Intel are massively ahead of their competition on the desktop and server. I don't believe for a moment that HSA is going to change that around. Especially for application servers.
You do realise I presume, that Intel have gone from pure CPU's what 6-8 years ago to dedicating a HUGE amount of on die transistors to the GPU. What on earth makes you think Intel aren't going to go towards the same types of software for compute, server, desktop and applications. You think Intel spent billions on R&D to add huge gpu's to their highest volume selling parts and will entirely ignore them while the rest of the industry starts to utilise them better?
Intel has absolutely made all the same moves in terms of hardware. They have made all the same cpu>apu>soc moves that AMD and mobile chips have, you think this was for minecraft alone? No one rates Intel drivers for 'proper' gaming, no one buys a gaming laptop or desktop looking for an Intel GPU. They put a huge GPU in every architecture but have almost no plans to use it, even though as soon as they went APU they added things like quicksync?
Why haven't they got loads of gpu compute performance, applications? Because the industry doesn't support it, AMD the same, they have the hardware but not full industry support. There are no real native languages that help do parallel acceleration WITHOUT explicit complex coding and software companies don't like specific code to target multiple different architectures on multiple platforms because it's a huge amount of extra work. That is what Java(and other languages) are adding. HSA is the industry support side of what AMD added in hardware. The only difference is with HSA support in Java, coders can VERY easily add commands to tell Java to use gpu compute and it will be able to use this code on any HSA compatible hardware. IT means writing both simplified/easier code to take advantage of gpu compute AND it means writing it ONCE which gets used on every platform rather than complex/difficult code that they need to code lots of times and differently and support that for every platform they use.
Hardware is the complex but still easy part... getting the industry as a whole to move to new languages and code differently... that is the exceptionally hard part.
AMD have ostensibly had DX12 hardware for what, 3-4 years... and software support is coming end of this year from MS. it took them starting Mantle to push MS then the industry in general to support 4 year old hardware to the best of it's capability.
Your own link showed in multiple cases where the FIRST 64bit ARM chips were within the same ballpark in many instances of LONG established 4/5th generation Intel chips. Not least the software you spoke about has been optimised for Intel chips/architectures for a decade, you don't get the same level of optimisation with a first gen product that has never been available before. There is a huge amount of improvement that will come from ARM chips in server space in a short space of time, software optimisation and second/third gen hardware with MUCH shorter development cycles to fix all the biggest problems. More to the point is the cost of an ARM server vs the cost of an expensive Xeon server.
ARM is making big inroads into servers, this will continue, software support WILL improve dramatically for ARM servers, ARM chips will improve very quickly in the area. HSA will make it easier for software devs to target more platforms, in server, in mobile, in desktop without having to change as much code.
If you think Intel aren't also ready for gpu compute, you're blind to everything Intel has done in the past 7-8 years. The industry has just been waiting on the tools to move forward with supporting gpu compute properly. HSA is such a tool which many major software devs and multiple MASSIVE hardware companies are going to support.
