I am using my old marijuana grow tent with lighting removed.
2 x 4 ft tent holds two 8 GPU rigs no problem with an old exhaust.

I don't think it'll happen ever. The problem is not only in memory per se but in memory bandwidth. Fast memory = expensive memory. And that is what makes a good ETH miner, not the speed of SHA rounds. So I don't expect any breakthrough in ETH ASIC mining. Somebody can try though, I might be totally wrong.

Think "stripped down GPU with SHA3 hardware units instead of general-purpose compute units". Won't be cheap though, nothing near Ant Miners.

Actually, yes we are. One thing that keeps traditional cryptography alive (we're still talking about asymmetric schemes based on large number factorization, correct?) is the fact that the largest quantum unit today (that we know of) is 50 qubits. Make it 5,000 and your SSL session can be decoded in seconds, no matter how good your entropy source is. EC-based is a little bit tougher but not by far.
One technology is there you'll need quantum devices to secure your information. But we're talking about decades timeframe, cryptocurrencies will be long dead by then.

I doubt it. Ethash was designed (please /lcybzq6]read the design paper) to require huge memory bandwidth.
ASICs give you advantage in SHA Hashes/Watt ratio, but for Ethash that represents a small fraction of the process (unlike the Bitcoin, that, essentially, IS a SHA).

In other words, ASICs can give you cheap SHA rounds but can't provide you enough memory. You have to have 1GB dataset per ASIC.
Technically you can regenerate parts of the dataset as you go - but that is going to kill the performance (+16MB ASICs aren't cheap).

It is still doable in ASICs - however not economical.

Don't worry, mining is going to be your smallest problem. As soon as somebody put together even 1k qbit unit traditional cryptography as we all know it is dead - and so does the world as we know it. Think about it: Internet commerce is gone, secure communication is gone, financial system is back on dedicated dial-up lines. You can't pay with a credit card anymore, you can't trade unless you're physically on the Wall St. Welcome back to the 70s, man.

You're missing the point.

It doesn't matter how good these new algorithms are going to be. You brand-new shiny fancy "supercoin" is worth as much as people want to pay for it (a.k.a market cap). Bitcoin is a king here and is going to stay the king until bubble pop -just because billions of $$$ are already invested in Bitcoin. Your "supercoin" (and I assume it is mathematically/cryprographically is superior to Bitcoin) is going to be the third class member of this society at best. Most likely it won't worth the electricity spent mining it.

So I don't expect any new algorithms or any new coins of significant value; Bitcoin will stay #2, ETH #2 and I'm not sure about #3.
Frankly, I think Bitcoin will collapse rather sooner than later. However, it's not time yet. I think by 2019 it'll be all over.

ETH and others aren't CPU heavy but rather memory heavy (please read ETH design paper). So any ASIC for ETH has to have TONS of fast memory. You can still make ASIC for these algorithms but they won't give you such tremendous advantages over GPUs as for CPU-heavy Bitcoin.
However, Chinese guys can kill ETH and others just by concentrating their money on the Bitcoin (so nobody will want to buy ETH at the first place).

OTOH, this all madness is a bubble / Ponzi scheme. So I think it is going to crash sooner rather than later. I wouldn't invest anymore at this point; just make sure you cover the equipment/electricity that you already bought to at least break even. To get rich on the criptocurrencies you should've invested big $$$ back in 2010.

Disclaimer: This is just my opinion. I am not your financial adviser / lawyer / rabbi / etc.

Hashrate is very unstable, going up and down from 15 to 29 MH/s.

Applied default patch using Polaris 1.6.7, no overclock.
When I replace Sapphire with any other brand I have - problem disappears and hashrate is stable at ~29MH/s.
So I assume the problem is with the card brand itself (as I have two of those and they behave identically).

PN: 299-1E387-0008A
SKU#: 11265-05

Thoughts?

Performance graph: https://i.imgur.com/UUeL8iX.png
Model: https://i.imgur.com/IPDPGjt.jpg
Actual card: https://i.imgur.com/KCbDQci.jpg