AMD really needs 300 series to be good. They basically need whatever it takes to hold out for the next 2 years until 14/16nm for GPUs drop and Zen is out. Hopefully 300 series and booming XB1/PS4 sales are enough for them to stretch it to late 2016. After that things should be easier. I am far less concerned about AMD's 300 series competing well on the desktop considering just how low the performance bar that was set by 970/980.
Pretty much AMD is holding its own with outdated 290/290X designs, which themselves were not even an improvement over Tahiti in terms of IPC and perf/watt. I basically view 290/290X as a larger Tahiti with a bigger memory bus and more units. The underlying architecture is more or less intact. Even geometry performance hardly improved with 285 beating 290X by nearly 70%. This shows that the original GCN was really a great architecture as it has now kept up for 3 years but a drastic improvement is needed for it to evolve and be relevant for 2015-end 2016 as games become even more advanced.
I maintain that AMD's 390X must be a big improvement over a 290X because I don't think it would have been that difficult for AMD to squeeze 15% more performance from a hybrid cooled and more mature 28nm 295X. If AMD didn't even bother refreshing Hawaii like they did 7970-->7970Ghz, sounds like they put all their effort into 300 series as a do or die.
What's more concerning though is AMD's ability to gain mobile dGPU market share. Once the major OEMs lock in 850/860/965/970/980M chips, they don't tend to release newer products for 6-8 months. Usually major laptop designs are updated around new Intel CPUs and GPU launches. These tend to come around June/July and Late fall/Holiday season. If AMD wants to be relevant in 2H of 2015 for mobile wins, they really need to have mobile 300 series ready by Computex, to align with Intel's Broadwell/Skylake launches. I suppose this time may be a slight exemption since Windows 10's launch could result in the notebook makers and OEMs updating their products more than the usual 2x a year, but generally speaking NV has the next 5-6 months already locked in with Maxwell's mobile design wins.
Apple's MacBook Pro has basically remained unchanged for 12 months with just a minor bump in CPU/RAM/SSD specs. AMD needs to capitalize on M290/295X Apple design wins, although considering how awesome the performance/watt of Maxwell is, R9 M300 will need to pull off a miracle to regain mobile dGPU market share in 2015 and get inside Mid-2015 MBP Retina and refreshed Mac Pro.
I said 2 years ago that AMD should not waste money on fighting Kepler as that battle was not even important. Unfortunately, looks like my hunch was true as Maxwell is devastating R9 200 series with AMD having no response other than price cuts. If AMD can't win on perf/watt, they would need to outright beat Maxwell in both price/performance and top end performance. This is one of the toughest times for AMD's GPU division and their silence isn't instilling much confidence as thousands of gamers got tired of waiting and jumped on 970/980. This month 960 will start to do even more damage to AMD's sub-$300 lineup. NV already added 965M too basically dominating mobile dGPUs now against nearly 3 (!) year old 7970M lineup. Ouch.
Luckily for AMD, Sony and MS have a spectacular line-up of console games in 2015, which should allow for the console sales momentum to be maintained during the 1H of 2015, just enough time for AMD to launch R9 300 series. It would also help AMD a great deal for high end 390/390X sales if 4K monitors started dropping in price rapidly in 2015 and if FreeSync monitors started coming out more aggressively than the slow roll-out of G-Sync monitors.
Whether or not R9 300 is 20nm is less relevant than true gains in perf/watt, price/perf and absolute performance. If AMD can also add some extra features such as DP 1.3, they will definitely have something extra over Maxwell. DP 1.3 would be huge as as it would allow the card to drive dual 4K monitors, 5K and even 8K, albeit at 4:2, but still.
http://www.anandtech.com/show/8533/...rt-13-standard-50-more-bandwidth-new-features