Good news!

I also revived my old XGR wallet from 2014 and resynced the blockchain with the bootstrap database file.
Now it's alive and mining some interest

Power consumption is not yet optimized, but currently 122W @ PSU DC side (measured with my clamp meter), so ~130W @wall @92% eff PSU.

Some more tuning on my 570  
http://imgur.com/1naFKkM

I just tested an RX 570 and got the same results as Wolf0 regarding power consumption.
28.3mh/s with the usual mod with 120W@wall.
It is about 3-4W less than a RX 470 on the same settings.

You can simply go with the newest driver+patch+bios mod.

Micron:
You can find posts saying you can achieve 30mh with using 1750 strap above 2050MHz, but the micron cards that I tested were all unstable above 1950. No wonder the manufacturers limited even the 8GB versions to only 1650MHz...
Is there such a big variance among micron chips?

Any success with micron?
I tested an msi 470 8GB Micron yesterday and most I could get was 27.3mh/s with the 1750 strap. Above 1950 it was not stable.
I heard of 30mh. Is it possible?

There's been HBM mentioned earlier before in this topic....

I got my hands on my first card with HBM (Sap Fury Nitro).

It's behaviour is f@ckin magic! Totally different from GDDR5.

I got ideas from modded Nano and Fury X roms from the OCN thread and copied 400 timings to 500&600.
But nothing changed even a bit, although I copied the straps to both 00 & 01 tables! The hashrates remained exactly the same.

I discovered there is a much more complex mechanism that determines timings dynamically.
Clearly there are frequency ranges about ~20MHz wide, where the timings change gradualy.
I found a sweet spot where I get above 33mh with stock timing, but there is much more I think.
And also there is a strong connection between GPU&HBM freqs, the performance gain is not linear.

And what is with that 100MHz first strap? (There are straps for 100, 400, 500, 600)

So many questions...so much pain..!?

No, it 's a obvious sign of too much OC.
Hynix won't be stable over 1950 MHz with the common timkng straps.
Even with custom straps you are lucky if you can go over 2000 Mhz.
1950Mhz with 1500 timing should bring you about 28 mh/s.

...finally getting interesting...

somebody might share Hynix AJR and Sam datasheet perhaps..?

I wonder how the optimized custom timings are related to the given minimal values in the official datasheet?
I mean how loose are the official values, are there even room for improvement below them, or custom straps simply are adjusted to the recommended official values...?

Decoding is already done by many of us.
The hard part is finding the best values, that is not simply guessing.

An encoding/decoding tool could help that process very much.
Maybe someone publishes his own, I don't have time to write one.

What I wonder the most about is how custom straps improve hashrate through reduced latency and not through increased bandwidth. I think for example ZEC mining is much more sensitive to latency than bandwidth. Obviously fragmented memory operations rely more on latency than bw.
I'd like to get some insight on these different approaches in timing adjustments and their relation.

Yes, you did it right.
You should get around 28mh with the 1500 elpida strap.
There is also a slightly better elpida strap published in this forum not long ago. Search for it

That is too theoretic

There are some known attempts with more or less success. You can find them here on bitcointalk from the past few years.
I think the answer to your question is that:
1. Yes, it is that much hard, that not many can do it. Also special (internal) hardware/software is required to do it.
2. Details are not made public for obvious reasons.

jstefanop leaked some info on his new hw/sw solution that will be available soon.

No, it won't make difference. Address allocation is the key.

I did some research on this matter and it seems that the difference between RX and previous series is that the RX cards allocates more resources for configuration space.
The only successful methode bypassing this is to modify the resource allocation in the mobo BIOS but it's not that easy.
Maybe the other way would be modding the card's BIOS disabling some unused entries in these allocation tables...

I even tried server mobos with 10+ pcie slots...and guess what.... it does not boot with 6+ 4GB cards. If I put in a 3GB card for 7th card it boots up fine.
Definitely memory addressing issue. It would be nice to see a mobo bios mod for this issue.

Just to make clear

ASIC quality != OC potential

It has an impact on the required Vddc for a certain freq.
But it does not correlate much with the OC potential.

As far as you want to run your GPU underclocked & undervolted higher Asic quality is better regarding power consumption.

The number of memory errors is OK below the million range.
212 is practically nothing.

The segmentation of the timing straps differs for different series, so you have to use many different parsing rules to get the right numbers.
You have to gather all the required data.

We are curious about your results

I didnt have time in the past few days to continue experimenting, but hopefuly at the weekend.

your trp is still messed up.