The only difficulty I have is GCN assembler. I've spent a lot of time to create efficient environment for development in asm, that's the only reason of the delay. It's one-time delay related to all my miners, now I see that with assembler I can improve them all.
Does that mean that the eth miner and also xmr miner will get a boost?
What about rx support at xmr? Still get sometimes the fan bug.
At least for GCN 3 devices (like Tonga), ethminer is within 1-2% of the memory read performance limit of the cards. GCN assembler won't make it any faster for those cards.
Plus with the profitability of eth mining to drop significantly in the next few months, miner developers are generally not interested in doing more work into eth.
Now, this is technical and i'm out of my element, but the Tonga cards should be able to workout 384-bit mem bus, problem is, after a bit of discussion i found out, you would have to pretty much write a bios from scrap, but, theoretically remove that bottleneck giving a 50% increase in performance, the 380x would suffer less if there are any bottlenecks, and the memory straps could also be modded in the way that certain ETH mods gave hashrates of 25+/s (mind you with wild results with regard to heat and stability the higher you went)
Basically, Radeon release the R9 285 at a strange time, and it was the first GCN 1.2, and only i believe, but they couldn't have it be too good as to remove from their 290/390s or the fijis they planned to release a few months later, as well as the 16nm stuff (RX series) So in theory the 380 is capable of a lot more, but AMD settled on a theoretical number they felt would be good for their entire line, and so the 285 meant to have 32 CUs had 4 of them locked,1/3 of it's memory bus knocked, put on PCBs meant to take their mediocre power draws and heat output. Honestly if they went full-pin and made a flagship with it, it would have fallen somewhere fiji and hawaii, closer to fiji. I'm sure they tested it too, but it was just good business.