How do you guys feel about the current state of overclocking?

[yoadknux]

I'm not exactly an overclocking veteran, but I've had my fair share of hardware throughout the years. I always overclocked not to break records or get high benchmark scores, but to gain actual performance, especially for gaming. Somehow I feel that the gains from overclocking have dropped over the years.
Back then, I overclocked my Q6600 & Q9400 on abit IP35 Pro to I think 3.2GHz each (~25% increase). I overclocked my i7 2600k to 4.5GHz (~16% increase), overclocked my 4930k to 4.5GHz (~14% increase). Now with 8700k, turbo is 4.7GHz, and I can only manage 4.9GHz... that's less than 5% increase from stock turbo. I am limited by high temps that can only be resolved by delidding, but even then it seems most processors aren't that much stable over 5.1GHz.
I also liked GPU overclocking, especially 680GTX since it had a good custom bios. Now with my 1080ti, with watercooling and overclock, I can only reach about 10% higher memory speed and 4% higher core speed, which translate to roughly 6% more performance over stock. The card has a lot of power protections and doesn't scale too well with voltage. Pascal and Turing also do not have a bios editor, which is another limitation.

What do you guys think? Do we currently have less room to operate than in the past? Do you still overclock your hardware for daily use?

 

[Tír na nÓg]

In one word, bad...

OC's on life support.

Could squeeze 40 to 50% out of my Sandy bridge (4.9 to 5GHz on 2600k's), same on 3770K and 4790K, 4.2GHz+ on Phenols/Thuban.

Last CPU was a [email protected].

No desktop since, but when I look at OC results with last gen CPU's on both Intel and AMD side, I am somehow sad.

Was having a lot of fun with GPU's too. 6950 mod, 7970's up to 1250, even a 780ti to 1500MHz+. All the fun stopped with the 980's. I might have squeezed 5% max out of the 1080 and the 1080ti. Meaningless...


I guess my next build will be a Ryzen 4xxx coupled with a 3060/3070 or the AMD counterpart.

Won't be running after the extra performance with a desktop anymore.

I have fun tweaking those little amd's chips on my last laptops though. Got a good 20% extra performance on both the 2500U and the the 3700U by using liquid metal and playing with TDP and p states.

 

[Mr.Scott]

No point in OC'ing todays hardware. It's already pretty close to running on the edge.
I'm a competitive benchmarker and I'm telling you that.
Want to OC like the old days?, buy some old hardware to mess around with.
It's still fun.

EDIT- and yes, I still OC my everyday hardware, 'cause that's how I roll. lol

 

[Tír na nÓg]

No point in OC'ing todays hardware. It's already pretty close to running on the edge.
I'm a competitive benchmarker and I'm telling you that.
Want to OC like the old days?, buy some old hardware to mess around with.
It's still fun.

EDIT- and yes, I still OC my everyday hardware, 'cause that's how I roll. lol

Yes yes yes! Agreed on the old hardware! A lot of fun with 20bucks old GPU's! You can event experiment hard mod, won't be a big loss, lol!

P.S. : sorry Scotty for having left like a thief, but I really needed rehab... Only way was to go HWBot AWOL!

 

[trents]

I'm very okay with it. Yes, miss the challenge to some extent but don't miss all the time it took away from more important things. Neither is there the specter of warranties not being honored because you overclocked. Today's hardware is so powerful out of the box and so near the edge of what it's capable of anyway there just is not point in it anymore. That's why I suggested a while back that we should consider changing the name of this forum to something that would appeal to a wider audience. But my suggestion was poo pooed. What worries me more than the overclocking issue per se is what I perceive to be a lack of general activity on the forum. Just a new thread here and there these days it seems.

 

[Tír na nÓg]

^Agreed on all points.

 

[Mr.Scott]

Yes yes yes! Agreed on the old hardware! A lot of fun with 20bucks old GPU's! You can event experiment hard mod, won't be a big loss, lol!

P.S. : sorry Scotty for having left like a thief, but I really needed rehab... Only way was to go HWBot AWOL!

I totally understand.
You were not the first and won't be the last.
There were never any hard feelings.

 

[Nebulous]

The current state of overclocking is down the chitter tube. Those days of raw performance are long gone. Everything out now is 2 or 3 clicks and bam, you have your performance. It's no longer an art.

I retired from the game long ago. Sold all my old junk. Since I didn't have anyone to pass it along to, the rest I just tossed out. Hell I don't even have a back-up rig anymore. As a back up I think I may have an old Dell XPS M170 laptop with a 7800 GTX Go and a puny single core.

Yup, dem days have faded away....

 

[David]

I keep some LGA775 gear for when I have the time/space to play with it, but this won't be until August after we move house. The new place will have a study so I'll have somewhere with space to set up a benchmarking/tweaking rig. Keeps the cost down as well - if I break something it's not the end of the world.

Planning a Ryzen 4000 build this autumn but there doesn't seem to be any point overclocking it.

 

[Tír na nÓg]

I totally understand.
You were not the first and won't be the last.
There were never any hard feelings.

But I kept the Avatar!

 

[Woomack]

On ambient temps, at least for me, it looks like:
CPUs:
- AMD - is not overclocking, lucky users can reach boost clock on all cores but hard to call it overclocking
- Intel - non-K series are not overclocking what is obvious but they are about as good for OC as unlocked Ryzens, you get for what you pay and nothing more. K series OC a bit but most users are lucky to reach 100-200MHz above turbo clock.

GPUs - all are overclocking up to about 5% regardless of the brand and series

RAM - the most popular IC for overclocking for like 3 years is Samsung B. Most new and popular IC can make higher clocks but it's pretty pointless as on ambient temps, memory controllers hit a wall at 4800-5000, memory is not scaling well past ~4600 and on AMD is required too much work to make something out of IF ratio 1:2 so most stick with 3600-3800 mem clock.
Still, RAM is probably the only thing that is worth playing with as can set up to 100% higher clock. The only problem is that RAM OC doesn't give significantly higher PC performance.


Now sub-ambient OC. Well, what can I say, it's almost dead. Most hwbot members are inactive. Most good overclockers are not active anymore or work as pure marketing showing top scores on usually closed events. Most new overclockers have no idea how to OC and have no basic knowledge. This leaves maybe 50-100 overclockers in the whole ranking (count maybe 100k people who post something from time to time) who are posting good results every now and then.

As always I play a lot with RAM but recently I care more about power usage and low noise. For example, I found i5-10500 a great option for lower power PC like up to 250W that still works well in games up to 1440p. I picked a locked CPU over anything unlocked because of what I said before, unlocked chips are not overclocking anyway. Also for home, anything more than 6 cores seems a waste of money and electricity.

Old hardware still seems fun but my luck looks more like a long list of dead motherboards or various problems that end on let's say limited fun.

 

[stompah]

- AMD - is not overclocking, lucky users can reach boost clock on all cores but hard to call it overclocking

This explains my situation. Bought my 1700 hoping it being a lower tdp and lower clocked cpu meant I might get lucky and get a chip that goes past the boost clock. It overclock to boost speed and a little higher. Beyond that, it looks like it needs more voltage and better cooling than the wraith it came with. The extra speed isn't enough to justify the cost.

I always hold out hope that Intel or AMD will let one slip by the gate and we have a neutered chip that screams.

 

[trents]

"a neutered chip"?

 

[stompah]

"a neutered chip"?

A higher end chip that bins lower because of a fault. Loses a core, lower locked multi because a core can't boost fully, cache on a higher end chip clipped to fill lower end supply.

 

[trents]

A higher end chip that bins lower because of a fault. Loses a core, lower locked multi because a core can't boost fully, cache on a higher end chip clipped to fill lower end supply.

Okay, so you're talking about for instance, the old AMD Thuban's, say a 960T.

 

[stompah]

Yes.

 

[Zerileous]

With AMD that is unlikely because they have a SKU for every one of those scenarios. The exception would possibly be a 4 core part, but I doubt that would happen. The best 6 core dies are going into the 3900x, the next best the 3600x. Of course there are a lot more 3600s selling, so maybe it could happen, but now we have XTs. Sure the process is improving, GN had an early 3600 that took 1.4v to do 4.2GHz, clearly not a good idea for 24/7, and recently tested an XT that did 4.4GHz at 1.28v or so (going on memory). So that's 200mhz and 100mV improvement for $100. Of course, the 3600 is improving as well, but to a lesser extent for the most part. Even if you got a 3600XT grade core in the 3600 box you're not getting a significant improvement. Adding 200MHz to a 3600 is a less than 5% increase in clock speed, and it costs $100.

I remember when Barton was pretty mature I got an 11x 333MTs CPU and changed the FSB to 400MTs for a 367MHz increase in core clock by strapping a pound of copper ( <3 thermalright ) and a delta to it and hitting it with maybe 150mV over stock. That's a 16% increase in core clock for free (or for the cost of careful shopping and a decent HSF).

And don't even think about getting a great 3700x, the 8c parts are going into so many different SKUs: 3950x, TR, the new 8 channel TR, Epyc, not to mention 3800x. Even stuff like the 1600af which is basically a 2600, are having trouble even overclocking past their rated boost of 3.6GHz (as trents pointed out in previous threads), meaning that rather than down-badging the 2600, they're pulling chips out of the rubbish bin and selling them for bargain basement prices.

edit: this post isn't meant to be seen as bashing AMD, Ryzen is a great product and I have no problem with a company doing things that are profitable. My point is that Ryzen is not an overclocking product.

 

[trents]

With AMD that is unlikely because they have a SKU for every one of those scenarios. The exception would possibly be a 4 core part, but I doubt that would happen. The best 6 core dies are going into the 3900x, the next best the 3600x. Of course there are a lot more 3600s selling, so maybe it could happen, but now we have XTs. Sure the process is improving, GN had an early 3600 that took 1.4v to do 4.2GHz, clearly not a good idea for 24/7, and recently tested an XT that did 4.4GHz at 1.28v or so (going on memory). So that's 200mhz and 100mV improvement for $100. Of course, the 3600 is improving as well, but to a lesser extent for the most part. Even if you got a 3600XT grade core in the 3600 box you're not getting a significant improvement. Adding 200MHz to a 3600 is a less than 5% increase in clock speed, and it costs $100.

I remember when Barton was pretty mature I got an 11x 333MTs CPU and changed the FSB to 400MTs for a 367MHz increase in core clock by strapping a pound of copper ( <3 thermalright ) and a delta to it and hitting it with maybe 150mV over stock. That's a 16% increase in core clock for free (or for the cost of careful shopping and a decent HSF).

And don't even think about getting a great 3700x, the 8c parts are going into so many different SKUs: 3950x, TR, the new 8 channel TR, Epyc, not to mention 3800x. Even stuff like the 1600af which is basically a 2600, are having trouble even overclocking past their rated boost of 3.6GHz (as trents pointed out in previous threads), meaning that rather than down-badging the 2600, they're pulling chips out of the rubbish bin and selling them for bargain basement prices.

edit: this post isn't meant to be seen as bashing AMD, Ryzen is a great product and I have no problem with a company doing things that are profitable. My point is that Ryzen is not an overclocking product.

3900X is 12 cores, 24 threads.

 

[Zerileous]

yep and it's made by taking two 6c dies (really 8c dies with two cores fused off) and packaging them onto one substrate. It has been said that it's a 3600x + a 3600, but with a higher boost clock than the 3600x, I'd imagine that it's higher performing die actually bins better than most 3600x dies. The entire ryzen product line up from 4 - 64 cores, desktop to server, is based on the same 8c dies or chiplets. It's the IO die that changes across platforms to support a certain number of memory channels and PCIe lanes. A 64c part is 8x8c dies on one package. The 4c part is the same die with 4 cores fused off. The difference between the 3100 and 3300x is whether the cores that get fused off are in the same compute complex or CCXs or if the remaining four are spread across 2 of them, since each 8c die has 2 compute complexes that have to communicate through the IO die.

 

[SPL Tech]

Your numbers are wrong. You cant compare your OC with the TurboBoost OC because you're comparing OC to OC. You need to compare to the stock speed, in which it's about a 30% OC on all the chips. So it's not that overclocking has gotten worse, it's that Intel has more aggressively started pushing OEM overclocking on their products and so they are taking a lot of the skill and work out of the picture by just overclocking the chips at the factory.

However, when it comes to GPUs overclocking has gotten way worse. Nvidia cripples the BIOS settings so aggressively that you cant do jack. You can still get a VERY solid overclock on memory. 50% of the FPS gain comes from overclocking the memory on the card. However, core is much more limited due to the magical 1.09v limitation in the BIOS. This has been done to make overclocking a marketing scam. Like brand 1 has a higher stock OC than brand 2 so now they can charge $300 more for their card even though there is only like a 2 FPS difference between the two. Like, if you look at the absolute top of the line 2080Ti and compare it to the cheapest 2080Ti, the price difference is basically double, but the performance increase is maybe 10% and you'd clearly be much better served by just going SLI on the cheaper card.

What is perplexed me is that no one has found a way to crack the Nvidia encryption on their BIOSes. I mean, literally every software program out there has been cracked and can be found on the Internet with key generators and such. Not sure why the Nvidia BIOS cant be cracked.