Playstation and Xbox sales are ~5% of PC shipments.
Not to mention, the fastest compiler for AMD chips is...the Intel compiler.
unless they're allready doing it for playstations and xbox'es of course ..
It doesn't improve the hardware, it improves the software. AMD is never going to out-fox Intel on the high-performance x86 front, so AMD needs to leverage the advantage they do have: GPUs.
Stream, OpenCL, C++ AMP, etc have yet to catch on with most developers because all of it still requires the GPU to be accounted for and programmed separately. With HSA, programmers are effectively programming against a single virtual ISA that covers both the CPU and the GPU, meaning they no longer have to explicitly program for the GPU. And that is how AMD intends to get programmers using the GPU, thereby allowing them to exploit their GPU performance advantage and outperform Intel.
The main problem is if nVidia and/or Intel doesnt back it. Then it wont turn universal and as such...and suffer the same fate as SSE5, SSE4a, 3Dnow! and so on. You aint gonna code/compile for a 17% and decreasing segment of the market.
Then the discussion if its better or not is purely secondary.
What about if ARM, Samsung, TI, Qualcomm and all the other members back it? Are Intel and Nvidia volume leaders in the total semiconductor market?
Its irrelevant. Their software base is volatile and they tend to code for an upper layer like Java (Not Oracle Java). Samsung for example gonna ship Tizen devices this year with Intel.
Also, what companies are actually making HSA hardware?
It doesn't improve the hardware, it improves the software. AMD is never going to out-fox Intel on the high-performance x86 front, so AMD needs to leverage the advantage they do have: GPUs.
Stream, OpenCL, C++ AMP, etc have yet to catch on with most developers because all of it still requires the GPU to be accounted for and programmed separately. With HSA, programmers are effectively programming against a single virtual ISA that covers both the CPU and the GPU, meaning they no longer have to explicitly program for the GPU. And that is how AMD intends to get programmers using the GPU, thereby allowing them to exploit their GPU performance advantage and outperform Intel.
Playstation and Xbox sales are ~5% of PC shipments.
Not to mention, the fastest compiler for AMD chips is...the Intel compiler.
Tizen...an OS backed by hardware companies. They will get very, very far >.>''
If performance is any metric - then a Intel\NVidia GPU will win..... many times over.
If performance\watt - then well probably ARM no?
(If this gains traction - any Compute oriented semiconductor IP creator will obviously have to support it so there).
To exploit HSA, you need a high performing GPU component with a low-latency shared memory/shared cache architecture, along with a unified address space and fast context switching on the GPU side of things. In other words, you need a tightly integrated CPU/GPU combination.Yea, but couldnt you just have an intel cpu and add a discrete gpu? Have the best of both worlds. I suppose it would be a bigger advantage in laptops and other mobile devices where it is not easy to add a discrete gpu.
To exploit HSA, you need a high performing GPU component with a low-latency shared memory/shared cache architecture, along with a unified address space and fast context switching on the GPU side of things. In other words, you need a tightly integrated CPU/GPU combination.
Pairing a high performance CPU with a high performance dGPU does not work in this situation due to the lack of unification between the CPU and the GPU (you'd die shuffling data between them). This effectively excludes NVIDIA in the x86 space (now Tegra would be another matter), while Intel would be going up against AMD's faster GPUs.
Not at all. I'm just pointing out AMD's rationale for going this direction. HSA is how they intend to exploit the advantage they have for now; whether they'll still have that advantage in 5 years or whatever is anyone's guess.So virge foresees the great AMD - will always and forever carry the GPU peformance advantage in Compute on a single DIE?
Not at all. I'm just pointing out AMD's rationale for going this direction. HSA is how they intend to exploit the advantage they have for now; whether they'll still have that advantage in 5 years or whatever is anyone's guess.
very much so in trying to get rid of the x86 monopoly we've had on the heavy compute side.
I don't see a monopoly here. For what I'd call "heavy compute" nvidia has had the market cornered for the last few years. x86 is just the feeder architecture.
This is where having an EE/CE in here would be helpful. I don't think you're wrong, however even with an infinitely fast bus I'm not sure that would work. At some point it comes down to latency, as electrons only move so far in a nanosecond. What AMD wants to do with HSA requires very low latencies (in order to mitigate the penalty of having to context switch from the CPU to the GPU), so I'm not sure even a HT/QPI bus would be fast enough.If it became a real need, a new bus could be made in physical and logical proximity to a CPU to allow these types of transfers (think QPI). You may end up with a socket/slot near the CPU socket, and then a dongle, or motherboard based hardware for the video connection if using the same hardware to drive video.
It would be a far more open and upgradable option than being mated to a franken CPU that is decent at one thing and crap at everything else compared to its competition. You could buy both the right CPU and the right GPU for the task.
PCIe is getting long in the tooth anyway.
Its irrelevant. Their software base is volatile and they tend to code for an upper layer like Java (Not Oracle Java). Samsung for example gonna ship Tizen devices this year with Intel.
Also, what companies are actually making HSA hardware?
Which doesn't make sense.
HSA in-itself is a brilliant idea - very much so in trying to get rid of the x86 monopoly we've had on the computing in general side for 20 years.
But how that translates to AMD - i do not understand.
And no one can explain that.
It's like they're making the groundwork - so others can come benefit with greater execution\R&D budgets once it's actually done.
Well AMD can't do this on their own. They don't have enough market share or enough clout among developers. If they don't bring on partners then this would never take off. So they've essentially formed an anti-Intel coalition.Which doesn't make sense.
HSA in-itself is a brilliant idea - very much so in trying to get rid of the x86 monopoly we've had on the computing in general side for 20 years.
But how that translates to AMD - i do not understand.
And no one can explain that.
It's like they're making the groundwork - so others can come benefit with greater execution\R&D budgets once it's actually done.
That's actually a good comparison. The big difference being that HSA is intended to compile down to native code rather than being executed in a JIT runtime fashion. You wouldn't have a virtual machine like you do with Java.As it is now, HSA looks more to be another Java.
You are certainly not getting rid of x86 without an unacceptable performance penalty