Originally Posted by Opcode
That's the key to the whole operation. If AMD goes APU only for the next couple of generations (or forever). It will push developers towards optimizing their software for HSA. AMD doesn't hold a very large portion of the market, tho the little bit that they do have accounts for millions of dollars in software revenue. So it definitely wont be something that large software vendors will ignore. Even if they do, a quad core with Nehalem-Sandy Bridge performance is more than enough for today's consumer. Tho I can almost guarantee at some point all of your commercial grade software will be HSA optimized. Software like Adobe Suite will be quick to adopt the technology. Keep in mind OpenCL wasn't exactly an easy SDK to utilize. Tho it still made itself known in lots of today's software.
- Adobe PhotoShop CS6
- Battlefield 3
- Final Cut Pro X
- Mathematica 8
- VLC Media Player
HSAIL on the other hand is to be released as a extension for high level languages such as C++ and Java. Making it that much easier for developers to adopt the
technology. The problem that I foresee isn't the lack of developers adopting HSA. It's what will become of GPGPU once a major software vendor does. The
difference in the software's performance will be astronomical even in comparison to the Socket 2011 line of CPU's. An analogy would be a city full of gas stations.
If one gas station dropped the price of their gas by $0.50, all the other nearby stations would follow suite. Simply because if they didn't they would lose a large
amount of revenue to their competitors. The software world works exactly the same. Your software suit has to remain up to date with the latest that technology
has to offer. Otherwise someone else will fill that gap with their own implementation and claim half of your consumer base (you still got the "I am familiar with it,
so i'll keep using it anyways" bunch). HSA wont replace what OpenCL already is to the consumer. Tho in the perceivable future it will help GPGPU become closer
to being a standard than the OpenCL SDK ever did alone. You also have to think beyond the consumer, APU's are more cost effective to businesses and schools. I
am also certain a super computer could be made of APU's. Web servers, game servers, and other software can be written to utilize the power of GPGPU. Which
would lead to growth in AMD's server market share. I am not sure if anyone here has ever written a game server before, but there is a lot of math that happens all
the time (especially floating point calculations in MMO's). I mean you both are right, that it will be limited to what software is readily available to utilize it. Tho I
personally think GPGPU should be standard by at least 2016. Intel and AMD could keep piling on the x86 cores to raise the performance of their processors. But it
doesn't scale as good as having a few x86 cores paired with GPU cores. Especially with both of them taking up the same space and the APU consuming equal or
less power. AMD certainly could stack 12 steamroller cores into one hell of a FX monster. Tho if you stood it up against their 4 core flagship Kaveri APU in
software optimized for both CPU and GPU cores. The APU would run circles around it all day long. This is why AMD is moving in the direction that they are
moving. They are tapping into a new resource that Intel hasn't even given much thought to yet./Quote
The theory is fine, but how quick software vendors adopt it is another issue. AMD must survive the transition period. It could be another 2-3 years before massive adoption is a fact. Will the public buy a 4 core steamroller or excavator if they are told it won't be better or equal to Intel? No they won't . That is why you have to have a product in the current market with enough horsepower to kickass on I 3 and I 5 processors. That can only happen with 6 or 8 core steamroller or excavator.