So we've still got a potential 9 months for SR.
I wonder if AMD are actually going to bring any half way product out to replace the FX83XX.
If not, then again it's evidence of AMD not putting on the pressure.
And I'm still sceptical about SR APU's, Richland isn't even properly out yet and we're a month away from Q3.
I don't see AMD releasing Richland and then bringing out another APU straight after.
"AMD have technology, but lack urgency" - Roy Reed
Yes and no, some things you can't rush. AMD have had HSA locked down pretty much for a while and could have pushed for HSA and earlier fusion products, but you have to weigh up your options when you're losing cash. You can bring out chips as often as Intel, but it costs a twice as much to release chips twice as often, and it gives you less time between releases to recoup R&D costs, when you sell to 80% of the market, thats easy, when you sell to 20%, you'd only lose money even if you increased sales.
Likewise, the market/world/devs/OS's have to be ready for fusion, and other advances, often they can't support new instructions, or more to the point they often can't get new instructions USED until Intel uses them and pushes new compilers and their partners to use the instructions.
But in terms of Bulldozer > PileDriver > steamroller, full on cpu's tape out to release is a 18 month process, for AMD or Intel, give or take a couple months. They also can't release a 300mm2+ 4Ghz + chip till GloFo can make it, its a 28nm chip, and they can't make it in volume or sell it before the process is ready and can support it.
Also AMD has pretty much said Xbox One will be worth 3billion + to them (I also think this is minimum and will be that even if sales or low, if sales are better they will only bring in a lot more), PS4 is likely to be worth even more to them, potential $250mil per quarter, or a billion a year, maybe more. Putting money/effort/man power into Jaguar over potentially bringing Steamroller forwards was easily the correct decision.
Steamroller the cores will probably be launched this year in APU/Kaveri form, and even potentially(rumours are gathering pace) gddr5, which would not give it Intel style edram bandwidth, but hugely increased bandwidth on current chips, with FAR more powerful cpu's and more bandwidth than any desktop versions of Haswell as well.
octo core Steamrollers have the potential to be significantly competitive with Intel hex core products and potentially leave Intel quad cores in their wake, with Intel potentially not having an awful lot in reply for a year + themselves.
Doing it right is better than doing it earlier and AMD have been doing things right for a few years now, the results just take a significant amount of time to pan out in products in the market.
Bit of a shame for Dirk and a bit lucky for Read, he'll get huge credit for the deals being done now and the products releasing in the next couple years, but these were all products basically taped out(or very very close) before he even joined AMD.