 6 core westmere would take 3-5 secs, and how much did you need to downclock it?

 

Yea new cpus arent fun to oc when all their potential has been squeezed out and then some if the 13900ks is anything to go by, but cpu oc in general can be pretty boring when its just set vcore + set multi and done, even on older bclk/fsb platforms, cant really crank volts either since newer platforms arent as volt resistant as old ones where i could run 1.6/1.7v without issue now i cant go past 1.5v without noticably degrading, not to mention heat so youd only have fun once you cool them so subzero temps so straight up unreachable for anyone short of oc nuts crazy enough to build and run a phase cooler not to mention actually setting those things up would be a massive pain

 

hence why ive taken my oc interests to ram simply because its underappreciated, shitloads of headroom for oc if that 2900 screenie is anything to go by (1333 transcend 1gb sticks with pscs), and dirt cheap since all you need is a cheap bare pcb with a specific ic, its straight up cheaper than buying normal 3200/3600 rams while being far faster. Not boring either since theres an actual challenge with the crapload of timings to adjust, high freq can be hit with ease but performance at those high freqs is a different story

[Somerandomtechyboi]

4 minutes ago, Mark Kaine said:

that's the thing, the chips will degrade, just varies by how conservative/ knowledgeable someone is.

 

in almost all cases you're better off undervolting than overclocking nowadays,  that goes for gpus too ofc.

 

 

^which btw is more a software issue than hardware,  its just how modern boot algorithms work, and rightly so!

Yea manufacturers just pushing the things too far and too conservative stock settings so they can sell the crappier binned stuff hence the power savings from undervolting

 

im very liberal with voltage but since older hardware looks to be mostly bulletproof even running 1.7v doesnt seem to phase any of my cpus, ill have to do some longer testing for degradation but thatll probs come up soon since theres a cheap untested w3680 for sale, or ill just daily a nehalem because ram speed go vroom vroom without a crappy asus board that cripples nehalem with that 1:2 bullshit

[Mark Kaine]

6 hours ago, HorseBattery said:

I'm underclocking my CPU (5800X3D) with 4GHz all core set in the BIOS. It doesn't get to boost so it always runs very conservative voltages and never gets hot or loud. When cores aren't being loaded they use minimal power.

i mean thats also a way of doing it, but if you want more performance undervolting with pbo2 is the thing,  mine boosts to max frequency *on several cores* (which is impossible "according to internet lol") and just runs cooler while also using less power (slightly) 

[Mark Kaine]

3 hours ago, Somerandomtechyboi said:

Yea manufacturers just pushing the things too far and too conservative stock settings so they can sell the crappier binned stuff hence the power savings from undervolting

 

im very liberal with voltage but since older hardware looks to be mostly bulletproof even running 1.7v doesnt seem to phase any of my cpus, ill have to do some longer testing for degradation but thatll probs come up soon since theres a cheap untested w3680 for sale, or ill just daily a nehalem because ram speed go vroom vroom without a crappy asus board that cripples nehalem with that 1:2 bullshit

yeh, hence newer cpus benefit massively from uv, with older cpus its different ig. 

 

also if you don't see degradation after 2+ years (?) its probably fine,  just a lot of people do and thats just un-economical .

[Somerandomtechyboi]

8 minutes ago, Mark Kaine said:

yeh, hence newer cpus benefit massively from uv, with older cpus its different ig. 

 

also if you don't see degradation after 2+ years (?) its probably fine,  just a lot of people do and thats just un-economical .

why keep a cpu for more than 3-5 years anyways? As long as it dont degrade much within that period you are fine, rip to whoever you sell it to tho xD

 

Welp with my i7 930 having a far superior imc than my i7 950 i think thatd be a prime candidate for me to daily simply because 3200+ ddr3 on a bloody nehalem on air, honestly starting to doubt wether a w3680 can beat its imc, and it needs like 1.6v vtt minimum which is problably what ill daily it at unless i can crank the uncore a decent bit if i run it at 1.7 or 1.8v though i may be abit more conservative on this thing cause i kinda dont wanna wreck a golden sample incase this thing has a golden imc, core 2 are simply just stupid to try and daily, 1156 has shit imc why the heck would i use that over a 1366 westmere, same goes for a 2500k, and a 4670k is still like 20$ with z87 still being pricey af, so this nehalem is a perfect daily cause the imc isnt dogshit and i can daily ludicrous freq

[IR Baboo]

Actually, now that I think of it, it's quite irrelevant if one locks the clock & voltages because unless one finds out how to completely disable core sleeping states (which I don't btw), the CPU (AMD or Intel) will behave the same in idle scenarios, meaning it will drop the effective clocks immensely and only respect the clock fixed ceiling when under load. So in conclusion unless one needs peak single core performance, in that in default (stock) CPU config the turbo boost will opportunistically boost best cores to highest frequencies with a ton of power to sustain it, locking the CPU all core frequency to a desired level with a stable fixed voltage (yes, that requires testing) is actually best for multithreaded workloads and draws probably less power than the default under turbo scenarios with like ~5% or so loss in overall performance, based of course on the desired all core ceiling