Do I need to disable power limit? I looked under the hood and the frequencies are higher per clock state so going to try it out. Thanks.

So I was able to get 1506/4000 stable at 1.2 v as far as I can tell. Tested it with various 3D benchmarks then customized my rom to 'Bake-in' 1506 as the default TDP Boost clock entry. Unfortunately for some reason while gaming it Boosted down to 1341!

I couldn't understand why it would Boost down or throttle as I raised my TDP values in MBT but it was using clock state #61 instead of #74 which I had set to 1506. So I had to shift down the values in the Boost table and shift up the voltage table to match.

Now it will auto Boost to 1506 and throttle down normally but I kind of cheated here it feels like.

If anyone knows why please comment.

Thanks.

Can you send me the bios (here or via PM). I can look into it. On normal 980 Tis there is no need for "cheating". The "right" way is to edit the voltage table (voltage table tab in MBT) and set your desired voltage in the top 3 rows, like you did I guess, and then scroll down all the clk states and search the first state which does have a higher voltage like you set in the top rows. Let me describe it with real values. Let's say you set your card in the top 3 rows at 1245 mV, then you must search the first clk state down the table (maybe its 63 or 68) which does have a higher value than 1245 mV and set all the following states to 1245 mV. This let's the card use all the states up to clk 74 because the voltage of all states isn't higher than the voltage set in the top rows. Example, as you can see in every "Mod" picture the values were set to 1225 because the bios was locked to 1225 in the top rows.

As far as I know on general 980 tis the last value in clk 74 is 1281 mV and if you have put your top 3 rows to 1200 mV the last state should be accessible all the time. Have you tried setting the last value (clk 74) to 1200 mV too? Maybe the card was irritated that there is no state for 1200 mV at all? If the "cheating" method is working which forces 1200 mV at clock state 61 and it should also be possible to extend it to clk 74.

Im also not a fan of that "bake in" thing going on. Was it that tutorial on overclock.net which sets clock state and boost state to the same MHz value? Not liking that. Let the card boost and change the clock like it wants, you are just confusing the boost technology implemented in the bios by forcing boost clock and base clock to the same. If the card needs full boost clock and the bios is correct then the card will boost up to max clock on its own. But that's another story.

How high did you set your power limit? And have you set it "correct"? Always set the percentage values that way, that its logic for the card. Lets say your pcie slot has 66 watt and your two 8 pin connectors are set to 150 watt then your total wattage for 100% should be 366. Not 350 and not 370. And don't use the power limit slider. Just set your desired wattage at the middle row, the 100% row and the third row in the "block" set it a tad higher like 102% or something. Because there is a bug with windows sleep mode if you set the middle row (100% row) and the third row to the same values.

Chef,

Let me describe how I did my mod because I am at work and do not have access to the final modded rom file to share right now. Will post them later this evening. In anycase this is what I noticed. My cards for the most part did not boost past clock state #61 which on the stock OC bios for my card was 1430.5mhz. I ran GPU-z and the sensor data said that the perf cap was because of Vrel which I guess is voltage reliability. Only rarely did it ever go up higher to 1544 or 1551 which only happened like once and confused me because when I tried to manually clock 1551 the system locked up. I could not even get 1520 barely stable nevermind 1544mhz despite pumping 1.275v using Precision XOC with overboost and overvolt enabled which I found was the only way to raise the voltage on the 980 Ti KPE short of a +25-+50 mv boost using dip switches. Trying to unlock the three top voltage tables using Kepler bios tool and setting them to 1.281 caused Windows to lock up at the spinning wheel stage of boot up process. Read there might be some software lock via Microsoft that prevents that kind of mod but not sure.

Anyhow the problem I had was that no matter how I modded my firmware the cards would auto boost to clock state #61. Even if I manually set boost in common tab to lets say 1464mhz it would down boost to 1430 or what ever value was in clock state #61. So this is what I did. I set all entries in common tab to 'entry #0' and set TDP mhz, 3D mhz, and boost all to 1506 mhz. Set GPC values in P0 and P2 to 1506mhz. Then I rolled back the boost tables to have clock state #61 be 1506 mhz which was the max clock state my cards boosted to no matter what I did. Maybe it was limited by Asic score. I don't know.

Then I had to edit the voltage tables to have every clock state above #61 to be 1281 which was out of the range set by max voltage of 1212. Then I just copied the voltage tables for clock state #67 which corresponded to 1506 up on to #61 and up to #35 raising the min and max a couple notches just to be safe. This works for me and I stress tested it all weekend and it works. Unfortunately setting #74 to 1506 resulted in down boosting but at least this way the cards down clock the volts and mhz when idling. It would have even been possible to set a max boost clock if I set GPC, XBAR, L2C the same but didn't because I didn't want to take XBAR and L2C that high. As far as the power tables go I set TDP to 350000,350000. PCI-e to 75000,75000, two Eight-pin rail to 162000,180000, 162000, 180000, and power limit to 350000,351000. Read that you got to set the limit 1 watt higher than max tdp.

My temps are amazing at 35 to 37 degrees per core which is sad because Maxwell has like an overclocking wall around 1510 mhz where the voltage requirements are so high I didnt feel it was safe to try. Also wanted to advise others attempting the same to disable Windows Fast Startup and Fast Boot in bios so Windows gets a fresh copy of the updated firmware. I noticed sometimes my new updated flash didn't get recognized until a complete cold shut down. Still feel I have a little head room left I can do via software. My cards allow manual vdroop compensation via dip switches but Maxwell is a pain to overclock.

Will post my firmwares later tonight,

Thanks

Chef heres the rom file:

http://www.mediafire.com/file/jpf70nrfccwy1ko/roms.zip/file

Also I studied your png image you posted and I think that part of the voltage range on right side being past the voltage map limits the use of those clock states is wrong. Because if that was the case I would have never been able to boost up to 1430 on the stock OC bios for the 980 ti KPE which I verified multiple times at boosting to 1430 (CLK 61) even though according to your image would be limited at #52. To get a CLK state ignored both left and right have to be out of bounds as the right side is a max value. Doesn't that make sense?

Thanks for the rom and your detailed explanation. It looks like you are pretty much through the whole bios modding process as far as MBT goes. I thought you maybe mixed up the voltage table but after reading your text above I guess you understand how the voltage table works, which was the only point I wanted to take a look in your bios.
I will look into it anyway but I suspect its more and more a KPE specific behavior like not showing the right voltage above 1212 mV in tools. I know that the max. value for standard cards is 1281 mV. All values above they can't recognize and the bios is acting weird with such settings. This leads me to the question, because we know that the Ti cards always put one voltage step more then set. Have you tried setting the top three sliders at 1275, does windows still won't boot? I'm now confused myself because in my thread I wrote that the max values are 1281 mV even for the Ti cards and after all these years there wasn't anyone pointing out that his 980 Ti won't boot after setting it to 1281 mV. So a possible try would be to limit your top 3 sliders to 1275 and try booting windows again.

When I'm home (now I'm at work, stupid timezones) I will take a look into my old bios collection and if there is no ti bios with 1281 on the top sliders then you have found a 3 year old mistake which nobody ever noticed .

Regarding your card boosting to clk 61 instead of 52, was that with or without the +25 mV switches at the back of the card? Also the voltage table isn't that precise. It does not mean that you can't access a clk state which does have a higher value on the right side then your top slider is set to. The left value of the clk state does have a role too. That's what I hate about nvidias gpu boost. You just can't tell exactly at which clk state the card stops only by looking at the voltages, because of the voltage range each clk state has. And thats the reason you could boost up to clk 61 even the right values where higher (and the left ones were lower) than your stock bios voltage settings. But for full voltage table access you need to edit the right and left side like on the picture to make the card boost up to clk 74. I hope that's somewhat understandable.

And yes the maxwell wall is defiantly starting at 1500 and being really hard at 1550. Even more sadder is that also a KPE card does suffer from that wall. I mean whats the point of the KPE card when every good standard 980 Ti can do 1510 +- 20 MHZ too

No problem. Someone on EVGA's forum told me that the Kingpins voltage is locked down to 1212mv (+50mv if use dip switches) and that only way to raise the voltage and get real time monitoring of the voltage is either through Precision X with Overboost and voltage control enabled or use EVBot module. There might be a third way. I was able to verify that using Precision the voltage was in fact raised and accurately reported shy of 6mv but unfortunately that voltage was locked and wouldn't down volt under idle conditions.

Trying to set a higher than 1212mv voltage by unlocking the three voltage control tables to 1281mv using Kepler Bios Tool caused windows to not complete booting into the desktop. Read that was an intentional driver fault from NVidia though I can't be sure what was going on to cause Windows to fail booting all the way. I should have taken a volt meter to the cards and checked but didn't think of it.

About the boost state going to CLK #61 that was with and without. I'm like 90% sure because I turned off the voltage boost and it still was boosting to 1430 at Clk state #61. Anyhow I think your wrong about the right side of the table controlling weather or not the CLK state will be accessed or not because I looked at my tables and certain clock states still got boosted into despite the max exceding the voltage map max value. I read an article about Kepler boost states last week that stated nvidia has some algorithm it uses for that but can tell you if both CLK values are over the max map value it definitely will not be used. Sure we can both agree on that.

Anyhow it was disappointing that despite the much cooler temps that my Kingpin cards cant go over like 1510 mhz despite voltage raises. I mean I was loading at 32 degrees last night with Furmark! Maybe because I got them in SLI I'm limited. Maybe because my ASIC scores are 75.0 and 76.5. Though found out after the fact that the Kingpin cards are only meant for LN2 in regards to overclocking despite the custom PCB and extra power phases and VRMs. Putting them on water doesn't help.

Going to do more experimenting with overclocking them further later this week. Got an Evga Probe-It cable on the way. I personally think that the Boost states up and down are what cause instability at high mhz. Much like when your old school over clocking your cpu its best to disable EIST and lock in your frequency at high volts as the adaptive overclocking methods are so much more difficult though not impossible. So got more testing to do.

Thanks