That doesn't make any sense or have anything to do with what you quoted. What study are you referring to? I can't find anything even slightly related to this on google. Llano does not have any HSA features, period. The very first minor HSA features were introduced with trinity. The first major leap is going to be huma, which will be introduced with kaveri. Full HSA support doesn't come until 2015 with carrizo. Llano is literally just silicon level integration of the GPU, just like sandy bridge. Nothing more. It was the first stepping stone towards further integration. And they chose to call it an APU despite it being nothing more than a regular IGP. That makes your statement incorrect which is why I pointed it out. Trinity and richland are also pretty much just a regular IGP as well.
Doing a study on the theoretical benefits of HSA to llano doesn't make any sense. What exactly was this study about? The vast majority of code can't even be properly ported to openCL so the only real thing they could study would be the effects of using HSA acceleration vs. no GPU acceleration in a specific application. I'm not even sure how they would go about testing this since it doesn't exist yet.
Most likely this study you're referencing doesn't refer to llano, but something more recent. Or it's just talking about openCL, not HSA. Without a link I can't say for sure. But what you've implied so far doesn't add up.
Both AMD and Intel have spent a lot of money in recent years reaching out to universities and big software businesses to try and push education of parallel programming both on and off the GPU. And they have both been known to use universities for 3rd party data collection in regards to this. So you're likely just referencing one of these openCL demos.
As for why they are focusing on HSA that's a whole other topic. But I'll go ahead and delve a little bit into that here. They've been working on HSA since at least 2005 so I really doubt this study that you're referring to had any bearing on whether or not they went through with it. Their plan for HSA wasn't even disclosed to investors for at least a year let alone the public. The decision to invest billions of dollars and many years into this long term project was made by dirk meyer even before the merger with ATI was proposed. HSA can provide significant benefit for certain tasks if the code is properly optimized (which despite what they've been saying publicly can be tricky) but is completely useless for most things. That being said I still look forward to it since it will help us do some very interesting things with computers in the future. But yes they did decide to go with it "just out of the blue" regardless of how useful it will actually end up being. Of course "out of the blue" also includes internal research and theories devised by the engineers working on it. But they started developing it before they started trying to convince people to use it. They didn't wait around until other organizations thought it was a good idea. If they had they would have come into the game way too late to make any difference in the market. Big businesses pretty much always operate this way.
If HSA really is an open platform Intel will easily be able to copy everything that AMD is trying to do and might even make a better implementation since they have access to better manufacturing technology and a crapton more financial resources for R&D. In recent years openCL actually seems to be the one area where Intel IGPs are dominating AMDs. Which is a bit ironic considering AMDs current marketing platform for APUs is all about the how openCL acceleration for future applications will make their APUs faster than the competition. Everybody seems to be making HSA out to be some sort of super revolutionary platform that will make Intel cpus obsolete and shift the balance of power back to AMD. After seeing people say the same things about x86-64 before I am skeptical that it will really give AMD that much of a boost.
The following is entirely baseless speculation:
I think that what will probably end up happening is it will give AMD a slight lead in a few applications for 6-12 months until Intel implements it. Effectively killing any advantage they worked for. They won't be able to get strong enough software support fast enough for it to be a significant tide turning event. That's what happened with x86-64 anyways. I do find it very surprising that Intel didn't already get on board with HSA. But they could still be secretly planning to support it in the future if it takes off. Intels massive capital has allowed them to do sudden 180 degree turns in the past with ease in order to survive.
Doing a study on the theoretical benefits of HSA to llano doesn't make any sense. What exactly was this study about? The vast majority of code can't even be properly ported to openCL so the only real thing they could study would be the effects of using HSA acceleration vs. no GPU acceleration in a specific application. I'm not even sure how they would go about testing this since it doesn't exist yet.
Most likely this study you're referencing doesn't refer to llano, but something more recent. Or it's just talking about openCL, not HSA. Without a link I can't say for sure. But what you've implied so far doesn't add up.
Both AMD and Intel have spent a lot of money in recent years reaching out to universities and big software businesses to try and push education of parallel programming both on and off the GPU. And they have both been known to use universities for 3rd party data collection in regards to this. So you're likely just referencing one of these openCL demos.
As for why they are focusing on HSA that's a whole other topic. But I'll go ahead and delve a little bit into that here. They've been working on HSA since at least 2005 so I really doubt this study that you're referring to had any bearing on whether or not they went through with it. Their plan for HSA wasn't even disclosed to investors for at least a year let alone the public. The decision to invest billions of dollars and many years into this long term project was made by dirk meyer even before the merger with ATI was proposed. HSA can provide significant benefit for certain tasks if the code is properly optimized (which despite what they've been saying publicly can be tricky) but is completely useless for most things. That being said I still look forward to it since it will help us do some very interesting things with computers in the future. But yes they did decide to go with it "just out of the blue" regardless of how useful it will actually end up being. Of course "out of the blue" also includes internal research and theories devised by the engineers working on it. But they started developing it before they started trying to convince people to use it. They didn't wait around until other organizations thought it was a good idea. If they had they would have come into the game way too late to make any difference in the market. Big businesses pretty much always operate this way.
If HSA really is an open platform Intel will easily be able to copy everything that AMD is trying to do and might even make a better implementation since they have access to better manufacturing technology and a crapton more financial resources for R&D. In recent years openCL actually seems to be the one area where Intel IGPs are dominating AMDs. Which is a bit ironic considering AMDs current marketing platform for APUs is all about the how openCL acceleration for future applications will make their APUs faster than the competition. Everybody seems to be making HSA out to be some sort of super revolutionary platform that will make Intel cpus obsolete and shift the balance of power back to AMD. After seeing people say the same things about x86-64 before I am skeptical that it will really give AMD that much of a boost.
The following is entirely baseless speculation:
I think that what will probably end up happening is it will give AMD a slight lead in a few applications for 6-12 months until Intel implements it. Effectively killing any advantage they worked for. They won't be able to get strong enough software support fast enough for it to be a significant tide turning event. That's what happened with x86-64 anyways. I do find it very surprising that Intel didn't already get on board with HSA. But they could still be secretly planning to support it in the future if it takes off. Intels massive capital has allowed them to do sudden 180 degree turns in the past with ease in order to survive.
"Normally if given a choice between doing something and nothing, I’d choose to do nothing. But I would do something if it helps someone else do nothing. I’d work all night if it meant nothing got done."
-Ron Swanson
"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony
-Ron Swanson
"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony