EVGA 3080Ti FTW3 overclocking on water

[Voodoo Rufus]

I finally got around to playing with my 3080Ti FTW3 with a Heatkiller V block mounted up. I've never had a card that OCs like this in terms of technique. The memory never seems to crash out or artifact at all, it just slows down performance and power draw. I haven't found the sweet spot there yet as a result. I went all the way to +2200 on the mem supposedly stable, but it needs to be +1000 or less I think. My memory shenanigans seem to result in huge swings in Speedway benchmark performance.

At least on Heaven, I cannot get the board to pull more than 400W. On Furmark, I can get it up to 430W, but it's still less than the board max of 450W.

On the core clocks, I can go right up to +175 / 2145Mhz average. +180 / 2160 crashes out no matter what I do. My GPU temps are right at 50-53C, and water temps at 34C in an ambient of about 22C, so I have tons of thermal headroom. Seems like it's begging for more voltage but there's none to be had.

3D Mark is giving me 9200 in Time Spy Extreme and 14k in Port Royal. Speedway is 4500-5400 but like I said I need to nail down my memory speeds.

Using OC Scan came up with +82 core and +200 mem. Seems very conservative.

Any hints on what I could do extra to try to get to a nice round 2200MHz? At another forum they told me to be happy with my 2145.

Anyway, one attachment for fun running ancient benchmarks to compare to older builds. I might have to gun for that 50k in '06. Feels like it's CPU bound in a big way since the GPU can't even get warmed up on it.

 

[Voodoo Rufus]

Played some more tonight dialing in the memory first then pushing the core clocks for benchmarking. Did memory at +500, 1000, 1500, 1650 then back down a bit until Port Royal plateaued. Hard to tell between runs how much is measurement error, but letting my water temps come back down to 25-26C then running again yielded pretty consistent results.

Won't surprise me if +175 is the game clock I'll use, but we'll see. Could try manually tweaking the curves as tedious as it is. Not sure how much I would gain there, and I'm kind of a noob at detailed OC'ing these days.

Managed to get 14969 in Port Royal and 5935 in Speedway with +207 core / 1450 mem. Would be nice to get an even 15k/6k but we'll see. I'll try again in the morning when the house cools off more.

I haven't figured out how to do a benchmark run for Borderlands 3 and ARK, but that would be nice to play with. Anyone have experience there?

 

[yoadknux]

At what voltage are you getting 2145? It's important because if it's a flat increase then what will usually happen is power limit triggers and the core clock runs much lower than you think it runs.
Memory OC was very weird for me, I would get artifacts at +900MHz, then look clean (and still increase performance) at +1000, and decrease performance at +1500.
I have a very similar setup to yours (mine is 3080 12GB not Ti, and 9900k not KS).

This is what I got at peak (with fans at 100%): https://www.3dmark.com/3dm/74763797?
And this is with the 24/7 OC (no crashes, standard usage, no 100% fans): https://www.3dmark.com/3dm/75319810?

I think you can hit higher on TSE since you have a Ti and run on water, maybe tweak the voltage-frequency curve a bit and search for a memory sweetspot.

 

[Voodoo Rufus]

At +215 is 1.018V
At +175 is 1.056V

Running other tests this morning. Did some Kombustor runs and there's zero difference in FPS or clocks when I change memory speeds from 0 to +2000, even at max res. +175 is the hard limit for Kombustor stability, same as Heaven. +180 crashes pretty quick.

I have yet to find a good memory speed/stability test.

Yes, I know Earthdog hates that power virus. No lecture needed.

 

[Nebulous]

My 3080 hates/dislikes any type of overclock. I tried Afterburner's OC scan tool and when it finished showed "Overclock unstable". I tried with using power limit enabled/disabled, and even tho temps max out @ 59c, there's very little performance increase. Boosting on it's own I net 2050 on core. I also tested the OC bios via switch and again very little performance gains.

So figured I'd let the card do it's own thing and it's happy.

 

[Voodoo Rufus]

I hear you. I never took the time to really see what my 2080Ti could do even after reflashing with a 380W bios. Too impatient.

And chasing a max stable OC isn't really that practical to chase 5% extra performance that would be hard to impossible to notice in general gaming.

I just catch the OC bug and hate performance left on the table....

NEED MOAR VOLTAGE!

 

[yoadknux]

At +215 is 1.018V
At +175 is 1.056V

Running other tests this morning. Did some Kombustor runs and there's zero difference in FPS or clocks when I change memory speeds from 0 to +2000, even at max res. +175 is the hard limit for Kombustor stability, same as Heaven. +180 crashes pretty quick.

I have yet to find a good memory speed/stability test.

Yes, I know Earthdog hates that power virus. No lecture needed.

These are very high clocks for that voltage. These cards under heavy load cannot sustain much over 0.95V-1V. So I suggest that you monitor the core clock while running the benchmark, I wouldn't be surprised if it's actually in the 1950-2025 area.

By the way, these Furmark/Kombustor/whatever don't indicate anything about stability... Port Royal is probably the best because it predicts whether RT games are stable, FireStrike Ultra is good for estimating stability under high power and Timespy Extreme is OK all-around with sustained clocks. But make sure you run the multi-loop stress test and not just the benchmark.

 

[Voodoo Rufus]

I've attached my best Port Royal run specs.

I know Nvidia limits the amount of fun that the AIB's can have with overclocking, but with my temps at 50C, and an overbuilt card, I feel like there's more to be had here without going subzero.

This also tells me I have plenty of radiator capacity and my fans are still loafing along. First, winter house temps are great at 65F. Second, I have a single row 360 exhausting out the top with a double row 280mm after that on the intake. I feel like one double row 480mm on top exhaust (would need a different case) would do likely a similar job.

 

[Nebulous]

I did notice that my card will max core at 2050 either by manual or auto boost. And voltage doesn't help. If anything if I add more voltage, I get diminishing returns.