Overclock.net banner

1 - 9 of 9 Posts

·
Registered
Joined
·
9 Posts
Discussion Starter #1
Hi all,

I was just wondering, what makes the most difference in performance, overclocking the GPU or overclocking the RAM. I have heard conflicting stories about both. Currently I have my 1080 strix SLI OC'd to +75 GPU and +500 ram. Which should I try to OC further to increase performance. By the way, my GPU won't OC past +80. Haven't really tested the limits of RAM yet.

Thanks, all suggestion and comments appreciated.
 

·
Registered
Joined
·
333 Posts
It depends on the game and the GPU

Usually I find performance increase to be somewhere between the percentage increase of each.

For example if I could add 5% to GPU and 10% VRAM I would expect the FPS increase to be somewhere in between those.

Are you gaming clock speeds anyhow? +75 doesn't mean anything these days as all these new GPUs have different base clocks
 

·
Overclocker in training
Joined
·
14,059 Posts
Hi,
Pushing both will give best results
Start with a nominal core + say +65
Then push the memory as high as it will go before memory spots start to appear +600......
Then back down until they go away and start increasing the core +10-20-25 until it become unstable and you'll find your maximum clocks.
Each benchmark is different that's why there are multiple profiles :)

I believe an average clock I've used on many benchmarks was +65 core and +666 on mem and go from there.
 

·
Desktop builder & OC
Joined
·
78 Posts
I will use "bins" which are approx 13MHz. Most cards speed up or slow down by "bin," so 12.6xxMHz (13MHz input) is the smallest increment of GPU core overclocking. It's somewhat irrelevant for VRAM as pacal can usually be overclocked by 400MHz and way above in some cases. So in MSI AB or Precision for the core my OC steps are as follows. Just example. Base shown in AB is 1,500MHz, first OC is 1513MHz, second is 1526MHz - you get the picture, but seems most folks don't know about this "speed bin steps." Not surprising as all such programs let the user input in 1MHz steps, but infact it doesn't work like that. Ever wonder why you see the same speed on core often, say 2012MHz, 2025MHz when trying OC? Those are exampes but steps are 13MHz.

So to the question. I have an evga GTX 1080ti ftw 3. In games and in 3DMark etc, the core speed is the most important. The trick is to find the right max bin your gpu core can run at if really pushing. So from years of testing, if choosing between the two it is certainly CORE speed in the vast majority of games apps that has the most impact.

But, it would be a waste to find a good stable max core, then leave VRAM at stock. So as THRASHZONE said correctly pushing both gives the best results. Don't bother with multiples of 13 for VRAM, just put in say +500MHz and increase in steps (of your choosing) until it crashes with core at stock. If already found max core speed, in theory that + your highest stable vram is your OC.
But in practice often one needs to be lowered. Pascal hates voltage. Absolute max is 1.093v. My gpu will crash if I set that max voltage. Highest is 1.081v, for me, so a small voltage increase is good/necassary with Pascal, but max TDP (Power is more useful in summary) On evga card base BIOS is set to a max of 117%. It's vBIOS hard locked. Any gpu with two BIOS try the 2nd (slave). In the case of EVGA pascal GPUs the result of just switching to slave BIOS can be profound. Starting point (before any OC) of slave has a more aggressive fan curve, and will allow power up to 127% if you do so. The most I use is 125% with heat target around 86c or 87c typed in. But, JUST switching to slave vBIOS will give noticable improvement in many games, and stress tests vary, but do show an increase.

This is a kind of free OC beyond the cards advertised maxes. Makers with dual bios don't tend to empasize switching to the slave, as it can and will push harder. (and NV don't like it. Ha Ha.) These slave BIOSes do something else too, not just max power input and fan. But I can't figure what it is. It's not voltage. Trade secret of EVGA and any other DUAL vBIOS makers. Fair enough. I don't bother with custom vBIOS because the slave is in effect a conservative custom vBIOS which allows Power TDP and Wattage above what Nvdia loosely enforce. Custom BIOS from the web can be dangerous, and remember pascal does benefit from a bit of extra voltage, but not much. So with pascal, if you can add a small voltage increase, usually good, (but some will crash). Adding custom BIOS max voltage above 1.093v will have no benefit, crash, or even damage the card (unlikely with many safeguards these days.) BUT, a few folks who won the silicon lottery can get away with it and add very big core OCs, plus, say 1,000MHz vRAM. But this is rare.

@Thrashzone notice your 65MHz is 5 x 13. Spot on for a decent 5 bin OC on a factory OCed card. Not on basic founders.

I know some of you will think *** am I going about with "bins," blah blah. Fair enough no need to use 13MHz steps as the GPU will settle at something divisible by 13 (it is actually 12.6??v) so 13 is the closest. I encourage anyone who thinks I am full of it to research this first well before adding to this thread.

But in all honesty for a casual gpu overclocker all of this bin stuff can be ignored. Those who are really into it will be interested, but probably already know. I was amazed when I found out a couple of years ago, and it explained a lot of questions and mysteries for GPU core OC. This is too long so I won't go into that, but this "bin," thing is not widely known, is not really necessary, but is interesting and very helpful for those pushing the limits.

Simple answer Both is best. Core is the more important of the two for increasing card performance. ([email protected] actually auto downclocks vRAM by 500MHz as it's not helpful with that type of simulation program.) vRAM OC on fast DDR5X memory is the icing on the cake!!

BTW everything @Thrashzone, above post wrote is correct and plenty to go on. Don't worry about bins, the card will round up or down, and it's just interesting and useful for crazy GPU OCs
 

·
9 Cans of Ravioli
Joined
·
19,889 Posts
FWIW - both my GTX 1060 and GTX 950 will seem stable with ridiculous vRAM OCs (>+1000) despite not really being stable. Anything past about +550 on my 1060 will give lower FPS in games/3DMark despite not showing any artifacts.
 

·
Desktop builder & OC
Joined
·
78 Posts
Yes, counterintuitive

FWIW - both my GTX 1060 and GTX 950 will seem stable with ridiculous vRAM OCs (>+1000) despite not really being stable. Anything past about +550 on my 1060 will give lower FPS in games/3DMark despite not showing any artifacts.
I know what you mean having experience that with a different card. Strange thing vRAM. I guess that's one reason my focus is on core, power TDP, and setting higher max temps to prevent early throttling.

In fact my 1080ti will appear fine with high (not as high as you) vRAM clocks, but - as you said!
I tend to put a max of 400MHz vRAM on 1080 and 1080ti (I have one of both). Usually just 275MHz and leave it. Just not worth it IMHO. :sneaky: { Mr. vRAM.
 

·
9 Cans of Ravioli
Joined
·
19,889 Posts
I know what you mean having experience that with a different card. Strange thing vRAM. I guess that's one reason my focus is on core, power TDP, and setting higher max temps to prevent early throttling.
Same here - forcing a custom fan profile to keep it at max boost clocks longer helps far more than a +500mhz vRAM OC.

I could have sworn that NVIDIA started using ECC vRAM with Fermi and I figured that was the reason for it being "stable" but losing performance - but my Google-Fu says that NVIDIA only uses ECC vRAM in the Quadro series so I really have no idea.
 

·
( ━☞´◔‿ゝ◔`)━ ☞
Joined
·
6,541 Posts
I have a 1080Ti FTW3 with an EK waterblock. I can have my 1080Ti run at 2087-2100MHz core without memory tweaks or 2061MHz with +500 MHz memory. Performance is pretty similar between both settings.

Honestly, I typically run stock settings since it goes to 1975-1988MHz core and my custom loop runs a bit cooler/quieter. 280W vs 330W output makes quite a difference.
 

·
Desktop builder & OC
Joined
·
78 Posts
vRAM

Same here - forcing a custom fan profile to keep it at max boost clocks longer helps far more than a +500mhz vRAM OC.

I could have sworn that NVIDIA started using ECC vRAM with Fermi and I figured that was the reason for it being "stable" but losing performance - but my Google-Fu says that NVIDIA only uses ECC vRAM in the Quadro series so I really have no idea.
Well, we fully agree.

Interesting on ECC vRAM comment. Sadly I have no answer for that.
 
1 - 9 of 9 Posts
Top