Not to get off topic, but do you run LLC7 for that extreme 5.4 clock? Any other goodies you've found to make it stable? I also don't run a constant voltage for daily use, so it can clock down and stay cool and hopefully extend the life.

Nope, I just change the voltage to 1,375 and set mesh at 5.0, and all the rest is on auto. This is not my daily PC, I retired it back in Feb when I build my Threadripper rig. It only comes out of storage for Port Royal, no kidding.

Finally joined the 15k club tonight!
 
https://www.3dmark.com/pr/676999
 
15,146
 
Jealous of you guys hitting 2,200mhz out of the box, but proud to be able to hit this with my not-so-great silicon and low clocks. #1 PR score for the 10850k/3090
 
NVVDD: 1.15v, FBVDD 1.375v, MSVDD 1.1v
 

https://www.3dmark.com/pr/679035
 
Using normal BIOS, +175/1150. Never really done this before, so don't really trust myself to go higher or move off Normal BIOS ... 

I think it's the intel chips? Not sure?
 
I ran some runs this morning at ambient, this time with GPU-Z open. Without touching classy tool, the run used only 480 watts power, and was hitting Vrel, Vop constantly. With classy, it was stuck on Pwr limit the entire time, pulling 540w on the stock LN2 BIOS.
 
Well, that makes me realize my card probably wont make much more than 15,300 without an ice bucket to keep the core temp <30, and beyond that, it needs XOC as its smashing its head off the power limits no matter how much more voltage I throw at it. 
 
This has just been a fun experiment for me, as I have never been much of a benchmarker at all. I just wanted an idea of how well my card was binned and how these KPs performed. The XOC BIOS seems a bit extreme for me, and I'm mostly a gamer, so I wouldn't get it. If there was a BIOS released down the road that was say 600w for little more headroom, I'd get that, but with the protections in place, but I'm doubtful.
 
I need to learn how to set up VF curves, PX1 doesnt make sense to me. I might be able to squeeze a little more with that, as I can undervolt or something? 
 
Oh and kris, how do you mitigate condensation when using an ice bucket with the rad? That makes me nervous!

USMC1419
That's impressive, nevertheless. I just ran some PR runs on my 5.3 setup. 
 
https://www.3dmark.com/pr/676918
 
15,302 with colder air. I kept bumping the memory speed up, ran stable at 1500+. However anything above +150 Core seems to fail at 38 seconds. I was looking at the average core clocks, and I feel it should be higher with the temps compared to other high end runs. 
 
Do you think this is bin limited (card just wont do it) or it needs more volts? Would 37-38C not allow 2160+ mhz? I see other runs at 2200 with the same temps, but I don't know what you guys are running for volts.
 
NVVDD 1.15
FBVDD 1.4
MSVDD 1.15
Level 1, On for both
 
The OLED was showing as high as 1.3V for NVVDD during the run despite setting it at 1.15 in classified. Not sure about that? Can you set it higher than 1.15 and still be safe at 35C?

Probably a combination. Better bin will be able to hit higher frequency at lower temps at a lower voltage. Mine won't do above +150 either. That being said, I think stability at higher frequency is a function of temp and voltage. Lower temps will probably get you stable at the same voltage but eventually you'll need added voltage and low temps. From what I've read Vince said not to go above 1.2 nvdd. I got a little excited on my 2080 Ti and went to 1.3 for a few runs in the 30s and now it no longer overclocks on the core for longer than about half hour before losing stability. I haven't tried to re paste it or anything so I guess it could be a hot spot or something. It literally just went from being stable one night to not being stable at any oc the next night several months after over volting so who knows. Also, the screen reports more nvdd than software from what I've seen. I haven't even touched the classified tool and iirc it was reporting 1.18 when software was reporting 1.08.

SoldierRBT
Hi, I got my 3090 KPE and it seems a good bin but unfortunately, the score I'm getting in PR doesn't match with the clockspeeds. The wattage was around 480-500W.
https://www.3dmark.com/3dm/55398081
Core locked at 1.10v 2280MHz
Memory: +1200
Max temp: 50C
NVVDD 1.10v
FBVDD 1.4v
MSVDD 1.15v
Level 1, Off for both

That's a damn good average clock speed! I cant get anything over 2,16x in PR. Your score does seem low for that clock, I'm not sure why we are seeing these anomalies. I scored 15,302 with +150 core. My memory was at 1400+ however.
 
I've heard rumblings of the clock speed reported in PR is not actually reflected realtime, might want to investigate that or compare to GPU-Z average or something. 2200+ at 48C with 1.1 NVVDD would be exceptional AFAIK

https://www.3dmark.com/pr/653446
 
15226 ... Only thing I did was flip to LN2 bios, max power in PX1 and do a 150/1100 offset. Spent maybe an hour messing with it all. I'm sure I could get more if I wanted to mess with the switches but I'm happy with that. It's for gaming after all. 

USMC1419

Stickboy46
https://www.3dmark.com/pr/653446
 
15226 ... Only thing I did was flip to LN2 bios, max power in PX1 and do a 150/1100 offset. Spent maybe an hour messing with it all. I'm sure I could get more if I wanted to mess with the switches but I'm happy with that. It's for gaming after all. 

You must have been doing something about the cooling, 33C is not normal unless you're sub-ambient or doing ice bucket? I had 5C outside air blowing into mine last night and could only manage a lowest of 37C with ML120s pushing it through.
 
Regardless, great run, nice clock speeds too.

That might have been the run I had the window open.(window is a couple feet away) I do have it setup with a push/pull setup in my case.  The stock fans are pushing and 3 ml120s are pulling. 
 
Nothing crazy like an ice bucket or anything. 

chicagohawk
It really has been bothering me to find out why low clock cards like mine and Sajin's (USMC's too, albeit you can actually hit 2200 peak) can get higher scores than those that boost 2200+ out of the box. I will say, though, yes, those that do clock higher on average also are the ones getting the ridiculous 15.6k+ scores, however, it seems the theory that higher average clocks directly correlate with higher PR scores is out the window. I hit 15.1k with an average of 2,130mhz, but Soldier is scoring 500 points less with a 2,215mhz average.
 
It'd also be interesting to know how that translates into something like game fps too. Would the card with average higher clocks or higher PR scores create better frames?

I think the problem is MSI Afterburner. I uninstalled it and used X1 with the same tweaks in classified tool. Score 15,404
https://www.3dmark.com/pr/680812
+165 Core 
+1200 Memory
Max temp: 49C
I also discovered that with OCP enable, the card would draw 450W max. Core clocks and voltage are the same. With OCP off, it's 490-500W and higher score.  
 

A few things I noticed that help inflate the PR scores for me is:
-If on a 4K display, drop to 1080p destkop resolution right before the benchmark run
-ensure MSI AF and/or PX1 are not running during the benchmark run, although I do keep GPU-Z minimized and logging to disk
-Nvidia Control Panel > Manage 3D Settings > “Texture Filtering - Quality” set to “High Performance” (that gained me almost +300 points on PR)

4.6.3 beta 2.

I bet that's some bs. Not click baiting me sir.

SoldierRBT

chicagohawk
It really has been bothering me to find out why low clock cards like mine and Sajin's (USMC's too, albeit you can actually hit 2200 peak) can get higher scores than those that boost 2200+ out of the box. I will say, though, yes, those that do clock higher on average also are the ones getting the ridiculous 15.6k+ scores, however, it seems the theory that higher average clocks directly correlate with higher PR scores is out the window. I hit 15.1k with an average of 2,130mhz, but Soldier is scoring 500 points less with a 2,215mhz average.
 
It'd also be interesting to know how that translates into something like game fps too. Would the card with average higher clocks or higher PR scores create better frames?

I think the problem is MSI Afterburner. I uninstalled it and used X1 with the same tweaks in classified tool. Score 15,404
https://www.3dmark.com/pr/680812
+165 Core 
+1200 Memory
Max temp: 49C
I also discovered that with OCP enable, the card would draw 450W max. Core clocks and voltage are the same. With OCP off, it's 490-500W and higher score.  
 

I keep OCP on, and I still get 520-530 watts no problem. This should not matter. From what another user said, this will not serve a real purpose unless you are actually on LN2.
 
I don't know what else to tweak to try and get the clock to run higher. I'm at 1.15 NVVDD and anything over 150 core wont pass. But it was like that before using classy as well. 

flyingtoaster85

johnksss
I bet that's some bs. Not click baiting me sir.

It’s just a tool video playing on the kingpin oled

It would be pure gold if someone gets Rick Astley’s Never Gonna Give You Up playing on the little LCD screen. :-D