• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Why overclock a graphics card these days?

Soldato
Joined
31 Dec 2007
Posts
13,616
Location
The TARDIS, Wakefield, UK
The boost speed advertised is the guaranteed minimum. Its not a maximum limit. Therefore most if not all Nvidia cards will boost higher due to cooling, power etc. Its been mentioned many times in this hallow forum. Its nothing new or surprising.
 
Soldato
Joined
22 Nov 2006
Posts
23,390
The boost speed advertised is the guaranteed minimum. Its not a maximum limit. Therefore most if not all Nvidia cards will boost higher due to cooling, power etc. Its been mentioned many times in this hallow forum. Its nothing new or surprising.

Yea they start throttling at quite low temps for some reason. So they benefit a lot from a proper watercooling setup. I think AMD only start to throttle at around 75c.
 
Soldato
Joined
6 Feb 2019
Posts
17,595
^^

The worst part of the throttling is I have not come across any way to adjust the temperature where it starts to throttle - it maybe be a locked feature in the bios/firmware - but changing it anyway would likely void the warranty which could be way it’s locked away from users. Turing throttling starts when the card is in the high 40s


Then Gigabyte has been having issues with its Auros cards for a while now.

There have been Auros 2080ti cards going through RMA for heat issues - some are fine, some are not due to bad contact between the die and heatsink. Gigabyte quality control must be subpar
 
Soldato
OP
Joined
5 Mar 2004
Posts
6,399
Location
Southport
The boost speed advertised is the guaranteed minimum. Its not a maximum limit. Therefore most if not all Nvidia cards will boost higher due to cooling, power etc.

I see, that makes perfect sense, cheers.

Its been mentioned many times in this hallow forum. Its nothing new or surprising.

Been away a while, it's both new and surprising to me :)
 
Soldato
Joined
9 Mar 2015
Posts
4,550
Location
Earth
With Boost 3.0, when there is overhead (temp and power) the card will boost higher. With half decent coolers the boost clock speed on the Box is usually conservative. Skyrim is likely not demanding so the GPU is not stretching its legs / there is a FPS cap in place. Metro is a fair bit harder so very much imagine its clocking up fairly high. Haven speeds can be ignored, that is not as someone else mentioned a peak clock the card hits. It is rather a theoretical step which is usually very high, higher then what most cards will actually achieve. These cards do not boost up and down in single digit numbers, but rather steps (bins).

Now in regards to manually overclocking, usually you can extract more performance depending on the card / model even under water (so removing the thermal constraint for most the part) even with boost tech (at least on Pascal) I have 5 1080Ti's here (3 FE and 2 Strix OC) and they all under water manually tend to clock to around 1900 Mhz or so out the box with no tweaks on the FE and mid 1900 on the Strix. I can manually overclock them all to around or just under the 2.1 GHz mark stable in gaming. Point here I am making is I can add around 150 Mhz - 200 MHz which is a 8% or so increase in clock speed and add another Gbps in the VRAM, and while my own sample set is not massive, I can still manage to increase performance on all the cards. This in my experience when using my HTPC with a 4k TV adds around 6%-8% extra performance for notching up a few sliders in MSI AB when playing. Not too bad really for a quick 5 min tweak IMO.

Note: someone mentioned Boost speed is the guaranteed minimum. It is NOT (unless someone can show me contrary information to Nvidia's site). The minimum Guaranteed number is the Base clock speed. The Boost clock speed however is a function of power and temps and if there is headroom on both the card will normally hit those speeds, and in reality go much higher. With a solid cooler it will usually go much higher then that as boost tapers off slowly with as temperature increases and this starts from a pretty low number in fact. Why those under water can usually maintain a higher boost by a step or two across the board. In fact Nvidia have an article mentioning Base clock is the guarantee: https://www.geforce.com/hardware/technology/gpu-boost/technology However most people will similarly also RMA a card if it cannot reasonably maintain Boost clock speed advertised, its just not guaranteed however like the Base clock number.
 
Soldato
Joined
27 Feb 2015
Posts
12,621
The boost speed advertised is the guaranteed minimum. Its not a maximum limit. Therefore most if not all Nvidia cards will boost higher due to cooling, power etc. Its been mentioned many times in this hallow forum. Its nothing new or surprising.

I know, I never said otherwise.

But its logical that these are set based on how fast the worst performance cards can perform, so if e.g. your worst 1 percentile of tested cards runs at say 1800mhz then you would set at maybe 1750mhz to not get RMA's. I dont think its very many cards that would be that bad, it would be the extreme end of it.

What I do think is wrong 100% tho is making assumptions that things like 2ghz is a easy overclock that every chip can do, and then saying if you cannot make 2.1ghz you must have a cooling problem or faulty card.

good luck trying to RMA a card on the basis you cannot overclock it to 2.1ghz without crashing.

TDP limits are spec'd to sustain the base clock speeds, the advertised turbo clock is guaranteed to be stable in the sense you wont get artifacts or crashes, anything above that is simply put "a bonus" and should never be expected.

However also remember even without any o/c done by the user, many cards will clock right up way above the advertised speeds all by themselves. So when someone says they cannot get much from overclocking it doesnt mean they leaving there card at e.g. 1800mhz because that was the advertised clock, the card may already be going up to 2ghz by itself. If a card is going up to 2ghz by itself and you can push it to say 2020mhz with an overlock, a mere 20mhz, you deem that a poor result? Would it be somehow made better if the card only went to 1800mhz by itself and then someone got the 2020 by a 220mhz overclock instead?

I never claimed my poor card is average, I know its below average so as such the average is better, but at least I am not claiming that no cards like mine exist, and I am some kind of oddity in the system.

In the case of my card tho I accept it is faulty as gigabyte OC mode is a stock feature for the card advertised, so I would be able to RMA it if I chose to. My EVGA card was also technically faulty as well.

Its a bit like those who claim every coffee lake can do 5ghz, heads stuck in the sand.

However most people will similarly also RMA a card if it cannot reasonably maintain Boost clock speed advertised, its just not guaranteed however like the Base clock number.

Yeah doesnt surprise me, but I think those should be rejected RMA's as turbo is a burst speed not sustained by design, it becomes sustainable when you jack up the power limit.

For me returning a component because you got a silicon loser is bad practice but of course we all know it goes on.

As far as your 5 samples go, you just got lucky. 5 samples and you say that every card (that isnt faulty) should be able to do the same thing? Even people with review samples have cards that cannot do 2.1ghz under water. Yes they could configure the card to run at 2.1 but it crashed.

My own auros 1080ti was almost on par with a gamersnexus sample as he published a lot of data, he had a pretty poor chip as well.
 
Last edited:
Soldato
Joined
22 Nov 2006
Posts
23,390
^^

The worst part of the throttling is I have not come across any way to adjust the temperature where it starts to throttle - it maybe be a locked feature in the bios/firmware - but changing it anyway would likely void the warranty which could be way it’s locked away from users. Turing throttling starts when the card is in the high 40s


Then Gigabyte has been having issues with its Auros cards for a while now.

There have been Auros 2080ti cards going through RMA for heat issues - some are fine, some are not due to bad contact between the die and heatsink. Gigabyte quality control must be subpar

I've had a few geforce cards that weren't stable on "boost clocks", so not a fan or how they do things tbh. If you up the power (or lower it) it also alters the boost clock. So you limited on what you can fiddle with without messing with the BIOS.

On my AMD card I can alter all the clock states and power independently using the official driver tools, which is much more helpful.
 
Last edited:
Soldato
Joined
13 Jun 2009
Posts
6,847
Overclocking has been dead for a while now, in both the CPU and GPU space. There are a few reasons:
  • [GPU/CPU] Manufacturers have figured out that adjusting the chips' clock rate based on dynamic temperature/voltage/current readings rather than picking a safe constant value basically gets them "free" performance with the same silicon and improves yields;
  • [GPU/CPU] Manufacturers have figured out they can make more cash by segmenting the market differently (e.g. only top CPU SKUs being unlocked, or differentiating GPUs by shader count instead of clock speed);
  • [CPU] A severe slowdown in architectural improvements over the last decade or so has necessitated pushing process nodes to their limits each generation instead.
For example, a GPU might only be able to run at 1.5 GHz over several minutes due to heat output, but if it could easily run at 1.8 GHz for a short period when needed, why not do so?

Gone are the days when you could whack a 2.66 GHz i7-920 up to 4.0 GHz on air cooling, sadly.
 
Soldato
Joined
9 Mar 2015
Posts
4,550
Location
Earth
As far as your 5 samples go, you just got lucky. 5 samples and you say that every card (that isnt faulty) should be able to do the same thing? Even people with review samples have cards that cannot do 2.1ghz under water. Yes they could configure the card to run at 2.1 but it crashed.

My own auros 1080ti was almost on par with a gamersnexus sample as he published a lot of data, he had a pretty poor chip as well.

When I mean under 2.1 Ghz at or around, that means some of my cards actually do come in under the 2.1 GHz mark under water, so actually trying to say its still a mixed bag under water and there is variation as expected.

Never once said

As far as your 5 samples go, you just got lucky. 5 samples and you say that every card (that isnt faulty) should be able to do the same thing?

A card not able to do 2.1 Ghz is faulty. Infact that would be contrary to my own position would it not of highlighting only the base clock is the guaranteed speed (and AFAIK there are no 2.1 GHz guaranteed base clock ;) ). No what I was pointing out is there is some variation, even between similar models under water and mine come in under the 2.1 GHz mentioned (maby should have been more clear there when I said the around bit).

I expect most GPUs in actual use with solid coolers (after market that keep temps in check and below thermal target when clock's drop rapidly) will maintain advertised boost clock speeds and usually go higher. I expect most 1080Ti's will clock to around 1950 - 2050 Mhz with some samples either side of that. EDIT Here: when I say clock to 1950 - 2050 I mean manually, not out the box.
 
Last edited:
Soldato
Joined
28 Oct 2005
Posts
9,147
Location
Scotland
OP you can also overclock the memory sometimes quite a bit higher.

This is one of the things I was going to call out...I've been very conservative with my RAM clock and was only running +150 on my 2080 but +700 is what I have it set at now and it's absolutely fine. Will need to test higher.

It definitely does seem these days the boost can be sufficient on the core especially when you use the Auto Tune software available. Temps are definitely the most important thing to keep that boost as high as possible though.
 
Soldato
OP
Joined
5 Mar 2004
Posts
6,399
Location
Southport
So when someone says they cannot get much from overclocking it doesnt mean they leaving there card at e.g. 1800mhz because that was the advertised clock, the card may already be going up to 2ghz by itself. If a card is going up to 2ghz by itself and you can push it to say 2020mhz with an overlock, a mere 20mhz, you deem that a poor result? Would it be somehow made better if the card only went to 1800mhz by itself and then someone got the 2020 by a 220mhz overclock instead?

Certainly not but I guess the point is that by overclocking thats an extra 20mhz. The rest the card was doing without any intervention so in your example, is it worth messing to get an extra 20mhz? Personally I wouldn't bother but obviously some feel the need to squeeze out every last drop :)

Would it be somehow made better if the card only went to 1800mhz by itself and then someone got the 2020 by a 220mhz overclock instead?

It would certainly make it feel like it was worth bothering with yes.

Memory OC I've not yet messed with, thoughts on that? what kind of performance increase have people seen?
 
Soldato
Joined
19 May 2004
Posts
3,846
Power mods on my mobile 1080s with cooling tweaks took me from 1700-1800mhz boost clocks to 1950-2000mhz solid during gaming. Lows are more consistent and overall performance is up.
 
Soldato
Joined
9 Mar 2015
Posts
4,550
Location
Earth
Certainly not but I guess the point is that by overclocking thats an extra 20mhz. The rest the card was doing without any intervention so in your example, is it worth messing to get an extra 20mhz? Personally I wouldn't bother but obviously some feel the need to squeeze out every last drop :)

It would certainly make it feel like it was worth bothering with yes.

Memory OC I've not yet messed with, thoughts on that? what kind of performance increase have people seen?

Personally I have only seen a very small increase from the memory OC at 4k, but would try messing around with it. One comment however would be to be wary of pushing it too far. While you can actually clock the memory without seeing artefacts and it may all seem okay on the surface, error correction can kick in the background so there may be an actual performance drop which may only be apparent if your comparing run to run.
 
Soldato
Joined
27 Feb 2015
Posts
12,621
Certainly not but I guess the point is that by overclocking thats an extra 20mhz. The rest the card was doing without any intervention so in your example, is it worth messing to get an extra 20mhz? Personally I wouldn't bother but obviously some feel the need to squeeze out every last drop :)



It would certainly make it feel like it was worth bothering with yes.

Memory OC I've not yet messed with, thoughts on that? what kind of performance increase have people seen?

I dont think there is much performance to be gained, but it seems easier to overclock, so I do it anyway. But also if you overdo it then you actually lose performance as its ECC, and will correct errors (With performance penalty). Probably low single digit performance gains, with variance from game to game.
 
Associate
Joined
9 Apr 2017
Posts
188
Location
Eve Online
Overclocking has been dead for a while now, in both the CPU and GPU space. There are a few reasons:
  • [GPU/CPU] Manufacturers have figured out that adjusting the chips' clock rate based on dynamic temperature/voltage/current readings rather than picking a safe constant value basically gets them "free" performance with the same silicon and improves yields;
  • [GPU/CPU] Manufacturers have figured out they can make more cash by segmenting the market differently (e.g. only top CPU SKUs being unlocked, or differentiating GPUs by shader count instead of clock speed);
  • [CPU] A severe slowdown in architectural improvements over the last decade or so has necessitated pushing process nodes to their limits each generation instead.
For example, a GPU might only be able to run at 1.5 GHz over several minutes due to heat output, but if it could easily run at 1.8 GHz for a short period when needed, why not do so?

Gone are the days when you could whack a 2.66 GHz i7-920 up to 4.0 GHz on air cooling, sadly.

While I agree with you that manufacturers have looked closely at how auto-overclocking of their products can provide inbuilt gains, I don't agree with the comparison to the i7-920, and this is speaking as someone who had one (and also had the Pentium 4 Northwoods that overclocked like crazy). At present I'm running a 7980XE that I keep on 4.7Ghz all-core 24/7. The stock all-core Turbo is a matter of some debate on sites but I was getting 3.3Ghz all core at stock.

i7-920: 2.66 -> 4.0 Ghz 50% overclock
7980XE: 3.3 -> 4.7 Ghz 42% overclock

So technically, yes, the i7-920 had a slightly higher margin, although if I turned up the chiller I could probably get close to that overclock delta.

And on the graphics front, an RTX Titan stock does 14500 Timespy, and my 24/7 clocks will do 17400. That 20% is up there with the better clocking cards I've had over the past two decades (although short of the likes of the 9700Pro).

What has changed is that more mid-range hardware is artificially locked down. The high end has roughly the same margins as ever.
 
Soldato
Joined
6 Feb 2019
Posts
17,595
Not sure about AMD, but on Nvidia GPUs overclocking in 2019 is pressing 2 buttons for free performance, why not do it

Anyone who says "overclocking is dead" should explain what they mean by this because several examples have been posted here showing overclocking stills gets you significant performance gains in 2019
 
Back
Top Bottom