AMD CPU speculation... and expert conjecture

[simon12]

I forgot to add I assume a HSA enabled APU will still be able to use HSA when a dedicated card is installed?

 

[palladin9479]

I believe HSA will completely cut out discrete GPUs except for the highest of high end systems in the next 10 years due to how little we gain from pure graphics hardware these days. The first part for AMD would be to get the hardware out and then the entire ARM consortium will follow. Every soc in the world will eventually use HSA because they will always have a integrated GPU, this allows for higher hardware utilization. If AMD can standardize it and get enough support for it, they can be in a good spot for mobile cpus without discrete gpu parts. I doubt the GPU compute units will be integrated into the FPU any time soon. They would need both for the foreseeable future. What HSA does is make it easy to utilize the GPU portion of the APU so the programmer can choose to not use the FPU where the GPU cores will be faster. Once HSA become the common place, AMD can eventually do away with all but 1 FPU in the CPU, which would be purely used for legacy support. I don't believe GPU accessible registers that can process SIMD instructions will be in the works for a while. You will still need legacy support on FPU instructions even if you can do that simply because of scheduling to the GPU cores will get you conflicts and you can't always resolve these with pipelinable instructions on the CPU side.

Kaveri having huma is a big thing but also its not. In the PC front I don't see it coming to great use for a couple of years. The hardware has to be out before the software. Having a coherent memory across all heterogeneous processors is required but will not be utilized at this point. Simple tweaks to compilers should start make use of it eventually. I would like to see some good GPGPU performance from it but I doubt its a game changer yet. We will have to wait and see.

They won't be replacing dGPU's ever. This is physics speaking, it's impossible for an iGPU to be as powerful as a dGPU due to space / cost / electricity limits. It's like saying mopeds or "economy" cars will replace consumer transportation. Gamers and their ilk will always want the powerful dGPU to provide the best experience, the iGPU is there for budget / economy systems.

Also GPU "compute units" are "FPU's". They do the exact same math with the only difference being their organized into massive arrays.

 

[palladin9479]

I still don't really understand what HSA & HUMA will do in practical ways, are the following statements correct (Isuspect not in many cases):
HSA will only boost performance in applications that could be accelerated by a GPU
HSA will add no boost to performance in programs not optimized for it
Also some questions:
Can HSA be used by a dedicated GPU to increase performance using a dedicated card and no APU?
If so could it use current hardware ie a FX6300 + a Radeon 7770?
If not could it boost performance with a next gen GPU ie FX6300 + Radeon 8xxx or 9xxx?
If either of the above 2 can help is it possible we may see systems using a dedicated HSA card ie a FX6300 + a Radeon 8750 for HSA + a Radeon 8950 for the rest of the game.
Is there any possibility of HSA being used in Intel or Nvidia hardware in the future?
Do you think HSA will be a big deal in a gaming PC in a years time?
Hope this makes sense.

People here have been throwing marketing terms around and all it did was muddy the waters. "HSA" isn't a processor type, it's just a set of standards that allow two non-uniform (as in their not the same type of processor) processors to cohabit the same real estate and communicate with each other without requiring expensive glue circuitry. "HUMA" is just the acronym invented to say that the GPU will use the same physical address map as the primary CPU. HUMA is part of the "HSA" concept / standard. This will allow vector instructions (SIMD basically) to be rapidly processed by the iGPU without needing them to be packaged and shipped off in the manor that OpenCL / GPGPU does. It reduces latency and the overhead required when doing this.

This will not directly speed up a dGPU in any sense but what it can do is allow the video game to offload physics and AI calculated to the iGPU while using the dGPU just for rendering.

For everyone else, please read up on what NUMA is and it's role in the multiprocessing world. "HUMA" is just the inverse of NUMA, multiple processors sharing the same physical memory space vs each having their own dedicated memory space.

 

[juanrga]

). It is much easy to scale FPU performance in a GPU (or similar compute unit). I already provided a graphics that shows the increase in floating point performance in GPUs compared to CPUs for a span of several years.

Ask gamerk on this one about scaling to 12+ cores with 4 CPU cores running @ 3.5ghz and 8+CU @ ~1ghz.

Seeing this massive improvement will need massive software. Everything today won't see much benefit at all.

IF all you claim is true, AMD is banking on this APU when at release will be slower than BD. 2-4 years from now it might look ok but what reason is there to upgrade now? just wait the 2-4 years for the next big thing because upgrading now will be a downgrade.

With releasing only a 4 core APU, everyone that has PII x6, or FX 6xxx-8xxx will see no reason to upgrade, especially considering most FM2+ motherboards would be a downgrade.

This is why if AMD abandons the AM3+ completely will piss off any loyal customers.

The other side of the coin is AMD needs to get this hardware out there so that the software can come.

In the mean time they need to support both mainstram and high end, and not just pushing low end parts that are designed for "tomorrow's software"

Just in the same post that you are replying I wrote:

Evidently, HSA will be only effective in situations where the software/code is compiled for it. The term "HSA enabled software" has been used before. I have recalled several times how developers are finding up to 5x performance gains when HSA is enabled. Evidently, if the software ignores HSA, then we will see no gain.

I don't understand why this point about the software is being repeated forever in this thread. Some of Haswell improvements require software to be recompiled for it. However, I don't see the "only if software is compiled for it" repeated forever in talks about Haswell.

I believed it was made clear in many occasions that HSA harwdware requires HSA software to see the 'massive' benefits. Therefore, I continue without understanding why the software issue is mentioned again and again and again and again.

About upgrading. Take a look to server roadmaps. I already gave my estimations on Steamroller performance. Others seem to agree:

That should put the Berlin family ahead of the quad-core Opteron 3300s, but likely behind the eight-core Opteron 3380.

http://www.extremetech.com/computing/158901-amd-server-assault-2014-roadmap

Berlin shines with HSA software, because then has the potential to provide up to a 8x the gigaflops per-watt compared to Opteron.

The CPU in Kaveri would be close to the level of a FX-8100 (octo-core BD).

About software:

- Servers? AMD has shown HSA web server applications that improve server performance compared to traditional CPU based server.

- PCs? LibreOffice will come with HSA enabled in spreadsheets. Wait HSA enabled in other applications.

- Games? At least two popular game engines (unreal and Hawock) have already offloaded physics simulations to the GPU.

 

[juanrga]

I believe HSA will completely cut out discrete GPUs except for the highest of high end systems in the next 10 years due to how little we gain from pure graphics hardware these days. The first part for AMD would be to get the hardware out and then the entire ARM consortium will follow. Every soc in the world will eventually use HSA because they will always have a integrated GPU, this allows for higher hardware utilization. If AMD can standardize it and get enough support for it, they can be in a good spot for mobile cpus without discrete gpu parts. I doubt the GPU compute units will be integrated into the FPU any time soon. They would need both for the foreseeable future. What HSA does is make it easy to utilize the GPU portion of the APU so the programmer can choose to not use the FPU where the GPU cores will be faster. Once HSA become the common place, AMD can eventually do away with all but 1 FPU in the CPU, which would be purely used for legacy support. I don't believe GPU accessible registers that can process SIMD instructions will be in the works for a while. You will still need legacy support on FPU instructions even if you can do that simply because of scheduling to the GPU cores will get you conflicts and you can't always resolve these with pipelinable instructions on the CPU side.

Kaveri having huma is a big thing but also its not. In the PC front I don't see it coming to great use for a couple of years. The hardware has to be out before the software. Having a coherent memory across all heterogeneous processors is required but will not be utilized at this point. Simple tweaks to compilers should start make use of it eventually. I would like to see some good GPGPU performance from it but I doubt its a game changer yet. We will have to wait and see.

They won't be replacing dGPU's ever. This is physics speaking, it's impossible for an iGPU to be as powerful as a dGPU due to space / cost / electricity limits. It's like saying mopeds or "economy" cars will replace consumer transportation. Gamers and their ilk will always want the powerful dGPU to provide the best experience, the iGPU is there for budget / economy systems.

He said "except for the highest of high end systems". I agree entirely with him. Yes an iGPU has space limitations, but a dGPU has limitations as well, such as the slow bus that connects it to the CPU or the lack of hUMA.

Consoles prove how a iGPU can compete with much more expensive systems with a dGPU. Not everything is brute force. The customized Radeon hardware in the PS4 offers performance beyond top high-end dGPUs for some workloads.

 

[de5_Roy]

^^ lol. even after all the explanations.....

"Nothing official but a lot of info is hinted at"
http://semiaccurate.com/2013/08/22/gigabyte-goes-kaveri-crazy-with-seven-new-boards/

 

[palladin9479]

^^ lol. even after all the explanations.....

"Nothing official but a lot of info is hinted at"
http://semiaccurate.com/2013/08/22/gigabyte-goes-kaveri-crazy-with-seven-new-boards/

Yeah FM2+ will most likely become the new "high end" socket, or rather mainstream ~$200 USD per CPU socket.

I find it funny that anything with a dGPU is suddenly "the highest end system". Like the 660 / 760 never existed or something.

 

[juanrga]

I don't know from where comes the myth that gamers only play with the most powerful dGPUs.

Statistics show that most powerful dGPU are essentially marketing cards. How many gamers have a HD7990? A 0.1%?

Moreover, the trend is towards APUs and iGPUs

http://www.digitaltrends.com/computing/discrete-gpus-see-biggest-drop-in-shipment-since-2008/

Unfortunately for AMD and Nvidia, the decreased demand for dedicated graphics is even more dramatic compared to numbers in the past 10 years.

[...]

For those of you shopping for computers with all the graphics horsepower that AMD and Nvidia can deliver, the selection of PCs with any discrete GPU probably won’t improve as these numbers will hardly encourage computer manufacturers to introduce more computers with high-end graphics. It’s too bad because as discrete GPUs become more niche, it will only drive up the cost for consumers.

 

[hcl123]

It will be a disaster if an FM2+ cost ~$200 USD. Where i live an top of top FM2 costs ~$150, with A85. It will come considerably down after FM2+ (A88) appears.

That will be too high for mainstream...and this should be exclusively about mainstream. Ppl are only making too much suppositions out of the noise.

 

[CooLWoLF]

If AMD does announce something like a new AM4 socket line, what is the likely hood that the new AM4 CPUs are backwards compatible with the AM3+ line of motherboards?

 

[juanrga]

I still don't really understand what HSA & HUMA will do in practical ways, are the following statements correct (Isuspect not in many cases):
HSA will only boost performance in applications that could be accelerated by a GPU
HSA will add no boost to performance in programs not optimized for it
Also some questions:
Can HSA be used by a dedicated GPU to increase performance using a dedicated card and no APU?
If so could it use current hardware ie a FX6300 + a Radeon 7770?
If not could it boost performance with a next gen GPU ie FX6300 + Radeon 8xxx or 9xxx?
If either of the above 2 can help is it possible we may see systems using a dedicated HSA card ie a FX6300 + a Radeon 8750 for HSA + a Radeon 8950 for the rest of the game.
Is there any possibility of HSA being used in Intel or Nvidia hardware in the future?
Do you think HSA will be a big deal in a gaming PC in a years time?
Hope this makes sense.

HSA is a new architecture that allows compute units of different kinds (e.g. CPUs and GPUs) work together in the fastest and simplest way. The new hardware eliminates bottlenecks of the traditional architecture used in PCs. And the new software tools make programmers' work easy.

hUMA is a new memory model that is part of HSA. This model provides unified memory and allows the different compute units to share data more efficiently and even to work at the same time on the same data. Current PCs don't have unified memory but the memory of the GPU and the memory of the CPU.

HSA requires new hardware and software. You need a HSA enabled CPU, a HSA enabled GPU, and HSA enabled applications.

I hope this answer all your questions. For instance the requirement of HSA enabled hardware means that HSA enabled applications will not boost a FX6300 + a Radeon 7770

Many of us believe that HSA is the next big thing.

 

[de5_Roy]

after the new consoles come out and 4k gains more traction, current crop of $250~ gaming gfx cards will have a hard time. especially the ones that launched at around $250. current high end will be the new midrange( in terms of launch prices) because the ongoing stagnation of demanding gfx performance(pc gaming) is about to unravel (or dissolve or disengage etc..) very soon. i am pretty sure of those singing radeon 7950 or gtx770 praise will change their tune next year.

 

[hcl123]

People here have been throwing marketing terms around and all it did was muddy the waters. "HSA" isn't a processor type, it's just a set of standards that allow two non-uniform (as in their not the same type of processor) processors to cohabit the same real estate and communicate with each other without requiring expensive glue circuitry. "HUMA" is just the acronym invented to say that the GPU will use the same physical address map as the primary CPU. HUMA is part of the "HSA" concept / standard. This will allow vector instructions (SIMD basically) to be rapidly processed by the iGPU without needing them to be packaged and shipped off in the manor that OpenCL / GPGPU does. It reduces latency and the overhead required when doing this.

First HSA programmers reference manual is out (pdf in)
http://hsafoundation.com/standards/

Its a JIT approach alright, but with a dynamic runtime well suited for flexible apps... that wiil have several code paths attending the resources available

http://hsafoundation.com/hsa-programer-reference-the-formation-of-the-new-specification/

That will have its Hardware platform specification to (which will be minimal i think) and is basically 4 definitions or items.

In Development

HSA Platform Systems Architecture Specification
version 1.0 – In the working group
*includes IOMMU v2.5 official standard
* most probably will include hUMA (*cache coherency* + IOMMU makes hUMA, like cache coherency + same MMU/TLB makes NUMA)
*includes hardware task queues visible from userspace.

HSA Programmer’s Reference Manual: HSAIL Virtual ISA and Programming Model, Compiler Writer’s Guide, and Object Format (BRIG)
0.95 - Up for ratification in the Board of Directors
1.0 In working group

HSA System Runtime Specification
Draft Specification in working group for development

HSA Tools
LLVM to HSAIL Compiler
HSAIL Assembler

So its mostly "software" alright but there is a Hardware platform specification, not a processor uarch definition. That hardware definition is exactly what is needed for making diverse "processor element uarchs" act like an unified entity. Based that the gluing *magic* will be HSAIL not abstraction APIs. It will be more low level it will be considerably more flexible, allowing software developers to innovate. And hardly will be only about GPU, otherwise no TI or other DSP implementer will be there.

This will not directly speed up a dGPU in any sense but what it can do is allow the video game to offload physics and AI calculated to the iGPU while using the dGPU just for rendering.

That is exactly what it can do. The runtime environment(+ drivers) will be king saying where any piece of code runs, not the developer, that will code "exactly" has if everything is an single processor. It could be grand for beefing up tremendously the FPU parts in CPUs (XOP, AVX and or NEON) and start using ray-tracing more extensively (probably mixed with raster... and Imagination already has good ray-tracing boards based on ARM and their PowerVR)

For everyone else, please read up on what NUMA is and it's role in the multiprocessing world. "HUMA" is just the inverse of NUMA, multiple processors sharing the same physical memory space vs each having their own dedicated memory space.

Skewed exactly to the contrary. In NUMA the MMU/TLB part of the processing elements must be identical for not saying the same. In hUMA those MMU/TLB parts can diverge(edt), but they must adhere to the IOMMU standard. NUMA also presupposes that several DRAM pools are physically apart, and while that is true, there could be NUMA assigning unganged DRAM channels to different processor elements, only doesn't make much sense.

hUMA presupposes that there is only 1 DRAM pool, though this is the most straightforward, that is not necessarily true if there is *IOMMU complaint* and *cache coherency* over some interconnect for disparate "processing elements". And there could be several topologies, not only different boards over interconnect(like PCIe boards, but the problem is PCIe doesn't allow cache coherency... but Hypertrasnport and their HTX slots can, start to see intel cold sweats, and nvidia smell of rotten lol ? (edt) ), but also soldered to a PCB, inside a socket like MCM parts, or even 2.5/3D interposed.

[UPDATE 1 : and its not only Hypertransport (though this is what could impact more the game part of PC world, intel could have identical now if there were QPI slots), ARM AMB interconnect allows cache coherency over subtract, and even PoP (package on package) or 2.5/3D interposed parts... start to see the interest of ARM ? ]

[UPDATE 2 : any interconnect and the cache coherency mechanism "internals", is not and will not ever be part of any HSA standard... there is no need to... the only that matters is that is cache coherent and that it can work with the IOMMU guide lines ]

 

[palladin9479]

Obfuscation + smoke + mirrors + hand waiving.

Need I remind everyone that Intel and not AMD has the vast majority of the x86 CPU market. Developers will not suddenly start using a different compiler that produces code that doesn't work on their customers systems. So lets separate the PC market from the upcoming consoles. Everything I've posted is in regards to the PC market while I specifically said that consoles, being coded to the bare metal, could take full advantage of whatever special hardware is present.

You might as well post a link to the OpenGL working group stating that OpenGL is the next greatest graphics API and will drive the market of the future. Then we can all laugh at you.

And BTW none of the above actually disagrees with what I said.

 

[noob2222]

[
I believed it was made clear in many occasions that HSA harwdware requires HSA software to see the 'massive' benefits. Therefore, I continue without understanding why the software issue is mentioned again and again and again and again.

About upgrading. Take a look to server roadmaps. I already gave my estimations on Steamroller performance. Others seem to agree:
...

The CPU in Kaveri would be close to the level of a FX-8100 (octo-core BD).

because statements like this are repeated again and again and again and again without mentioning that this is only with "HSA software aware programs" wich won't be available

released today or in december ... hardly

2-3 years from now when excavator is out, maybe.

...

I don't know from where comes the myth that gamers only play with the most powerful dGPUs.

Statistics show that most powerful dGPU are essentially marketing cards. How many gamers have a HD7990? A 0.1%?

Moreover, the trend is towards APUs and iGPUs

http://www.digitaltrends.com/computing/discrete-gpus-se...

"Integrated graphics is good enough for most computer users who don’t want to spend a ton of money but want to be able to watch movies and play games like The Elder Scrolls V: Skyrim at the lowest setting.

"

lowest settings? no thanks. might as well play at 640X480 on a 1080p supported monitor or buy a console if thats all your going to be doing. maybe you didn't notice that referring to consoles, they usually have the lowest graphics settings.

Most "gamers" even on steam survey are using geforce ~560 ti cards. in the top 10 on steam survey of gpus are nvidia 560ti, 550ti, 460, 560, 670. 660. Id say thats hardly "the most powerful gpu".

The remaining 4 are all intel igp. What you don't know is what games they are trying to play. How many grandmas and grandpas are on steam playing just the games that came with their newly purchased Intel system that came with "steam"?

how many of the ~5% of mac systems are using Intel hd graphics?

Another interesting tidbit of information on their survey, 21% of the systems involved are 1366x768 screens ... aka LAPTOPS.

how reliable is the information regarding Intel HD gpus without knowing that?

still, 21% of the systems tested or more are laptops, ~5% are macs, and still only 14% of the total gpus are "intel", so how many "gamers" are using integrated graphics?

 

[juanrga]

Based that the gluing *magic* will be HSAIL not abstraction APIs. It will be more low level it will be considerably more flexible, allowing software developers to innovate. And hardly will be only about GPU, otherwise no TI or other DSP implementer will be there.

As I said before, HSA is about two generic kind of compute units: LCUs and TCUs.

A HSA GPU is an example of TCU, but HSA also considers others such as DSPs. In fact, if my memory does not fail, one of the members of the HSA foundation is developing a HSA DPS.

Yes, the magic starts at low level. E.g. a LCU is a generalized CPU (or HSA enabled CPU) that supports not only the CPU native instruction set but also the new HSAIL instruction set.

hUMA presupposes that there is only 1 DRAM pool, though this is the most straightforward, that is not necessarily true if there is *IOMMU complaint* and *cache coherency* over some interconnect for disparate "processing elements". And there could be several topologies, not only different boards over interconnect(like PCIe boards, but the problem is PCIe doesn't allow cache coherency... but Hypertrasnport and their HTX slots can, start to see intel cold sweats, and nvidia smell of rotten lol ? (edt) ), but also soldered to a PCB, inside a socket like MCM parts, or even 2.5/3D interposed.

[UPDATE 1 : and its not only Hypertransport (though this is what could impact more the game part of PC world, intel could have identical now if there were QPI slots), ARM AMB interconnect allows cache coherency over subtract, and even PoP (package on package) or 2.5/3D interposed parts... start to see the interest of ARM ? ]

[UPDATE 2 : any interconnect and the cache coherency mechanism "internals", is not and will not ever be part of any HSA standard... there is no need to... the only that matters is that is cache coherent and that it can work with the IOMMU guide lines ]

HSA hardware specification requires a HMMU that can communicate with both coherent and non-coherent memory.

However, I have one doubt about Hypertransport/HTX. Didn't this break hUMA? Logically memory would be uniform, but physically the system would consist of two kind of memories due to speed: a faster pool (physical dimms on the mainboard) and a slower pool (accessed through the HTX slot)? What is your thought?

Obfuscation + smoke + mirrors + hand waiving.

Need I remind everyone that Intel and not AMD has the vast majority of the x86 CPU market. Developers will not suddenly start using a different compiler that produces code that doesn't work on their customers systems. So lets separate the PC market from the upcoming consoles. Everything I've posted is in regards to the PC market while I specifically said that consoles, being coded to the bare metal, could take full advantage of whatever special hardware is present.

You might as well post a link to the OpenGL working group stating that OpenGL is the next greatest graphics API and will drive the market of the future. Then we can all laugh at you.

And BTW none of the above actually disagrees with what I said.

First, there are about 2.8 PC gamers that use Intel for each one that uses AMD. However, in the console space the 100% of gamers will use AMD. I recall reading in some place to some game developer claiming that their games will play better in AMD PCs than in Intel PCs.

Second, nobody said that all developers suddenly will start producing code that doesn't work in Intel PCs.

Some developers will ignore HSA and will continue selling ancient software that only improves about a 10% for each new generation of Intel CPUs. However, other developers will produce new HSA software which will runs up to 5x faster in the new AMD hardware.

I already provided a link to the news reporting AMD joining to the Document Foundation Advisory Board to accelerate LibreOffice. I will quote again:

It is great to work on LibreOffice with The Document Foundation to expose the raw power of AMD GPUs and APUs, initially to spreadsheet users,” said Manju Hegde, corporate vice president, Heterogeneous Solutions at AMD. “Bringing the parallelism and performance of our technology to traditional, mainstream business software users will be a welcome innovation for heavy duty spreadsheet users, particularly when combined with the compute capabilities of the upcoming generation of AMD Heterogeneous System Architecture (HSA) based products.

It is exciting to work together with AMD and their ecosystem to take advantage of AMD’s cutting edge innovation right inside LibreOffice,” said Michael Meeks, SUSE Distinguished Engineer and TDF Board Member, “The growth in performance and parallelism available in the GPUs of today, and particularly with AMD’s revolutionary APUs of tomorrow, is something we’re eager to expose to LibreOffice users.

The idea of that one needs a console to take advantage of special hardware is incorrect. One can take advantage of special hardware in a PC. One user here already explained how recompiling software for his AMD hardware he got up to the double of performance.

 

[blackkstar]

I don't see AMD abandoning the market served by FX 6000, 8000, and 9000 series. They are already on the heels of Intel quads with HT in multi-thread, and if you added 20% increase to that Intel quads with HT would be in serious trouble.

http://www.tomshardware.com/reviews/core-i7-4770k-haswell-review,3521-17.html

You can do some math there, but an FX with 20% more performance would do some major hurt.

You're looking at 100fps for Handbrake, 67 seconds for iTunes, very close for the last one.

Look at benchmarks like BF3 which are multi-threaded, even 10% increase in performance would put AMD well into the lead. Considering they are pushing so hard for gaming, I don't see them giving up on it. As I've said, what do you think AMD is going to do? Here is our gaming platform, it features Radeon Memory, Radeon Graphics Card, Radeon SSD, and an Intel CPU? Look at our Gaming Evolved program powered by Intel? We realize we are powering the two big consoles of the next gen, but if you're a PC gamer you're best with an AMD GPU and Intel CPU?

None of that make sense, that's about the worst thing you can do for marketing. It's like AMD coming along and saying we're not going to sell GPUs anymore, FX works best with Nvidia products.

I don't think we're going to see more AM3+ CPUs. AMD is waiting, I think, to give us a bigger upgrade along the lines of DDR3/4 compatibility, PCIe 3.0, HTX slots as mentioned earlier, etc. I wouldn't be surprised if we saw something along the lines of Steamroller FX type CPUs being pushed off until mid 2014, and then AMD just shelving the idea entirely and waiting for DDR4 and a new socket/chipset.

I also wouldn't be surprised by this at all. AMD would be focusing more on mobile parts, the markets served by APUs would get the new improvements first, then the desktop would follow. It would be much like Intel does with LGA2011 lagging behind in architecture.

However I do not think steam hardware surveys are enough to extrapolate if a company is going to stop making chips. 8 core CPUs have .29% marketshare according to SHWS. There are 9 options there with market share. Steam peaks at about 5.5 million users at a time. .29% of that is 159,500 8 core CPU users on at a given time.

As for high end graphics cards.

GTX 690, Titan, 7990, don't even show up on the list as it looks like SHWS cuts things off at <.30% (ED: forgot decimal) for graphics cards. So Nvidia and AMD flagship graphics card products sell at best the same as FX 8 core and at worst far worse.

I wish there were a way to weed out hex core Intels from FX 6000 series, but if FX 6000 is doing strongly on SHWS, it's very well possible that FX line could have more marketshare than 7970 and 7950 combined, and AMD GPUs are supposed to be their strong point. It may be because they can charge more for them, but if AMD could actually beat Intel quads in multi-thread games, and it should do very well against them given the speculation on SR improvements, I see no reason why AMD couldn't manage to charge $300 or so for 8 thread SR chips. THat's just about how much AMD charges for an entire graphics card with 7970 on it, and the FX dies are smaller and it's just a die on a small package, not an entire board.

 

[hcl123]

Tough anything is possible, i think much of the hysteria about the end of so called "big cores" and AM3+ and FX, comes exactly from pieces like this... i call it less attentive journalism, or anyone sees what it likes to see that fits better his official "sensationalist" mantra.

http://www.extremetech.com/computing/158901-amd-server-assault-2014-roadmap/2

http://www.extremetech.com/wp-content/uploads/2013/06/Opteron-Warsaw-640x353.jpg

Warsaw could be Piledriver... nevertheless will be Piledriver on a NEW socket. I think the author of that piece went bezerc with the statement of "Seamless migration" and "fully compatible socket and identical software certifications"...

But "fully compatible" is NOT the same socket, otherwise the word "compatible" doesn't make sense. And "and from Opteron(tm) 6300 family", to me means that the 6300 can be used on the new platforms "and" the new Warsaw also... not unlike AM3 CPUs on AM3+, the contrary, AM3 CPUs can be used on a AM3+ platform, its a "seamless migration with identical if not the same certification for most everything" *and* the new AM3+ CPUs can also be used...

But then Piledriver on a "new" server socket !?... and what socket will be that ?... and notorious enough better perf/watt, when its 32nm PD-SOI ? ... SOI is usually better at clock, so this "seems" to mean higher clocks for the same watt. But them if is 32nm SOI why a new socket ? ... if it is a new fab process why not Steamroller ?

Other news speculated an interesting new name... GC36
http://translate.google.com/translate?langpair=auto|en&u=http%3A%2F%2Ftweakers.net%2Fnieuws%2F82803%2Famd-voorziet-boulder-serverchip-in-2014-van-20-steamroller-cores.html

[ http://tweakers.net/nieuws/82803/amd-voorziet-boulder-serverchip-in-2014-van-20-steamroller-cores.html .. again, the original German link, TH breaks google translation links ]

New Socket GC36
Larger cache support
Maximum of 20 cores/24MB-L3 in MCM (Multi-Chip-Module)
DDR4, LR-DIMM and Power-Efficient DDR3 support
Integrated PCI Express 3.0
HyperTransport 4.0 (forth generation )
Quadchannel memory support

'Boulder' (the code name used) first to appear in 2014...

Seems clearly a replacement for G34... and the fact that is GC... seems to indicate is a replacement for C32 to, perhaps it could use single die chips with only 2 "channels" DRAM, or 3 or 4... yet could be single die chips. And is LGA (better for power and clock), and could be perfectly compatible with past G34 chips, only new GC36 parts will not be possible to fit in G34.

There is no technical impediments for any of this. And with it AMD will have a *single socket* for anything traditional server...as well FX ...

And the support for "Integrated PCI Express 3.0" and "HyperTransport 4.0 (forth generation )" is exactly what will be needed to have combo PCIe +HTX slots, and hUMA with discrete GPGPU adapters, serves enthusiast, serves professional, serves HPC, serves server... all from the same socket, which seems position to last long for a Open Platform definition that seems position to last also.

It will be more expensive since the socket like G34 will be quite larger, but no impediment here to fit even uATX boards. Other form factors there is FM sockets.

I know... in the end AMD is trowing the towel and no more FX etc, seems a very nice and easy prediction, and nobody will trow rotten eggs if wrong, OTOH i'm taking a lot of risk lol

 

[hcl123]

hUMA presupposes that there is only 1 DRAM pool, though this is the most straightforward, that is not necessarily true if there is *IOMMU complaint* and *cache coherency* over some interconnect for disparate "processing elements". And there could be several topologies, not only different boards over interconnect(like PCIe boards, but the problem is PCIe doesn't allow cache coherency... but Hypertrasnport and their HTX slots can, start to see intel cold sweats, and nvidia smell of rotten lol ? (edt) ), but also soldered to a PCB, inside a socket like MCM parts, or even 2.5/3D interposed.

[UPDATE 1 : and its not only Hypertransport (though this is what could impact more the game part of PC world, intel could have identical now if there were QPI slots), ARM AMB interconnect allows cache coherency over subtract, and even PoP (package on package) or 2.5/3D interposed parts... start to see the interest of ARM ? ]

[UPDATE 2 : any interconnect and the cache coherency mechanism "internals", is not and will not ever be part of any HSA standard... there is no need to... the only that matters is that is cache coherent and that it can work with the IOMMU guide lines ]

HSA hardware specification requires a HMMU that can communicate with both coherent and non-coherent memory.

However, I have one doubt about Hypertransport/HTX. Didn't this break hUMA? Logically memory would be uniform, but physically the system would consist of two kind of memories due to speed: a faster pool (physical dimms on the mainboard) and a slower pool (accessed through the HTX slot)? What is your thought?

hUMA is not an HSA official standard. There is no need to mention it. For having hUMA it only needs to mention "cache coherency" is needed, without any other definition or mechanism or software, since "cache coherency" is a purely hardware mechanism and a lot of different implementers can have quite different mechanisms with different coherency protocols. So HSA only needs to mention CC is needed, IOMMU definition will do the rest, and then they will have hUMA.

[ UPDATE: not exactly, cache coherency can be implemented in software, that is , a reference mechanism "emulating" what an hardware implementation would do. And i think that will be case for the *Runtime*, it will have a reference mechanism for all those combinations of CPU + acceleratores that don't have an "hardware" mechanism... it could be clever enough that even a CPU +GPU can be cache coherent by hardware, yet a DSP that is not, will be also cache coherent by software, that is, clever enough to have multi configurations when only a part of those is cache coherent by hardware. IT could use PCIe boards as example (on intel), only software can have an appreciable overhead. But there isn't a Runtime definition yet, its in development... but that deosn't mean squat, only AMD is ahead of the curve, which is no strange since most of it is based on AMD tech anyway]

HT/HTX breaks hUMA as much as Kaveri with dedicated (rumor of kaveri with DDR3 + GDDR5(faster different)) memory for the GPU.

Being hUMA doesn't mean everything uniform or equal or tight physically integrated (though this is the simplest and more straightforward implementation)... the contrary, the letter "h" means "heterogeneous" a priori presupposes differences. NUMA also doesn't require equal DRAM pools for everything, Torrenza examples had some "accelerators" as co-processors on AMD platforms, with quite different memories and sizes.

[UPDATE: hUMA and NUMA are very similar. Both try to give a "single" view of the memory a system has.

While NUMA works with different physically addressable pools as if cascaded/hierarchical, and so each "processor" sees the memory in a different processor as its own, but is closed tied to a particular coherency mechanism and to physical addressing.

hUMA works similar but with "virtual memory" addressing. By ways of IOMMU definitions, it loses the close tightness with any particular coherency protocol, and so broadens the possibilities of having cache coherency between different processors with different mechanisms... its more about "memory coherency". In the end the result is the same, several processor elements will see the whole memory a system can have in a unified way.

That is how i see it, any correction is welcome... but all in all hUMA is not the reverse of NUMA, which many seem to interpret ]

 

[juanrga]

[
I believed it was made clear in many occasions that HSA harwdware requires HSA software to see the 'massive' benefits. Therefore, I continue without understanding why the software issue is mentioned again and again and again and again.

About upgrading. Take a look to server roadmaps. I already gave my estimations on Steamroller performance. Others seem to agree:
...

The CPU in Kaveri would be close to the level of a FX-8100 (octo-core BD).

because statements like this are repeated again and again and again and again without mentioning that this is only with "HSA software aware programs" wich won't be available

No. My above claim is for the CPU running an ordinary CPU benchmark suite for windows 7. That is why I said CPU and didn't mention HSA. In fact, I have a Linux Himeno benchmark where the CPU in Kaveri will be faster than a FX-8150. Again this is not about HSA.

HSA enabled software will not be run by the CPU alone but by the APU (CPU+GPU) as a whole. Using HSA enabled software developers are finding the APU is up to 5x faster than a CPU. 5x the performance implies that the APU will be faster than a hypothetical dual FX-9370.

I.e.:
Ordinary software ~ FX-8100
HSA software >> dual FX-9370

Before someone misinterpret me again: in some ordinary workloads a FX-8100 will be faster than Kaveri. Moreover, not all the HSA workloads will improve performance by a 5x. I considered only an optimal case.

I don't know from where comes the myth that gamers only play with the most powerful dGPUs.

Statistics show that most powerful dGPU are essentially marketing cards. How many gamers have a HD7990? A 0.1%?

Moreover, the trend is towards APUs and iGPUs

http://www.digitaltrends.com/computing/discrete-gpus-se...

"Integrated graphics is good enough for most computer users who don’t want to spend a ton of money but want to be able to watch movies and play games like The Elder Scrolls V: Skyrim at the lowest setting.

"

lowest settings? no thanks. might as well play at 640X480 on a 1080p supported monitor or buy a console if thats all your going to be doing. maybe you didn't notice that referring to consoles, they usually have the lowest graphics settings.

Most "gamers" even on steam survey are using geforce ~560 ti cards. in the top 10 on steam survey of gpus are nvidia 560ti, 550ti, 460, 560, 670. 660. Id say thats hardly "the most powerful gpu".

The remaining 4 are all intel igp. What you don't know is what games they are trying to play. How many grandmas and grandpas are on steam playing just the games that came with their newly purchased Intel system that came with "steam"?

how many of the ~5% of mac systems are using Intel hd graphics?

Another interesting tidbit of information on their survey, 21% of the systems involved are 1366x768 screens ... aka LAPTOPS.

how reliable is the information regarding Intel HD gpus without knowing that?

still, 21% of the systems tested or more are laptops, ~5% are macs, and still only 14% of the total gpus are "intel", so how many "gamers" are using integrated graphics?

Again I don't know what are you reading, because the research made by Jon Peddie firm is not about consoles but about PCs.

About Steam.

560ti + 550ti + 460 + 560 + 670 + 660: 11,27%.

"The remaining 4 are all intel igp.": 11.5%

My point remain: How many gamers have a HD7990? 0.1%?

You cannot redefine gamer to be only what you want it to be. Gamer is not only the young who is playing Crysis 3 on max settings in a dual top-end GPU config with OC CPU. That is only a ~0.1% of the market.

Moreover, the worldwide trend is moving to mobile. That is why Intel developed Haswell with laptops in mind and didn't care about the desktop. AMD is also focusing on the mobile.

 

[hcl123]

Look at benchmarks like BF3 which are multi-threaded, even 10% increase in performance would put AMD well into the lead. Considering they are pushing so hard for gaming, I don't see them giving up on it. As I've said, what do you think AMD is going to do? Here is our gaming platform, it features Radeon Memory, Radeon Graphics Card, Radeon SSD, and an Intel CPU? Look at our Gaming Evolved program powered by Intel? We realize we are powering the two big consoles of the next gen, but if you're a PC gamer you're best with an AMD GPU and Intel CPU?

None of that make sense, that's about the worst thing you can do for marketing. It's like AMD coming along and saying we're not going to sell GPUs anymore, FX works best with Nvidia products.

But that is how the wold is... doesn't make sense. The problem is that most of those IDMs are afraid of the "propaganda cloth" and "market twisting prowess" intel has (most included AMD). Remember the famous AMD lawsuit against Intel, and the "exclusivity deals" intel imposed most particularly against AMD server ? .. that is, if a vendor wants juicy intel deals, then they have to abandon any other competitor...

Everything went quite worst since then. No wonder the best TCO and performance AMD server had for some time has never picked any steam, the contrary went almost extinct. Once a thief always a thief, and the same entity with the same leadership, bent on the same methods and ideology, the contrary would be the "strange". That is why AMD went Seamicro, doesn't really depend on any other vendor or 3th party to sell.

The same is with "games", for long everybody had to put up with the big bad wolf... unless users directly boycott it, nothing will change anytime soon, even if it is a clear stagnating force in the market. If not clearly better now, Radeon on AMD could be miles away than Radeon on Intel... simple, just make the drivers to use XOP has much as possible, its not a lock-in tactic, it could make those drivers really fly (special the integer FMAC part, FMAC that is late on x86 almost a decade thanks to intel), and in the end Intel could had implemented XOP... use HTX exclusively for GPGPU boards, again not a lock-in, but incredibly better, and in the end intel could had implemented HTX, Hypertransport is a no-royalty standard ( stuck with crap PCIe is better lol)... etc etc...

But i think everything can start to change now. HSA is just a glimpse. The ARM armada knows it has no chance against intel if they don't battle together... that is why AMD managed to get all those big names in so short notice. Why on hell would ARM or Samsung or Qualcomm etc want to be in bed with AMD !??.. if its not the enemy of my enemy is my friend !?... AMD will have full HSA and more, looong before any ARM has it, it seems too much of a give way, isn't it ? ... Intel must had calculated similar, most probably will go nowhere very fast, but perhaps calculated wrong, the same with all arrogant bullies, they can never imagine they can be targets lol

And if that sentiment sticks... will be the end of it. A flush of rebellion.

 

[noob2222]

[
I believed it was made clear in many occasions that HSA harwdware requires HSA software to see the 'massive' benefits. Therefore, I continue without understanding why the software issue is mentioned again and again and again and again.

About upgrading. Take a look to server roadmaps. I already gave my estimations on Steamroller performance. Others seem to agree:
...

The CPU in Kaveri would be close to the level of a FX-8100 (octo-core BD).

because statements like this are repeated again and again and again and again without mentioning that this is only with "HSA software aware programs" wich won't be available

No. My above claim is for the CPU running an ordinary CPU benchmark suite for windows 7. That is why I said CPU and didn't mention HSA. In fact, I have a Linux Himeno benchmark where the CPU in Kaveri will be faster than a FX-8150. Again this is not about HSA.

ya, and pigs fly when a 4 core kaveri beats any 8 core FX cpu in an ordinary benchmark.

lets see some kaveri numbers since IN FACT you have one.

Before someone misinterpret me again: in some ordinary workloads a FX-8100 will be faster than Kaveri

I say almost all ordinary workloads the fx-8100 will be faster, unless your actually saying the 8100 as being the 2.8 ghz chip that barely even exists with OEM systems. The way you twist your words around and just throw 99% meaningless numbers around who knows what your actually thinking.

FX-8150 slower than a 4 core kaveri ... lol.

About Steam.

560ti + 550ti + 460 + 560 + 670 + 660: 11,27%.

"The remaining 4 are all intel igp.": 11.5%

My point remain: How many gamers have a HD7990? 0.1%?

I don't know from where comes the myth that gamers only play with the most powerful dGPUs.

first off, where did this "7990" statement even come from?

thats a myth from where? your the first person to bring it up.

maybe your defending your own statement?

Not everything is brute force. The customized Radeon hardware in the PS4 offers performance beyond top high-end dGPUs for some workloads.

...

You cannot redefine gamer to be only what you want it to be. Gamer is not only the young who is playing Crysis 3 on max settings in a dual top-end GPU config with OC CPU. That is only a ~0.1% of the market.

Moreover, the worldwide trend is moving to mobile. That is why Intel developed Haswell with laptops in mind and didn't care about the desktop. AMD is also focusing on the mobile.

ya and in some people's fantasy gaming world, gamers includes facebook games. Now farmville, thats a seroius game right there, graphics are totally awesome and the gameplay can't be beat.

your getting just as bad as hafijur with your wild claims about kaveri FACTS, more like fantasies on a hypethetical playing field drummed up by who knows what with hypothetical supporting information.

then again, when you use very vague general statements, no one can prove you wrong since you didn't really say anything solid.

 

[juanrga]

HSA hardware specification requires a HMMU that can communicate with both coherent and non-coherent memory.

However, I have one doubt about Hypertransport/HTX. Didn't this break hUMA? Logically memory would be uniform, but physically the system would consist of two kind of memories due to speed: a faster pool (physical dimms on the mainboard) and a slower pool (accessed through the HTX slot)? What is your thought?

hUMA is not an HSA official standard. There is no need to mention it. For having hUMA it only needs to mention "cache coherency" is needed, without any other definition or mechanism or software, since "cache coherency" is a purely hardware mechanism and a lot of different implementers can have quite different mechanisms with different coherency protocols. So HSA only needs to mention CC is needed, IOMMU definition will do the rest, and then they will have hUMA.

[ UPDATE: not exactly, cache coherency can be implemented in software, that is , a reference mechanism "emulating" what an hardware implementation would do. And i think that will be case for the *Runtime*, it will have a reference mechanism for all those combinations of CPU + acceleratores that don't have an "hardware" mechanism... it could be clever enough that even a CPU +GPU can be cache coherent by hardware, yet a DSP that is not, will be also cache coherent by software, that is, clever enough to have multi configurations when only a part of those is cache coherent by hardware. IT could use PCIe boards as example (on intel), only software can have an appreciable overhead. But there isn't a Runtime definition yet, its in development... but that deosn't mean squat, only AMD is ahead of the curve, which is no strange since most of it is based on AMD tech anyway]

HT/HTX breaks hUMA as much as Kaveri with dedicated (rumor of kaveri with DDR3 + GDDR5(faster different)) memory for the GPU.

Being hUMA doesn't mean everything uniform or equal or tight physically integrated (though this is the simplest and more straightforward implementation)... the contrary, the letter "h" means "heterogeneous" a priori presupposes differences. NUMA also doesn't require equal DRAM pools for everything, Torrenza examples had some "accelerators" as co-processors on AMD platforms, with quite different memories and sizes.

[UPDATE: hUMA and NUMA are very similar. Both try to give a "single" view of the memory a system has.

While NUMA works with different physically addressable pools as if cascaded/hierarchical, and so each "processor" sees the memory in a different processor as its own, but is closed tied to a particular coherency mechanism and to physical addressing.

hUMA works similar but with "virtual memory" addressing. By ways of IOMMU definitions, it loses the close tightness with any particular coherency protocol, and so broadens the possibilities of having cache coherency between different processors with different mechanisms... its more about "memory coherency". In the end the result is the same, several processor elements will see the whole memory a system can have in a unified way.

That is how i see it, any correction is welcome... but all in all hUMA is not the reverse of NUMA, which many seem to interpret ]

Thanks by the answer to my question.

If HT/HTX breaks hUMA then I don't think this has future at AMD. In my opinion, hUMA is part of the HSA specification, because all the properties of hUMA (cache coherency is only one of them) coincide with the properties of the memory model of the copy of the HSA specification that I have. AMD presents hUMA in the following terms:

The HSA Foundation has released a new multicore architecture specification called hUMA or heterogeneous uniform memory access. hUMA allows CPUs and GPUs to share the same memory in a heterogeneous system architecture (HSA). hUMA makes it easier for developers to create apps that use the individual powers of CPUs and GPUS and is compatible with mainstream programming languages like Python, C++ & Java. You'll soon be able to see hUMA in action in AMD’s Kaveri processors.

I don't know the rumour about Kaveri that you mention. I only know an old rumour (latter discredited) that said that Kaveri was coming with a dual memory controller with support for DDR3 and GDDR5. However, the controller only could run one type of memory or the other but not both at the same time. Either all the main memory was DDR3 (like in a Trinity APU) or all the memory was GDDR5 (as in PS4 APU). This restriction makes sense, because an arbitrary mixture of DD3+GDDR5 breaks hUMA (in the above sense).

 

[juanrga]

[
I believed it was made clear in many occasions that HSA harwdware requires HSA software to see the 'massive' benefits. Therefore, I continue without understanding why the software issue is mentioned again and again and again and again.

About upgrading. Take a look to server roadmaps. I already gave my estimations on Steamroller performance. Others seem to agree:
...

The CPU in Kaveri would be close to the level of a FX-8100 (octo-core BD).

because statements like this are repeated again and again and again and again without mentioning that this is only with "HSA software aware programs" wich won't be available

No. My above claim is for the CPU running an ordinary CPU benchmark suite for windows 7. That is why I said CPU and didn't mention HSA. In fact, I have a Linux Himeno benchmark where the CPU in Kaveri will be faster than a FX-8150. Again this is not about HSA.

ya, and pigs fly when a 4 core kaveri beats any 8 core FX cpu in an ordinary benchmark.

lets see some kaveri numbers since IN FACT you have one.

Before someone misinterpret me again: in some ordinary workloads a FX-8100 will be faster than Kaveri

I say almost all ordinary workloads the fx-8100 will be faster, unless your actually saying the 8100 as being the 2.8 ghz chip that barely even exists with OEM systems. The way you twist your words around and just throw 99% meaningless numbers around who knows what your actually thinking.

FX-8150 slower than a 4 core kaveri ... lol.

About Steam.

560ti + 550ti + 460 + 560 + 670 + 660: 11,27%.

"The remaining 4 are all intel igp.": 11.5%

My point remain: How many gamers have a HD7990? 0.1%?

I don't know from where comes the myth that gamers only play with the most powerful dGPUs.

first off, where did this "7990" statement even come from?

thats a myth from where? your the first person to bring it up.

maybe your defending your own statement?

Not everything is brute force. The customized Radeon hardware in the PS4 offers performance beyond top high-end dGPUs for some workloads.

...

You cannot redefine gamer to be only what you want it to be. Gamer is not only the young who is playing Crysis 3 on max settings in a dual top-end GPU config with OC CPU. That is only a ~0.1% of the market.

Moreover, the worldwide trend is moving to mobile. That is why Intel developed Haswell with laptops in mind and didn't care about the desktop. AMD is also focusing on the mobile.

ya and in some people's fantasy gaming world, gamers includes facebook games. Now farmville, thats a seroius game right there, graphics are totally awesome and the gameplay can't be beat.

your getting just as bad as hafijur with your wild claims about kaveri FACTS, more like fantasies on a hypethetical playing field drummed up by who knows what with hypothetical supporting information.

then again, when you use very vague general statements, no one can prove you wrong since you didn't really say anything solid.

I can do something better than showing you ordinary benchmarks (aka non-HSA) where a Kaveri quad outperforms an octo-core FX-8150.

I can show you ordinary benchmarks (aka non-HSA) where a Trinity quad competes or even outperforms an octo-core FX-8150.

This is the Linux Himeno benchmark that I mentioned above. A10-5800k is a bit faster than FX-8150.

There are more benchmarks where the A10-5800k is faster than FX-8150: FFmpeg, FFM Nero2D...

One can also find benchmarks under windows where an A10-5800k competes with an FX-8150

Of course, in other ordinary benchmarks the 8150 will be faster than Trinity and Kaveri, but as shown, it is not needed to see "flying pigs" (as you claimed) for a kaveri quad outperforming an octo-core BD FX.

 

[esrever]

I believe HSA will completely cut out discrete GPUs except for the highest of high end systems in the next 10 years due to how little we gain from pure graphics hardware these days. The first part for AMD would be to get the hardware out and then the entire ARM consortium will follow. Every soc in the world will eventually use HSA because they will always have a integrated GPU, this allows for higher hardware utilization. If AMD can standardize it and get enough support for it, they can be in a good spot for mobile cpus without discrete gpu parts. I doubt the GPU compute units will be integrated into the FPU any time soon. They would need both for the foreseeable future. What HSA does is make it easy to utilize the GPU portion of the APU so the programmer can choose to not use the FPU where the GPU cores will be faster. Once HSA become the common place, AMD can eventually do away with all but 1 FPU in the CPU, which would be purely used for legacy support. I don't believe GPU accessible registers that can process SIMD instructions will be in the works for a while. You will still need legacy support on FPU instructions even if you can do that simply because of scheduling to the GPU cores will get you conflicts and you can't always resolve these with pipelinable instructions on the CPU side.

Kaveri having huma is a big thing but also its not. In the PC front I don't see it coming to great use for a couple of years. The hardware has to be out before the software. Having a coherent memory across all heterogeneous processors is required but will not be utilized at this point. Simple tweaks to compilers should start make use of it eventually. I would like to see some good GPGPU performance from it but I doubt its a game changer yet. We will have to wait and see.

They won't be replacing dGPU's ever. This is physics speaking, it's impossible for an iGPU to be as powerful as a dGPU due to space / cost / electricity limits. It's like saying mopeds or "economy" cars will replace consumer transportation. Gamers and their ilk will always want the powerful dGPU to provide the best experience, the iGPU is there for budget / economy systems.

Also GPU "compute units" are "FPU's". They do the exact same math with the only difference being their organized into massive arrays.

It gets to the point where nobody needs the performance out of a discrete GPU. Just like Laptops eventually overtook desktops and how tablets are becoming pretty big things. Considering the next generation of consoles both already use APUs. All mainstream gaming will eventually be done on APUs even on PCs. If you are targeting 1080p 60fps, which is what I would guess 90% of PC gamers do, in 3-4 years, APUs will be able to do that no problem. Once sub 20nm comes out, APUs will be able to hold 7970 performance hardware in them. The performance you get at that point, will be so insignificant when you add in a discrete GPU for the average gamer.