Can someone explain Overclocking to me?

[Birblover12]

So I was wondering: If GPU/CPU Manufacturers test their products to go as fast as possible while still staying safe, what is the point of overclocking? If the manufacturers already made it go as fast as possible while being safe, why would they market "overclocking" as a feature? Wouldn't that be counter intuitive since if it was able to go faster safely, they would have made it do so from the factory? And in a similar train of thought, why would they market pushing their product past what's safe, wouldn't that void warranties and break more products?

 

Genuinely wondering is all! Keep in mind I've never overclocked anything before so it's possible I'm misinformed about many things lol 

[Eigenvektor]

31 minutes ago, Birblover12 said:

So I was wondering: If GPU/CPU Manufacturers test their products to go as fast as possible while still staying safe, what is the point of overclocking?

That's not what they're doing though.

 

After a waver has been cut into multiple chips, each chip is tested. Are all eight cores functional? Yes. Are they stable running at x GHz? Yes. Alright that chiplet will go into an R9 7950X. If either answer is no, they will test the next lower core count and/or clock speed and so on.

 

This process is called binning, where they sort their chips/chiplets into multiple discrete categories or bins. These categories are selected to (a) make sense from a marketing perspective and (b) be achievable by anything they produce. Anything that isn't able to make it even into the lowest tier is a waste and a loss of money.

 

However the chiplet they tested to run at 4.7 GHz might be able to run at 4.8 or maybe even 4.9 GHz. That's not a product in their lineup, so they don't care and don't sell it as such. Doesn't mean that you can't take the time to test whether that's possible.

 

It's true that silicon these days is usually very close to what is achievable and overclocking often can't give you much of a performance boost anymore, but it's still doable.

 

Another form of of overclocking is undervolting. The CPU/GPU you bought is run at a voltage that is known to work for anything they produce. Doesn't mean your particular one isn't able to run at lower voltages. Combine undervolting with a power limit and you may have a CPU that performs the same as before, while using less power.

[8tg]

They don’t make it go as fast as possible while still being safe. They make a very wide range of manufactured dies, which have a surprisingly large amount of variance between them, all adhere to one product spec requirement. And as long as that requirement is met, the individual die will be made to conform to it even if it can do more.

 

In all the 14900 dies manufactured, they’re all going to adhere to their required base and boost clock speeds. Even if the die is capable of more, they’re all 14900’s.

Obvious outliers in testing become 14900k’s, and are sold with the guarantee of being at least a 14900, but potentially more.

[RONOTHAN##]

15 minutes ago, Birblover12 said:

if GPU/CPU Manufacturers test their products to go as fast as possible while still staying safe, what is the point of overclocking?

There's a couple different reasons. The biggest is that manufacturers have something called "safety margin" in their products, where in order to make sure their yields are high enough and that they don't have to deal with as many RMA requests from things like mediocre power delivery, they artificially add extra voltage to take that variance into account. Because of that extra voltage at stock, you can overclock the CPU and get more performance out of it without even adjusting the voltage by reducing that safety margin that's unnecessary. 

 

There's also power efficiency, though this admittedly is less of a thing nowadays. Generally, performance goes up linearly with clock speed, but power draw goes up exponentially with clock speed due to the extra voltage needed, and so if the manufacturer decides that they want to target something like 65W but you are fine with a 200W+ CPU, you can overclock it yourself to that power draw and use the extra performance of it while still being safe. 

 

Another thing is just the silicon lottery. Different CPUs have different frequency and voltage characteristic, and because CPU manufacturers have to have their frequency characteristics broad enough to get good enough yields, there are chips out there that are at the bare minimum of those specs, while there are chips that can do much beyond that. Take 6th gen where I have the most experience with different CPUs, at the same voltage level I have had a 6700K that wouldn't do more than 4GHz (probably the worst 6700K to ever exist) I had a 6600K that would do over 4.7GHz. Those would both be rated at 4GHz, and therefore one of those chips has much more performance in it than the other. 

 

 

The biggest misconception here is that manufacturers will have it be as fast as possible while staying safe. In more recent years, they have gotten closer to that point with more advanced boost algorithms and more consistent silicon fabs, but they aren't at that point yet where everything can be right at the edge of stability and cooling. 

[8tg]

And then there’s the marketing side of things.

It’s not a new thing, people have been overclocking their hardware for as long as the concept of a pc has existed. It’s just around the time that Intel became their own dominant chipset manufacturer for their CPUs (because there was a point where they weren’t), they started to offer overclocking oriented options to make it more accessible, at a cost premium.

It took a lot of the niche knowledge and guesswork out of the equation. Instead of researching which via, nvidia, SIS, etc chipset had what features and worked well with specific processors, Intel basically just went “if it has an X in the name it’s meant to overclock”.

And so began a line of high end hardware designed around overclocking, pushing hardware beyond the functional norm. Intel simply saw it as profitable to make it easier to overclock, because if they made it easier to attract more users, those users would buy intels more expensive hardware.

 

Something like skulltrail as a concept simply didn’t exist prior to that. Hardware with the goal being to push it as hard as you can, and get the most obscene performance possible at the time. That wasn’t a marketing point prior to that, out of the box tweaking from a first party.

 

But it exists for the same reason now. The CPUs and platforms for them exist to make it easy, and to attract customers towards more expensive hardware.

[GarlicDeliverySystem]

To the previously mentioned things like binning and the effect of the silicon lottery I would add a bit of historic perspective and also marketing.

 

Many years ago, chips ran at fixed clocks and power consumption set by the manufacturer; the chip itself had no features or even protections in it. This meant having good/reliable cooling and power supply was important and chip manufacturers had to come up with specs that both the PC OEMs could meet, and that their own chips could reliably hit.

 

However, this also meant that a lot of chips could actually outperform these specs, provided they had enough cooling to avoid cooking themselves. Proper cooling was however expensive and most manufacturers/people didn't bother with extra coolers etc. And for a lot of applications it was more important to hit a certain performance threshold and make that as cheap/profitable as possible.

 

Over time features got added to chips to allow thermal protection to shut down before overheating, later on throttling instead. Then you had more dynamic control of the actual clock, with CPUs no longer running full clock all the time but going up/down as required by the workload. Still, in order to make all of this work reliably there was plenty of headroom in the specs to allow for sub-par cooling, silicon etc.

 

These were times where you had chips/product lines that could be overclocked a lot, compared to today. As long as you were willing to risk your chip and void the warranty, invest in some better cooling, other specialized hardware (better motherboard, power supply, memory) and spent time fiddling, you could get a lot of extra performance. Nothing was guaranteed and you could have a dud that did not perform stable above normal specs.

 

Overclocking was thus mostly for enthusiasts that liked to tinker and spend money on stuff, not really for the mass market. Plus, some manufacturers tried to lock it down for better product segmentation and to push people to buy the more expensive chips instead.

 

Now, fast forward and we now have a lot of features that are designed to push any chip as much as it can go, within certain specs. So manufacturers still guarantee minimum clock speeds that can be sustained, they advertise with "up to" kind of speeds attainable in certain scenarios. Overclocking now is mostly marketing imho, apart from scenarios where the performance of a chip is simply limited by the cooling solution and power delivery. There it could make sense to just unlock the chip and let it go as fast and hot as it can get.

[Issac Zachary]

One thing that isn't clear here, and IIRC, is that a certain die, such as a 14th gen CORE Intel, is the same die for all i3's, i5's, i7's and i9's. The problem is that there tends to be imperfections in each CPU that is built, so certain cores and other components may not work. So they test the CPU and see what it can do, and if it fails to run X number of cores, they lock out those cores and call it an i3 or i5 and so on.

 

But the performance differences aren't limited to set i3, i5, i7 and i9 categories. You may have one Core i9 that can reach a higher clock speed than Back in the day you could try to push a CPU beyond the limit. Some would do it, others would have to be set back to normal settings in order to work. So for a time there you could either play the CPU lottery and try to overclock a CPU, or buy one that someone else had tested and proved that would over clock by X amount (for more money of course).

 

But then CPU manufacturers started catching on to this and decided to control overclocking for their own financial gain.

[thrasher_565]

over the years ocing i think they didnt care of didnt have all the testing for. and probly some time whent by were people didnt oc as there was too much risk as it cost too much for a pc.

pc were only realy afordable in the $1000 range (ill have to ask my dad how much he paid for the first pc) but i  payed $800 for my p4 back in the day and that was with a $20 case and psu im guessing any way people found ways to oc it. some thow pins on the socket some by jumping a pin on the cpu. some mb had a jumper you moved and the goold old turbo button...but that was before my time. even ocing gpus there was not many safty's inplace and can easly push it too far and kill it. then the i7 920 came out and from 2.66 oc to 4.0 prity easly made it seem too good to be true. some people would buy the cheaper cpu and oc it to a hier tire cpu for free. well maybe more cooling was needed. maybe the made deals with cooler makers that they keep cpu low so others can sell cool parts

 

now adays the manufacture self bins and oc them and sells em keep all the moeny and making new tires. that is why there so much gpus and cpus on the market.

 

the good old quake and doom devs could not afford to buy a pc so they used there work pcs after work. later on a few guys saving moeny to buy one and put it in a basement like micro soft...

Edited March 18 by thrasher_565

[Spindel]

Overclocking nowadays is nothing really worth it compared to the hasstle in most cases.

 

As an old timer let me tell you when certain CPUs where shipped as a 600 MHz part but you could (with some fiddling) OC it fairly easy to 900 MHz and in a lot of cases to around 1000 MHz (all practically usable no strange sub ambient set ups) That gave a meaningful performance boost compared to today when going from 4,7 to 4,8 or 4,9 GHz.

 

According to me most stuff with OC ”capabilities” today is mostly to be able to get more money from consumers buying in to this (again for practical use and not sub ambient set ups).

[cmndr]

On 3/16/2024 at 10:10 PM, Birblover12 said:

So I was wondering: If GPU/CPU Manufacturers test their products to go as fast as possible while still staying safe, what is the point of overclocking?

Here's the thing, historically, the companies were NOT pushing their parts to the limits. They sand bagged because... why not? Slightly reduced warranty claims and it helped segment their products better, which was good from a marketing stand point. 

 

17 years ago I could overclock an Intel Celeron 420 from 1.6GHz to 3.2GHz easily. 100% more performance by changing ONE setting on a motherboard. I think I didn't even need to change the voltage. 

My Core 2 Duo e6400 went from 2.13Ghz to 3.2Ghz at stock voltage. It also hit 3.6Ghz relatively easily. 

 

These days getting 2x the speed easily doesn't happen. It's more like 2% more. With huge down sides. 

 

On 3/17/2024 at 10:37 AM, Issac Zachary said:

One thing that isn't clear here, and IIRC, is that a certain die, such as a 14th gen CORE Intel, is the same die for all i3's, i5's, i7's and i9's.

Intel actually has a few dies that they use for different SKUs. It's possible that an i9 die could become a very low tiered SKU but in practice it doesn't really happen. Yields aren't that bad. 

[Zodiark1593]

On 3/16/2024 at 10:10 PM, Birblover12 said:

So I was wondering: If GPU/CPU Manufacturers test their products to go as fast as possible while still staying safe, what is the point of overclocking? If the manufacturers already made it go as fast as possible while being safe, why would they market "overclocking" as a feature? Wouldn't that be counter intuitive since if it was able to go faster safely, they would have made it do so from the factory? And in a similar train of thought, why would they market pushing their product past what's safe, wouldn't that void warranties and break more products?

 

Genuinely wondering is all! Keep in mind I've never overclocked anything before so it's possible I'm misinformed about many things lol 

Yields are probably the largest impediment, as only an ever-diminishing portion of chips can reach higher speeds, at given power targets. At some point (whether power, or finding chips that can go fast enough at all), it becomes not worthwhile to make a dedicated sku. 

Additionally, nowadays, chips are pushed pretty close to the edge out-of-the-box. Unless you’re willing to dump a fair bit more power into it, it’s hard to achieve meaningful gains anywhere. Definitely unlike the C2D days, where you can push a 1.8 GHz chip to near 3.2, or even Celery days, where you can outright double clocks. 

[Mark Kaine]

this is also why most chips are getting too high voltage out of the box, because manufacturers can be sure, yes, a wide variety of chips will work at this voltage.

 

but in reality most chips can be undervolted,  super easy trick to get more performance,  because now it has a higher temp (and voltage) margin which means it'll clock higher at about same power usage as stock (or even less power usage) 

 

hence traditional "overclocking" where you tell the chip "clock higher" possibly even with more voltage than stock is outdated as heck. all you're doing is make it overheat long term...

 

(it can still work but you'll need top notch cooling,  preferably LN2)