emmett
I got just a tad more than my last score earlier today. I guess I am gonna have to just break down and try the XOC because at the beginning
of the test I am hitting 530-540 (GPUZ) power limit. I can see an immediate 1-2 FPS throttle. I will say that cooling the back of the card seems pretty important
and that using fans or heat sinks may not do the trick.
 
15801
https://www.3dmark.com/pr/819482
 

This is true, but most want to try to do it out of the box, on stock everything and hit 15.6K. No sure how that is even possible. the best I got was 15,280K on 450W stock vbios. And you have me beat on 520W vbios @15,662

Guys what's going nvidia and windows settings are you all using to get your best scores?

I had this long drawn out post, but not really worth it to explain.
 
https://www.3dmark.com/compare/pr/820304/pr/681933/pr/819320#
Bottom line I see a few things that are not the same as you seem to think they are.
1: steam version
2: 1412 on memory while your at 1375
3: older driver while your on newer one.
4: his average is higher than yours at 2181 mhz while you are 2177
5: you don't know what bios he is using.
6: you don't know if he is using classified or not.
7: you don't know if he is benching with open window or ac cooling.
8: he has an asus you have a gigabyte.
9: you dont know what his nvidia settings are.
10: you are running the same os and cpu speed as him. that seems to be about the only thing you have in common.
 
1: you need to reset your nv settings back to factory.
2: you need to drop the curve business for now.
3: you need to start by seeing how high your core can go.
4: you need to use the 520W version and quit slow dragging through the versions.
5: you need to use the classified tool to set voltage to 1.1V nvvdd and leave everything else alone.
6: you need to start at about 135+ on core and leave memory alone for now. work on one then the other. you pass, add +15 and run it again. don't worry about waiting. open gpuz and drop it to task manager. i use px1 for locking the clocks with boost lock
7: you will have crappy scores at first because you are just trying to see what the core will do vs the memory.
8: when you crash add .050mv to core and run the same clocks again. you are trying to see how high you can go before you max out at 540W as that's about where it stops for 520W vbios.
9: as was stated before. the count goes by 15's. so 105/120/135/150/165/180/195/210/225/240/255
10: as also stated before, you might crash because depending on which bin you are in your clocks will change. this is why on some days the same clocks will crash because they are running higher mhz at that setting. example +240 can go as high as 2295 if the temps are pretty low while on hot days some thing like 2265 or a bin lower.
11: can you break 15k with just core? yes you can.
once you get as high as you feel comfortable with, then start with memory. 1000 is a nice place to start as you don't need to move memory voltage. and go up in +100 increments till you crash. core is also at stock at this point. to get a 1400 mhz memory you need to be like +1500 to +1600 and if your room is a furnace, it's not gonna happen.
12: if your room and back of your card is like a furnace you can forget about getting any decent scores. just not going to happen. scores will be super crappy as you are looking for how high your memory overclocks for this time of day.
13: best to bench during the night with a window open and fan pointed towards your case or card.
14: learn the time, frames and fps points in the bench. if it isn't at a certain point no sense in letting it run till the end. you'll just be mad is all and wasted 2 minutes for nothing. stop bench and make a minor adjustment and run again.
14: never assume anyone with really good numbers is just hitting the go button and just miraculously has super great numbers. it's just not possible.
15: when you get your clocks set for a real good go at your best, then switch settings to performance mode in nv control panel and close out. and run bench.
16: best drivers are usually older drivers. 457.xx seems to do really well, although a few have been able to do with some newer drivers. not sure about the newest driver.
 
You can use that as a guide.
 Edit:
And last but not least.....You card is not a dud if you are 1 fps off the leaders. people get so invested in the score that they do not realize that it is like 1 to 2 fps in the grand scheme of things. if you were like 20 fps behind, then you can start with the rma campaign.
 

emmett
Great guide right there. Follow it and you will do well.
 
And breaking 15K with core like jonksss said is no issue. My best with a little classified tool prodding is 15402  +0 mem. Work with core first.

How much voltage are you pushing to achieve that and what core clock?  I just swapped fans on my radiator for the card, the stock fans on it are actually horrible, garbage stock pressure with giant gaps between the blades - now running EK Vardar EVO RGB ones, we'll see how that goes, gonna do some regular gaming on it shortly since I'm taking the day off from OCing.

johnksss
I had this long drawn out post, but not really worth it to explain.
 
https://www.3dmark.com/compare/pr/820304/pr/681933/pr/819320#
Bottom line I see a few things that are not the same as you seem to think they are.
1: steam version
2: 1412 on memory while your at 1375
3: older driver while your on newer one.
4: his average is higher than yours at 2181 mhz while you are 2177
5: you don't know what bios he is using.
6: you don't know if he is using classified or not.
7: you don't know if he is benching with open window or ac cooling.
8: he has an asus you have a gigabyte.
9: you dont know what his nvidia settings are.
10: you are running the same os and cpu speed as him. that seems to be about the only thing you have in common.
 
1: you need to reset your nv settings back to factory.
2: you need to drop the curve business for now.
3: you need to start by seeing how high your core can go.
4: you need to use the 520W version and quit slow dragging through the versions.
5: you need to use the classified tool to set voltage to 1.1V nvvdd and leave everything else alone.
6: you need to start at about 135+ on core and leave memory alone for now. work on one then the other. you pass, add +15 and run it again. don't worry about waiting. open gpuz and drop it to task manager. i use px1 for locking the clocks with boost lock
7: you will have crappy scores at first because you are just trying to see what the core will do vs the memory.
8: when you crash add .050mv to core and run the same clocks again. you are trying to see how high you can go before you max out at 540W as that's about where it stops for 520W vbios.
9: as was stated before. the count goes by 15's. so 105/120/135/150/165/180/195/210/225/240/255
10: as also stated before, you might crash because depending on which bin you are in your clocks will change. this is why on some days the same clocks will crash because they are running higher mhz at that setting. example +240 can go as high as 2295 if the temps are pretty low while on hot days some thing like 2265 or a bin lower.
11: can you break 15k with just core? yes you can.
once you get as high as you feel comfortable with, then start with memory. 1000 is a nice place to start as you don't need to move memory voltage. and go up in +100 increments till you crash. core is also at stock at this point. to get a 1400 mhz memory you need to be like +1500 to +1600 and if your room is a furnace, it's not gonna happen.
12: if your room and back of your card is like a furnace you can forget about getting any decent scores. just not going to happen. scores will be super crappy as you are looking for how high your memory overclocks for this time of day.
13: best to bench during the night with a window open and fan pointed towards your case or card.
14: learn the time, frames and fps points in the bench. if it isn't at a certain point no sense in letting it run till the end. you'll just be mad is all and wasted 2 minutes for nothing. stop bench and make a minor adjustment and run again.
14: never assume anyone with really good numbers is just hitting the go button and just miraculously has super great numbers. it's just not possible.
15: when you get your clocks set for a real good go at your best, then switch settings to performance mode in nv control panel and close out. and run bench.
16: best drivers are usually older drivers. 457.xx seems to do really well, although a few have been able to do with some newer drivers. not sure about the newest driver.
 
You can use that as a guide.
 Edit:
And last but not least.....You card is not a dud if you are 1 fps off the leaders. people get so invested in the score that they do not realize that it is like 1 to 2 fps in the grand scheme of things. if you were like 20 fps behind, then you can start with the rma campaign.
 

1: not using steam, pretty sure I stated that already
2. can't go any higher on vRAM, I tried, would likely have to increase voltage in Classified Tool, will probably try that later
3. I ran on 460.89 previously and my scores went up with the newer drivers, planning on using 457.51 as a lot of people seem to get good scores with those, for the next time I'm pushing for personal bests
4. 4MHz... seriously...
5 6 7 8 don't actually matter that much, if he's not using the 520W BIOS I'd be shocked, obviously it's not an open window because our average temps are the same, motherboard I probably have an advantage because of the memory layout being superior on the Gigabyte
9 - Unless I missed some recent discovery, my NVidia Control Panel settings are pretty "optimized" for benchmarking, with High Performance being set in filter etc.
10 that's literally why I chose THAT person to point out how their score is 400pts higher than mine for seemingly no reason... Glad you noticed.
 
2 (dunno why numbers started over, but swa) - As someone that's actually being helpful pointed out in PMs, the curve can actually gain you clock stability instead of it varying so much, but I am trying both with & without a curve set, obviously
3 Such help much wow
4: ??? You missed the screenshot with power draw a few posts before the score that is in the compare, that's fine - I am on the 520W BIOS.
5: Been there, done that
6: I started at where I started, 150 seems to be about the limit I could hit before increasing voltages in Classified Tool - and so far bumping the ones up that effect core voltage have no resulted in an increase in set-able clock, but have resulted in higher average clock & of course, higher power draw
7: I've done core first, then memory without core OCed, to see where each tops out, vRAM will not run much higher than 1250 without failing, score went up with 500, 1000 and 1250, then crashed at 1500 & 1350, so 1250 is where it is.
8: This is the part where I stopped once I got to 1.1V NVVDD because I was unsure how high people were going so I started looking over the thread, saw Sajin running 1.2V (assumably with higher than normal cooling) & figured I'd ask what others are running
9: Super aware, 15MHz increments has been the norm for a few generations now, much easier to calculate in the brain-parts than the increments on my old 1080 Ti was.
10: Aware I am
11: Can other people? Yes, can this system? Not yet, not even close, haven't gotten higher than 14.6k with just core tuning, yet
12: Temps are definitely not a problem:

Right side is max load temps in PR and that's with nothing special being done case-wise (no side off or whatnot)
13: I'll be letting the room get cool tomorrow when I resume pushing, but that's the most I'm doing, people can go crazy all they like with running in super-cool environmentals, I sleep in the same room, I'm not doin' that lol
14x2 Yes
15: This is exactly a problem I was having yesterday that made me get irritated & stop, I had clocks that would pass with Afterburner & GPUz open to monitor things, but once I closed them, for whatever reason, the benchmark would not pass again.  No idea what was going on there, but that happening repeatedly is what made me throw in the towel for the day... and that's with letting it cool down between runs
16: Yes as I said in an earlier number, plan on using 457.51s in the next runs, just to see how they do, but I'm also going to test with the 461.40s because like I said in an earlier post, they resulted in a stock-speed score increase of ~133pts from previous drivers I was using, which is well outside run-to-run variance for my system. 
 
And I know the card is definitely not a dud, I do think my vRAM is a bit weak though, nothing a lil' voltage bump might not fix, but it seems like +1250MHz on the memory is definitely sub-par.
 

bumfoto
https://www.3dmark.com/pr/823587
 
new score 16054
 
not ln2

 
I was about to say "Wow that's insane" then I saw your average temp & it made more sense, sick score either way though!

johnksss

Dabadger84
 

Now that that part is out of the way.
1: 1300 memory, try moving mem voltage to 1.400V. I would not recommend going above 1.425V
2: move nvvdd to 1.150V or even 1.200V. adding voltage is going to start raising the watts. And you may think he is on 520w but he could be on 1000w bios. you would need to ask him. at this point we are speculating and not actually talking facts.
3: run your highest clocks that crashed and see if it completes.
4: if it does not, then change nvvdd ocp to disabled and run it again. you will know if your over because your card will flash red for like a millisecond and show you what voltage is the problem, each time it hits it.
5: get a time of the exact crash. pretty sure it's not enough voltage at this point. since you are nowhere near using too much voltage.
6: these cards are built with pretty highed parts so don't be scared to hit 1.2V if need be. this is benching not everyday clocks.
7: yes, i have used the curve, but it cuts like 10 percent of the max power in my many test. so i never use it for benching.(impo)
8: max fans unless your name is Luumi who seems to do excellent on auto fans and +1600 memory. LOL
 
hummm, open px1 if you installed it. click the boost lock then exit out of it. and try what you were doing.
you really only need to run gpuz at this point as its enough to give you most of your answers that matter. (impo)
 
And here is a big hint as you start adjusting your clocks. if you don't start at around 64 fps or higher, might as well not even run the bench. starting at 67 fps is like 15.4 to 15.5k. 68 is like 15.6 15.7. 69 is like 15.7 to 15.8 or so....
I was at 15k last night, but in a 85F room and no open windows and using the exact same clocks i now get this.
(right click open in new tab to see full screen)

 
 
 

How happy I am to read from you johnksss,
Alienware overclocking times and notebookreview's forum seems so far away now...
Thanks for sharing your wisdom, I am having the same issues as fellow Dabadger84 with my Kingpin. But I am 99% sure it is temperature related. Even if my card is not strong enough as many of yours.
I will try to follow your tips so the 15K points barrier gets broken :P

Best,
Nikos
 

Just to clarify, I'm sure this is the case but thought I'd ask: Classified Tool settings don't stick after reboot, right?

So, this is with the newest drivers, CPU is at STOCK, RAM is at STOCK, GPU is the only thing messed with at all, no Classified Tool at all, 520W BIOS, GPU is set at +150 core, +1250 Mem, core voltage slider is at +100 though it seems like that actually does absolutely nothing, power target of course is maxed:
 

 
And that's with the newest drivers, all other background processes closed, just Afterburner & GPUz running in the background.
 
Is this lower than it should be, or not that bad?  Regardless of what your CPU is, if you could post a run on daily-use system settings with your GPU similarly clocked, so I can see some comparisons, that would be awesome.  Thanks for the help y'all.

Ran it again with GPUz closed, GPUz doesn't have a huge impact on score, but it did bump it just a tiny bit:
 

 
Looks like it was at 2175-2190 through 95% of the test, ambient is still a balmy 68F - room is slowly cooling off, turned my heater in the room off about an hour ago.  That bumps me up to 6th for my setup, passing the guy who's score I was questioning why I wasn't matching, by 4 points, and matching our very own markuaw1 (who has a STRIX card with the KINGPIN bios on it if I'm not mistaken) - and he was running sub-ambient/cold cooling not room temp, average temp on his run was 31C vs mine was 46C - I'm quite happy with that!
 
Guess it's high time I try a voltage bump eh?
 
Edit: testing with a bump, first result was a slight increase but also only a +15 increase on core: https://www.3dmark.com/3dm/57482788
 
Edit 2: Seems like I had some heat soak in the PCB so I called it for now, couldn't get +180 to pass even with 1.15V on the core which seems silly to me and it was crashing very early in the test.  Much happier now that I've cleared 15k easily - ended up getting 15410 with a re-run of the +150 settings, without GPUz closed, so I probably could get a tick higher, but that 15461 with +165 is my new personal best & good for 5th for my setup, only people that have higher have 5.3 or 5.4 or 5.5GHz on their 9900K & 2250+ on their core with average temps in the 15-33C range so much cooler than mine as well.  My ambient for those tests was still around 68F.
Maybe another day I'll let it get chilly in here but I doubt it. lol Next time I fiddle with it I'll try pushing the core to 1.2V & see what I can get clocks wise.

johnksss

Dabadger84
Just to clarify, I'm sure this is the case but thought I'd ask: Classified Tool settings don't stick after reboot, right?

They stick after a reboot, but not if your remove power or do a complete shut down or hard crash.
 
 
The dip switches are so you don't need to use classy tool. Some say they help while others say they don't. I never use it.
 
Congrats on the new personal best!!
 

Yeah I saw zero change with the dip switches. The ones I did are supposed to bump core voltage but seem to have done nothing. Perhaps they only work properly with the 1K bios.
And the classified settings stick? Interesting. I always set it back to standard before rebooting out of my extra-tight OC benchmark settings on the RAM. Good thing I guess. Of course I only did a small bump on the core voltage so not anything spectacular, 1.11 nvvdd & 1.0875 on the other one, I think. Going up on the memory voltage didn't help at all either, I can't get 1500 of 1350 to pass, it crashes out pretty quickly.
I haven't had a single hard crash yet. I have had to log out and back in to windows because I couldn't get a crashed/frozen PR screen to close though. Closest I've had to a hard crash.
I'm pretty happy with the results thus far considering every score above me on the 9900k is 15-33C average temp and 5.3-5.5GHz on their cpu. Hoping I can push up to at least 2250 once I give it some voltage.

I still have no idea what I did to magically make my score increase 400pts. I always ddu between driver swaps after uninstalling the ones I'm removing. But the 461.40s are definitely pretty good. Kinda curious if I can find a driver that performs even better. I'll futz with that next time.

johnksss
I see what you mean. I loaded up the 520W vbios and now am crashing on memory around 1300. I ran your settings and got a 15375, but tried to go higher and crash city or no complete. It seems like a whole relearn when you switch the bios switch. That was with 1.1V nvvdd only adjustment.
I think you may have mentioned it before...your memory timing adjustment. I think that's where you got your extra 400 points. First run on your clocks netted a 15048, but after adjusting the memory.... 15375. I'm using the same driver as well.
 
Edit:
Using my 10900K and not 4930K setup.
5.3Ghz/4400 Mhz memory.

The memory settings really aren't too awfully special, 4000MHz at 1.43V, CL17-17-17-38 2T with tRFC at 300, tREFI of 22500 (haven't really messed with that much), everything else is just what the BIOS or sticks read out at at stock, manually inputted so the BIOS doesn't do anything stupid like putting them tighter or extremely loose.  Don't think I could go much higher on speed even with looser timings.  I actually used that as daily settings for a while with 1.42V and tRFC of 350, but it gave me a crash or two in Cyberpunk so I reset the RAM to stock for my daily-usage settings, 3600MHz @ 1.35V CL16 isn't too bad. 
 
And those are the same RAM settings I ran before when I was only getting 15037 - switching drivers twice then reverting back to the latest just magically gave me almost 400pts and I have no clue why, but I approve.  I'm sure the cooler temps were a slight increase, but no way it was that much.  Maybe DDU caught something it didn't before and cleaning that up made the difference.
 
With what you experienced switching to the 520W BIOS, maybe it is indeed a memory timing thing or something preventing higher vRAM/memory clocks.  But even on the "Normal" BIOS switch position, I was able to run +150/+1250, couldn't go any higher without crashing, but my score was also only around 14.9k... but that was also with 72F ambient so who knows.
 
It definitely seems like I got some meh-worthy Memory on my card though - I'm going to try pumping it up to 1.4V on the vRAM next time I test, see if maybe that will let me run +1300-1350-1500 or thereabouts... with the fact that it crashes at +1300 but passes at +1250 every time, I have my doubts.  What is the stock FBVDD setting anyway, 1.36875 or 1.375?

I tried 461.40 to see if it was any better for me, I lose 50-70 points tested over several runs.
so then I reloaded 457.51 back on, and just for LOL's wanted to see where I topped out with just power-temp limit raised No classy tool. 
I just kept raising and locking 15 MHZ bins on Freq-Curve.
 
I think I must have glitched something?
 
https://www.3dmark.com/pr/827109
 
 
 

15438  
https://www.3dmark.com/3dm/57497662? 
 
Need to run again tonight with the windows open! A balmy -11C here today.
 
 
 
 

emmett
I tried 461.40 to see if it was any better for me, I lose 50-70 points tested over several runs.
so then I reloaded 457.51 back on, and just for LOL's wanted to see where I topped out with just power-temp limit raised No classy tool. 
I just kept raising and locking 15 MHZ bins on Freq-Curve.
 
I think I must have glitched something?
 
https://www.3dmark.com/pr/827109

This is exactly what happened to me, then I reverted to the 461.40s & got my high scores.  It was extremely weird.
 
So after feeling the exhaust air from my GPU radiator during a PR run, I decided I should indeed go Push/Pull on the radiator... unfortunately I'm plum out of EK Vardars (have one more, need 3 so lol) so the "Pull" fans are 3 Thermaltake 120mm Pure Plus I had laying around from the last time I ran a 120mm-variant radiator.
 
We'll see how those effect temps, going to have them running at 60% for starters.  Already idling at 23C while ambient is at 72F, which is usually resulting in 24C idle... maybe it will help, maybe I just wasted 30 minutes getting those fans in up there. lol

emmett
I tried 461.40 to see if it was any better for me, I lose 50-70 points tested over several runs.
so then I reloaded 457.51 back on, and just for LOL's wanted to see where I topped out with just power-temp limit raised No classy tool. 
I just kept raising and locking 15 MHZ bins on Freq-Curve.
 
I think I must have glitched something?
 
https://www.3dmark.com/pr/827109
 
 
 

Internal clocks drop a lot when you’re messing with the v/f curve. The NVVDD/MSVVDD table isn’t providing enough voltage that’s why the low score. You can check this with Thermspy or HWINFO64 (effective clocks).

Has anyone been able to reach 15K below 1.0v?