TiN_EE
AHowes
Getting close to 2200MHz of stable performance on TU102 is no easy, and there are many quirks in drivers/settings/VBIOS that can ruin the performance. That is why I kept repeating always test for performance, not just the highest clocks. It's impossible to troubleshoot your case remotely from just few scores without knowing details about your system, but if I'd be you I'd start with methodical approach, testing only one thing at a time. That means - test GPU clocks and GPU stability first. Have default voltage, max out power limit slider, pick one benchmark to run, and only adjust GPU clocks by +15 MHz at a time. Write down performance numbers. You don't need run whole benchmark suite, just one GT test would give you good indication if performance is scaling well. Go step by step, till you hit either stability problems (benchmark crashing/artifacts) or scaling issues (no more performance increase with higher clocks). This all with default memory, btw. Then bump the voltages on GPU a little bit (i'd say 25mV is good step). Check if that helps with stability or scaling in benchmark. I wouldn't go any higher than 1.30 on positive cooling for benchmarks, and even less for 24/7 use. By doing this for some time you will get a hint what GPU likes. Now back off on GPU clocks to default or near default, and do same for memory. Increase memory clock by 50-100 MHz, find out the max stabillity/performance scale. Bump voltage to 50mV, test again, more 50mV, test again some more. After few more days you will know what memory can do. Then repeat the whole shebang but this time with GPU and memory OC together. This will give you max clocks card can run on your case, with X temperatures and Y voltages. Now play with load-line settings, to find how that affects GPU clocks/performance. Difference can be a LOT, both temperature/power wise and clock wise. Sorry, there is no way around doing all this, Kingpin card, or no Kingpin card, as there is always some things that work bit differently for each system/benchmark. Drivers also can have large impact on performance, I would definitely try to use same driver you used with FTW3 before doing any score comparisons. When Vince benches for records, he spends days pre-testing all silly drivers to find good one before giving cards even a sniff of LN2. Doing all this just for one good benchmark run is a reason why any overclocker will laugh at those online statements like "oh, it's easy to make record, just slap the LN2 pot, pour some LN2, crank the volts, and record is yours, lol" :D. I would also try different clean OS just to rule out software issues. I've seen 3Dmark go haywire many times before after trying OC settings for so long, resulting wrong scores or loosing stability on perfectly stable hardware. Playing with CPU settings can be helpful too, like ring ratios, DRAM frequency/timings to make sure score is not capped by system side.
 
Some might not be willing to put all that effort for 200 more points on some benchmark (5 more fps, 5% performance boost for XX$$$, etc). Sure, that's okay too, just enjoy the card out of the box with higher power limits and hybrid quiet operation. There are no deliberate hidden secret performance caps to make everybody feel difficult, other than what's already in TU102 and drivers :). If you want performance only at all costs, try xoc BIOS and live with the consequences. You are right expecting same capability from same TU102 GPU, but default settings on one card may not be same as default settings on another card. Each CPU/GPU/memory have own sweet spots to deliver best performance. With RTX KPE we tried hard to get more for all around performance card, than track-only beast. And doing that for every KPE card is not simple. FTW3 you had running 2160 is great sample, however your neighbor with his FTW3 could be not as lucky for OC. Reducing that luck factor for KPE, given it's high cost was very important for us. With all adjustment settings and knobs available, it's not always easy and simple as plug card in system and get max scores, but we try to make all KPE cards do great scores at least :). Think of this elaborate OC experience as a penalty for YOU having advanced control to your own card operation, rather than everything-corkeddown-limited-and-predetermined by vendor like with any usual RTX card.

TiN_EE
AHowes
Getting close to 2200MHz of stable performance on TU102 is no easy, and there are many quirks in drivers/settings/VBIOS that can ruin the performance. That is why I kept repeating always test for performance, not just the highest clocks. It's impossible to troubleshoot your case remotely from just few scores without knowing details about your system, but if I'd be you I'd start with methodical approach, testing only one thing at a time. That means - test GPU clocks and GPU stability first. Have default voltage, max out power limit slider, pick one benchmark to run, and only adjust GPU clocks by +15 MHz at a time. Write down performance numbers. You don't need run whole benchmark suite, just one GT test would give you good indication if performance is scaling well. Go step by step, till you hit either stability problems (benchmark crashing/artifacts) or scaling issues (no more performance increase with higher clocks). This all with default memory, btw. Then bump the voltages on GPU a little bit (i'd say 25mV is good step). Check if that helps with stability or scaling in benchmark. I wouldn't go any higher than 1.30 on positive cooling for benchmarks, and even less for 24/7 use. By doing this for some time you will get a hint what GPU likes. Now back off on GPU clocks to default or near default, and do same for memory. Increase memory clock by 50-100 MHz, find out the max stabillity/performance scale. Bump voltage to 50mV, test again, more 50mV, test again some more. After few more days you will know what memory can do. Then repeat the whole shebang but this time with GPU and memory OC together. This will give you max clocks card can run on your case, with X temperatures and Y voltages. Now play with load-line settings, to find how that affects GPU clocks/performance. Difference can be a LOT, both temperature/power wise and clock wise. Sorry, there is no way around doing all this, Kingpin card, or no Kingpin card, as there is always some things that work bit differently for each system/benchmark. Drivers also can have large impact on performance, I would definitely try to use same driver you used with FTW3 before doing any score comparisons. When Vince benches for records, he spends days pre-testing all silly drivers to find good one before giving cards even a sniff of LN2. Doing all this just for one good benchmark run is a reason why any overclocker will laugh at those online statements like "oh, it's easy to make record, just slap the LN2 pot, pour some LN2, crank the volts, and record is yours, lol" :D. I would also try different clean OS just to rule out software issues. I've seen 3Dmark go haywire many times before after trying OC settings for so long, resulting wrong scores or loosing stability on perfectly stable hardware. Playing with CPU settings can be helpful too, like ring ratios, DRAM frequency/timings to make sure score is not capped by system side.
 
Some might not be willing to put all that effort for 200 more points on some benchmark (5 more fps, 5% performance boost for XX$$$, etc). Sure, that's okay too, just enjoy the card out of the box with higher power limits and hybrid quiet operation. There are no deliberate hidden secret performance caps to make everybody feel difficult, other than what's already in TU102 and drivers :). If you want performance only at all costs, try xoc BIOS and live with the consequences. You are right expecting same capability from same TU102 GPU, but default settings on one card may not be same as default settings on another card. Each CPU/GPU/memory have own sweet spots to deliver best performance. With RTX KPE we tried hard to get more for all around performance card, than track-only beast. And doing that for every KPE card is not simple. FTW3 you had running 2160 is great sample, however your neighbor with his FTW3 could be not as lucky for OC. Reducing that luck factor for KPE, given it's high cost was very important for us. With all adjustment settings and knobs available, it's not always easy and simple as plug card in system and get max scores, but we try to make all KPE cards do great scores at least :). Think of this elaborate OC experience as a penalty for YOU having advanced control to your own card operation, rather than everything-corkeddown-limited-and-predetermined by vendor like with any usual RTX card.