The GPUs are the same but the PCB's are far from it. Therefore you can not necessarily match any card's overclock with any same series card. That would be equivalent to saying a specific sports car is as fast as every other model with the same base engine model, despite the fact that some may have turbos, superchargers, transmisson, gearing, tire, exhaust, fuel supply, suspension and other performance related modifications.
Differences between reference cards and non-reference cards over the last few generations may include for example the following:
1. Coolers..... better coolers / fans / fan control mean lower heat, lower heat means higher/ more stable OC's.
2. Bigger / better VRMs .... this was most evident with the 570 cards where owners of EVGA's SC series for example were frying their VRMs 4 phase when over clocking while Asus and other users with 6 / 8 phase VRMs were just fine. Aside from "not frying", better VRMs result in more stable power which in turn results in more stable OC's.
3. Better componentry .... super ferrite chokes, better japanese capacitors all lead to more stable power and better overclocks
4. Better PCB / VRM Cooling .... the VRM is the hottest point on the card and when it hits up near 100C, your OC stability is compromised. With the 900 series, this is less of an issue than it has been in the past.
5. Better Memory - Look no further than the Asus 780 DCII which arrived in two flavors, one with Samsung memory, one with Elpida. Those who drew Elpida were sorely disappointed as overclocks were substantially less.
6. Better Memory Cooling .... some vendors provide heat sinks / thermal pads on the memory, some do not. This affects ultimate overclocks.
7. Binned GPUs .... some vendors have "binned" the GPUs handpicking the best ones for the premium lines. Asus 670 DCII TOP is prolly the most well known example.
8. Allowable tweaks ... some vendors have permitted additional tweaks within their overclocking (i.e. voltage) on some of their factory OC'd lines.
9. Physical Voltage Modding .... some vendors have provided the means and tools to allow one to bypass nVidia's voltage locks.
There is the reason that some cards consistently overclock better than others .... when you see the same results in review after review, it's obvious that the silicon lottery is not at play here. If all cards could overclock the same, when overclocked by reviewers, we wouldn't see things like this:
Reference 970
http://www.guru3d.com/articles_pages/nvidia_geforce_gtx_970_and_980_reference_review,28.html
Core Clock 1326 MHz
Boost Clock: ~1452 MHz
Memory Clock: 8002 MHz
Asus 970
http://www.guru3d.com/articles_pages/asus_geforce_gtx_970_strix_review,26.html
Core Clock 1264MHz
Boost Clock: ~1443MHz
Memory Clock: 8002 MHz
MSi 970
http://www.guru3d.com/articles_pages/msi_geforce_gtx_970_gaming_review,26.html
Core Clock 1325 MHz
Boost Clock: ~1501 MHz
Memory Clock: 8002 MHz
Gigabyte 970
http://www.guru3d.com/articles_pages/gigabyte_geforce_gtx_970_g1_gaming_review,26.html
Core Clock 1328 MHz
Boost Clock: ~1516 MHz
Memory Clock: 8002 MHz
Now if you read the bit-tech article above, you will note that the Asus has physically little to differentiate it from the reference model, and as such, has similar (though actually in this case worse) performance than the reference model. The MSI model which has substantial differences (as does the Giga) and we find performs better. HighTechLegion hit 1567 and other sites have broken 1600. I have not found a review as yet where either the reference model or Asus 970 broke 1500.
nVidia has increasingly clamped down both legally and physically as to what it board partners can do with voltages and overclocking potential and as time has gone on, the differences we see on the non-reference boards have shrunken. But even though the performance differences have shrunk over the last 3 generations, we still find that for little or no cost increase, we are getting better performance, quieter systems and better components.