Memory speed capability mostly determined not by gpu itself but the memory chips installed and its cooling. Asus afaik has no mem cooling so this combination - slow hynix plus bad design yields low stable mem speeds. I run 1950mem with 1375 strap on sapphire ref-like rx470 without mem errors in hwinfo64.
I was a bit skeptical, since I've worked with lots of servers before and the RAM would only get warm to the touch. However, I rely on data rather than my ego to determine what is correct, so I got out my IR thermometer and was surpised to find spots on the back of the card as hot as 98C. I bumped the fan to manual 60%, and that brought the hot spots down to the 80s. I then got out my infrared camera, and I could actually see the rectangular hot spots from the RAM on the other side of the board. The worst is above the GPU where the aluminum heatsink blocks any airflow over 3 of the DDR chips.
With a small rectangular piece of aluminum plate and a thermal pad I could probably make a thermal bridge from the RAM to the heatsink to cool them down...
See my post here: https://bitcointalksearch.org/topic/wich-is-better-rx-470-refference-or-custom-fan-1615394
Actually I remember seeing that when you posted it. But nobody actually gave any concrete info on the effect of DRAM cooling. Had I known that they can get over 95C without heatsinks, I would've paid more attention to that when buying a card. With the back of the board in the high 90's, the DRAM die likely was over 110C, or 25C over the maximum operating temperature for most Hynix RAM (85C).
https://www.skhynix.com/static/filedata/fileDownload.do?seq=267
Since power scales exponentially with voltage (close to V^2), if I can undervolt the memory a bit, I might be able to keep it cool enough to actually overclock it a bit more. I'm sure the Digi+ VRM can be controlled through I2C, but I don't want to dig up a data sheet or reverse engineer the register set in order to do it...
- Measuring the chip itself (Tcase) or the back of the board really is not relevant at all. GDDR5 is low voltage and it would be unlikely that the temp will have any affect at slightly overclocked speeds (Particluarly on Reference Cards as Wolf has pointed out). The outside of a Ram Package and the Board are suppose to be hot to dissipate the heat. If the cooler is modded and/or the thermal pads are missing then that may be different but even then I would say that at the speeds used in mining that it is rare to overheat ram. 
