maybe it is the game bugs that caused it to crash? cannot rule out that possibility i think

because that method is having one point stable but the rest all running lower than stock. of course will be more stable but slower than stock lo. Let’s say light game like csgo where the gpu will load around 70% only. you will lose alot of frames.

yesterday I asked you to provide some screen of both curves and the logs from gpu-z so I can tell exactly where (and possibly why) it is crashing.
post this in the curve oc thread and we can discuss.

but if you don't all I can say other than what Bonchi/I told you is that if based on your feeling the cut off point is more stable using another method (and don't mind the loss of performance when clocking down) just do so lah, no one force you.

Oh n yimingwuzere afaik that error correction mostly for ram though 
But I left that part out on purpose, there's vast differences in cooling of the vram and as long as nvidia doesn't permit the sensors to be read out by consumers there can't be one solution for all cards.

Post#9439, probably you didn't read as I did not tag you.

If you like your way more that's fine, the difference in most cases is not substantial. The higher your cut off point the more you will be able to tell a difference though.

To be fair I just briefly tested the way you followed once on Turing and results back then confirmed what I said here.
I can test again with my 3080 if I can reach higher clocks (on low voltage points) by setting the curve in a different manner, but I basically just wanted some data first.

Don't worry, if you're happy where you are now just leave it at that.
If you wanna spend more time to tweak you can just take the offer on posting said stuff in the curve oc thread so we can follow up on it n see if there is a 'better' way to do it n others can benefit from the discussion as well.

well to put in perspective. the youtube guide is dragging everything down a couple of hundreds then push one point up to 1920mhz (as example) so when the gpu is on lower utilization, it will be like 100mhz off from stock. meanwhile nrw guide is dragging everything up, overclocking those lower voltage a couple of hundred more.

from my own test..following those youtube (cuz i turn on vsync) my gpu core is only 900mhz @60%utilization...... whilst nrw guide, it’s 1300mhz. The difference is not small.

On the contrary, the peak you get stable with the single point is 1920mhz but nrw guide is only stable at 1905mhz for example... that 15mhz difference will be less than 1fps difference

the youtube guide works well with benchmark on that targets only at one single point but everyday use it will suffer... along with overall effeciency as well which nrw explained.

or u can try my method i mentioned before, u just reset the curve to default, find the max point u want and drag that point there. then just make all points to the right flat. this method only oc the max point and the few points before it, so less instability at lower voltages compared to the curve method, and no loss in performance using the single point method where u offset all the lower voltages to lower freq

yeah cuz cp2077 is maxing out. but if you go launch lighter or older games like dota, csgo, diablo3 etc with much lower graphic intensity, then you will start to notice. especially stutters with the youtube method.

All in all, these are just a way to maximize your undervolt and especially to prevent frame stutters caused by a huge spike on the curve. So id personally advise to try your best to smooth out the curve near the max regardless of which method you use.

no u reset to default curve first, drag only the 0.875v point to 1920, then u see the points at the right side still higher than 1920 u just make that whole part go down using the shift highlight so when u apply the entire part to the right of 0.875v is also 1920.
for example here is my default curve

i want my 1800mhz@0.85v, i first drag my 0.85v to 1800

then i click apply

now u see some points to the right of 0.85v still above 1800, so i just shift highlight and drag that whole part down

and click apply, then the part will be flat


do this and compare to the original one point method, u can see the clocks on lower voltages will be diff. that method decreased them compared to stock curve, which is pointless because the stock curve is technically supposed to be 100% stable d

1) Confirm that you have sufficient (direct contact cooling) - if you have a 3090 be careful as the temps between front n backside of the pcb is day n night.
2) Keep adding 100 and bench every time until you stop gaining performance (or even lose as error correction kicks in)
3) Go back to the last 'stable' (without performance stagnation or loss) point and take off another 50-100 for safety as other applications might stress it differently.

Don't overdo it in general... as I said ... if Nvidia wanted you to, you'd be able to see the temperature reporting of the 8/12/24 sensors.
Curve OC is pretty safe but I wouldn't say the same thing about VRAM OC. It will clock down at a defined temp but I doubt it's good for long term if its running right below the max.

Personally despite the chips actually being produced for 21GBit/s and the card being on water and +1300 being stable (+1250 on air) ... on my 3080 I just use +500 as a casual setting and even on my performance setting I 'just' went for +1000.

reminder though, this is still technically worse than just moving the entire curve. that one has higher risk for instability but potentially less stutters. just pick ur tradeoff

Guys, just updated 3080 to latest driver. Idle temperature around 28~29c. Are you having the same experience? I was like WOW lol.

well it’s silicon lottery. but youre basically running FE clocks at such temps is damn good.

yeap that’s correct. so if it goes over 70C it will drop down to 1905.

I thought the tuf should be super low temps. are you on the silent bios?

ahhh as i expected. Asus fan sucks lol... I had the same experience as the 2080 strix... their fans will rape your ears the moment it goes above 60% fan speed. or 1.8k rpm

read about asus tuf has issues with manual fan control at the moment as there’s many complains online as well. just wait for a bios fix i guess. But yeah. asus fans are generally noisy. I would deshroud it but it voids warranty worst is their gpu fan headers are some proprietary 7pin sommore.