HSA in the 14nm generation - summit ridge with [some] shaders, or, HSA from AI dGPU?

R3MF

[H]ard|Gawd
Joined
Aug 14, 2006
Messages
1,803
Is HSA still AMD's future?

If so, how does this work with high-end Summit Ridge CPU's?

To me there are only two possible choices:
1. Everything is an APU, i.e. Summit-Ridge will come with around 256 shaders to permit HSA operation.
2. PCIe latency is overcome allowing Summit Ridge to work with Arctic Islands dGPU for HSA operation.

Which is it?
 
Last edited:
AM4 will be able to use both FX style (no gpu built in) and APU (gpu built in)
not sure besides this, though I am quite certain because arctic islands as well as Zen in the first place
will both be top to bottom new Uarchs, HSA concepts and decisions were probably heavily used in the design.

Never know what AM4 really brings to the table as pretty much zero has been brought forth yet and we have quite awhile yet to wait until much of this information surfaces, same with Zen, same with Arctis Islands.
 
Not to sure if you understand what HSA is.
http://developer.amd.com/resources/...hat-is-heterogeneous-system-architecture-hsa/

HSA creates an improved processor design that exposes the benefits and capabilities of mainstream programmable compute elements, working together seamlessly. With HSA, applications can create data structures in a single unified address space and can initiate work items on the hardware most appropriate for a given task. Sharing data between compute elements is as simple as sending a pointer. Multiple compute tasks can work on the same coherent memory regions, utilizing barriers and atomic memory operations as needed to maintain data synchronization (just as multi-core CPUs do today).
 
It's neither. HSA is essentially useless (and will be for some time, probably at least 5 years) without large amounts of memory on/near the die(s).
 
I do know the basics of what it is, thanks for the link, but while I could always use more brains, I am decently well versed in anything AMD, not so much for Nvidia or Intel at this point. Thing is ANY AMD product CAN be designed with HSA in its DNA so for example, an AMD processor AND Graphics card both designed with HSA in mind has all the likely ability to just scream performance and efficiency then either on its own, being that AMD is of course the one who started it A and B is the largest partner behind the foundation.

As for the one who said its useless, you OBVIOUSLY do not at all know anything about it, nor do you know all the many different very high end industry leaders backing it, read into it at least a bit before you spout a useless comment such as its useless and will be for at least 5 years without requiring large amount of memory on/near the die, yes memory has a factor in it, but also keep in mind it is also meant for ultra-low power/mobile devices as well both of which have very little amount of memory to begin with.

The foundation to my knowledge was set-up to begin with to use architecture more efficiently no matter the form factor, taken from their page "The HSA Foundation seeks to create applications that seamlessly blend scalar processing on the CPU, parallel processing on the GPU, and optimized processing on the DSP via high bandwidth shared memory access enabling greater application performance at low power consumption. The HSA Foundation is defining key interfaces for parallel computation utilizing CPUs, GPUs, DSPs, and other programmable and fixed-function devices, thus supporting a diverse set of high-level programming languages and creating the next generation in general-purpose computing."
http://www.hsafoundation.com/

Hmm, does this say any where it requires boat loads of memory, NO, it does however point out using things far more effectively, this can be further summed up by ARM is also a massive partner, and generally speaking, most ARM devices are built around the as much performance from as little power and overhead as possible, which in most cases means very little memory as this would take up precious die room, BOM cost, and of course power.

Anyways. HSA probably will be a major design point for Zen, Artic Islands and AM4 in general to get as much efficiency from the cpu-gpu-memory and various sub-systems simple as that.

As for the original question, I am pretty positive HSA is now a huge part of AMD DNA seeing as how much they have invested and built towards multi-thread/task workloads 64bit etc over the last couple of decades, of course they will want to work with major industry partners to get the most out of many core/massive scale designs from software/hardware :D

furthermore, take a look at the HSA Foundation Partners from a Google search, yeh, it is useless, with so very many industry leaders focusing ALL their various resources and designs towards it, surely it is useless "here's your sign"
 
It's neither. HSA is essentially useless (and will be for some time, probably at least 5 years) without large amounts of memory on/near the die(s).
Huh? No. HSA just makes it easier to program GPGPU stuff by simplifying the memory model for sending data to and from the GPU. You don't need huge amounts of on die/package memory to make that work.

Right now its not worth much on the desktop but in the future it should be. Just like with anything that is new in the programming world. Look how long it took (is taking really) for X64 software and OS's to become common place.
 
AMD has been talking about "HSA next year!" for like, ten years now.
 
AMD has been talking about "HSA next year!" for like, ten years now.

except it has actually made it into some software this last year. Even has benchmarks to showcase its strengths. The biggest issue isn't AMD pushing its adoption but the complacency of users that software devs then don't feel the need to incorporate it.
 

To whom is that addressed?
I have a pretty good idea of what HSA is.
It appears to be widely accepted that HSA does not deliver on its promise (at least under current architecture) if there is not a tight coupling of the CPU and GPU with a shared memory allocation, and also affected by latency problems with PCIe.
Hence, i can buy a £300 laptop which supports HSA, but I cannot build a PC that leverages the power of my £600 graphics card and £300 CPU.
Either they sort out this limitation of shared memory pool and latency over PCIe, they stick shaders on Summit Ridge, or HSA has no part to play in their high-end offerings.
Which is it?
 
Last edited:
To whom is that addressed?
I have a pretty good idea of what HSA is.
It appears to be widely accepted that HSA does not deliver on its promise (at least under current architecture) if there is not a tight coupling of the CPU and GPU with a shared memory allocation, and also affected by latency problems with PCIe.
Hence, i can buy a £300 laptop which supports HSA, but I cannot build a PC that leverages the power of my £600 graphics card and £300 CPU.
Either they sort out this limitation of shared memory pool and latency over PCIe, they stick shaders on Summit Ridge, or HSA has no part to play in their high-end offerings.
Which is it?

Think he may have been confused by your original statement. HSA first and foremost required an iGPU to get the efficiency and boost. There has been talk of allowing dGPU but that latency that you are aware of makes such gains minimal. AMD has alluded that much of their primary CPU releases will be APU. Reason being HSA. This is also what drives the premise of multi-adapter which we saw with Ashes of the Singularity, though they didn't test iGPU that I am aware of.
 
Last edited:
^ People are expecting Sandy Bridge performance from 2016 AMD Lineup, i think Intel doesn't even mention AMD in their board meetings.
But if they are able to exceed everyone's expectations and fuck Intel in the ass, that would be great for the market and will easily get them back in the game.
 
To whom is that addressed?
I have a pretty good idea of what HSA is.
It appears to be widely accepted that HSA does not deliver on its promise (at least under current architecture) if there is not a tight coupling of the CPU and GPU with a shared memory allocation, and also affected by latency problems with PCIe.
Hence, i can buy a £300 laptop which supports HSA, but I cannot build a PC that leverages the power of my £600 graphics card and £300 CPU.
Either they sort out this limitation of shared memory pool and latency over PCIe, they stick shaders on Summit Ridge, or HSA has no part to play in their high-end offerings.
Which is it?

And that is why I linked that quote, What are you expecting from a cpu or dedicated gpu to have a design that has an influence on this problem?

The solution in hardware was for the complete (SOC/APU) package to allow the design through hardware to function better without the normal problems.

The HSA team at AMD analyzed the performance of Haar Face Detect, a commonly used multi-stage video analysis algorithm used to identify faces in a video stream. The team compared a CPU/GPU implementation in OpenCL™ against an HSA implementation. The HSA version seamlessly shares data between CPU and GPU, without memory copies or cache flushes because it assigns each part of the workload to the most appropriate processor with minimal dispatch overhead. The net result was a 2.3x relative performance gain at a 2.4x reduced power level*. This level of performance is not possible using only multicore CPU, only GPU, or even combined CPU and GPU with today’s driver model. Just as important, it is done using simple extensions to C++, not a totally different programming model.

HW Configuration

4GB RAM; Windows 7 (64-bit); OpenCL™ 1.1
APU: AMD A10 4600M with Radeon™ HD Graphics
CPU: 4 cores @ 2.3 MHz (turbo 3.2 GHz)
GPU: AMD Radeon HD 7660G, 6 compute units, 685MHz
 
Last edited:
Huh? No. HSA just makes it easier to program GPGPU stuff by simplifying the memory model for sending data to and from the GPU. You don't need huge amounts of on die/package memory to make that work.

No. One of HSA's goals is to unify memory between cpu/gpu (so that no copying has to occur to run GPU accelerated operations), a unified address space is a shitty half-baked solution that came to be because the technology for interconnected cpu/gpu/memory only became viable recently with HBM and the like.
 
a unified address space is a shitty half-baked solution that came to be because the technology for interconnected cpu/gpu/memory only became viable recently with HBM and the like.
LOL dude a unified memory address space is all HSA is about and that is hardly a half baked shitty solution. They were never going to try and pool all or most of the memory on die or through HBM. Its waaay to expensive and will be for years. Communicating with the GPU and coordinating processing between it, the CPU, and main memory is pretty hard and HSA makes it much easier. Almost easy, with a few caveats of course.
 
No. One of HSA's goals is to unify memory between cpu/gpu (so that no copying has to occur to run GPU accelerated operations), a unified address space is a shitty half-baked solution that came to be because the technology for interconnected cpu/gpu/memory only became viable recently with HBM and the like.
A unified address space is what gpgpu developers have wanted for a long time and it's been possible before HBM. HBM doesn't actually change anything anyway as it's just memory sitting behind a memory controller on a PCI device.

One of the more recent developments out of AMD is a c++ compiler that lets you make use of a GPU without demanding that you learn openCL. I'm not sure if it will make use of a GPU if you have a non-hsa system, but in theory it's possible. You can read about it here if you like
http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/p0069r0.pdf
 
A unified address space is what gpgpu developers have wanted for a long time and it's been possible before HBM. HBM doesn't actually change anything anyway as it's just memory sitting behind a memory controller on a PCI device.

One of the more recent developments out of AMD is a c++ compiler that lets you make use of a GPU without demanding that you learn openCL. I'm not sure if it will make use of a GPU if you have a non-hsa system, but in theory it's possible. You can read about it here if you like
http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2015/p0069r0.pdf

Yeah HSA needed HUMA or NUMA (never can remember which or if both) Which existed before HSA. HSA is a way to leverage either of them in a practical way using APUs be them AMD APUs or ARM processors in tablets and phones where low power horsepower is needed and desired.
 
I am going to read this information and enjoy it. That said, HSA from AMD is just a pie in the sky, fat chance of every happening sort of situation. Now, if Intel and NVidia start supporting it, then it will take off even on AMD hardware.
 
And that is why I linked that quote, What are you expecting from a cpu or dedicated gpu to have a design that has an influence on this problem?

The solution in hardware was for the complete (SOC/APU) package to allow the design through hardware to function better without the normal problems.

What am I expecting?

I am expecting to find out whether HSA is [actually] the future of AMD as they have claimed.
If it is the future then we [must] presume it will one day function on their high-end/high-margin products, and not just the $300 Best-Buy boxes from which they scrape a few miserable dollars in profit.

If HSA [is] still the future of AMD then its high-end/high-margin products have two possible solutions to the current problem that i can see:
1. Summit Ridge comes with some shaders. Not many, not a significant proportion of total die space, but a useful number to enable HSA functionality. 256 high density shaders on 14nm would be very acheivable with 8c/16t at 14nm inc L3 cache.
2. They find a way to extend shared memory allocation across PCIe, and look at technical solutions to reduce latency (3.0 was supposed to be better than 2.0, will 4.0 be better still?), and software solutions to mitigate the impact of that latency.

Which is it?

Is it getting any clearer for you now?
 
I am going to read this information and enjoy it. That said, HSA from AMD is just a pie in the sky, fat chance of every happening sort of situation. Now, if Intel and NVidia start supporting it, then it will take off even on AMD hardware.

I don't know man. People thought the same thing about X86-64 and multicore CPUs. AMD was first to come out with them for the mainstream market, it took a few years but Intel followed suit.
 
What am I expecting?

I am expecting to find out whether HSA is [actually] the future of AMD as they have claimed.
If it is the future then we [must] presume it will one day function on their high-end/high-margin products, and not just the $300 Best-Buy boxes from which they scrape a few miserable dollars in profit.

If HSA [is] still the future of AMD then its high-end/high-margin products have two possible solutions to the current problem that i can see:
1. Summit Ridge comes with some shaders. Not many, not a significant proportion of total die space, but a useful number to enable HSA functionality. 256 high density shaders on 14nm would be very acheivable with 8c/16t at 14nm inc L3 cache.
2. They find a way to extend shared memory allocation across PCIe, and look at technical solutions to reduce latency (3.0 was supposed to be better than 2.0, will 4.0 be better still?), and software solutions to mitigate the impact of that latency.

Which is it?

Is it getting any clearer for you now?

Just FYI... Summit Ridge means Zen CPUs, not Zen APUs. So no HSA or GPU for Summit Ridge. Bristol Ridge is Carrizo for desktop (AM4 platform).

Zen APUs = Raven Ridge if the leaks are accurate.
 
What am I expecting?

I am expecting to find out whether HSA is [actually] the future of AMD as they have claimed.
If it is the future then we [must] presume it will one day function on their high-end/high-margin products, and not just the $300 Best-Buy boxes from which they scrape a few miserable dollars in profit.

If HSA [is] still the future of AMD then its high-end/high-margin products have two possible solutions to the current problem that i can see:
1. Summit Ridge comes with some shaders. Not many, not a significant proportion of total die space, but a useful number to enable HSA functionality. 256 high density shaders on 14nm would be very acheivable with 8c/16t at 14nm inc L3 cache.
2. They find a way to extend shared memory allocation across PCIe, and look at technical solutions to reduce latency (3.0 was supposed to be better than 2.0, will 4.0 be better still?), and software solutions to mitigate the impact of that latency.

Which is it?

Is it getting any clearer for you now?

No it is still gibberish.

HSA is a hardware compatible design for integrated gpu. Which needs software for it to "function".
This has nothing to do with your #2.
HSA does not address issues with PCI-E bus.

HSA is the future for any company really not just AMD.
 
I am going to read this information and enjoy it. That said, HSA from AMD is just a pie in the sky, fat chance of every happening sort of situation. Now, if Intel and NVidia start supporting it, then it will take off even on AMD hardware.

You mean you actually do not care about technology until it comes in the form when a 3 year old can understand it, you obviously do not know what it does for hardware in general even if you got an example of what it does for computing rather then trying to understand you point to Nvidia and Intel to support it because of what exactly ?

Intel never cared for anything but their profits and Nvidia is in the same boat why would they bother with better integrated design.

HSA is not about maximizing profits but "better" chip design across all platforms ..
 
You mean you actually do not care about technology until it comes in the form when a 3 year old can understand it, you obviously do not know what it does for hardware in general even if you got an example of what it does for computing rather then trying to understand you point to Nvidia and Intel to support it because of what exactly ?

Intel never cared for anything but their profits and Nvidia is in the same boat why would they bother with better integrated design.

HSA is not about maximizing profits but "better" chip design across all platforms ..
To be fair AMD also "only" cares about their profits, but they're in a position where the best thing they can do for their bottom line is to change the fundamentals of chip design. Intel and Nvidia have market lock in and AMD's best chance to make more profit is to change the market (My opinion here).
 
AMD cares about not losing truck loads of money more then anything else profit is the next step :)
 
AMD cares about not losing truck loads of money more then anything else profit is the next step :)

I think we can agree that they care about money coming in then yes? :)

All I mean to say is that given their situation, AMD have a nice alignment of their own financial incentives with pushing the envelope of system architecture.
 
I think we can agree that they care about money coming in then yes? :)

All I mean to say is that given their situation, AMD have a nice alignment of their own financial incentives with pushing the envelope of system architecture.

Explain where HSA is making money for AMD ?
 
Just FYI... Summit Ridge means Zen CPUs, not Zen APUs. So no HSA or GPU for Summit Ridge. Bristol Ridge is Carrizo for desktop (AM4 platform).

Zen APUs = Raven Ridge if the leaks are accurate.

thank you, i do know that. my question is whether/how forms any function of their high-end/high-margin products.
 
No it is still gibberish.

HSA is a hardware compatible design for integrated gpu. Which needs software for it to "function".
This has nothing to do with your #2.
HSA does not address issues with PCI-E bus.

HSA is the future for any company really not just AMD.

the only thing that is gibberish is the value of HSA to AMD if it forms no part of the high-value / high-margin products.
 
Back
Top