Was AMD holding back on their fusion line?

[NTMBK]

8 excavator cores with a ps4 sized gpu and gddr5 would kick ass

not sure how you would go about buying gddr5 so you might need it built in with 32 gb since programs are likely to start using up to 16 gb soon

as for graphics you would go with a descrete card even with that large gpu built into the apu. you can use the gpu built into the apu for open cl compute tasks

There is a standard for GDDR5M on SODIMMs, but nobody seems to have actually implemented it in a product. To be honest we probably won't see the bandwidth we need for better PC APUs until stacked memory takes off. Thankfully, AMD is working on that: http://www.microarch.org/micro46/files/keynote1.pdf

 

[ShintaiDK]

q4 was a profitable quarter no? they sold a facility to make that profit but non the less they still made a profit

the predictions for 2014 seem to say profitable all the way

they have the console business to provide a stabilized and significant source of revenue for several years

the apus in low power applications like notebooks have been more sucessful than desktop cpus or hpc

rory read seems to be working out just fine for amd

they also hired a lot of old talent

so i think the prospects look good for amd

I guess you didnt check their financials. The PC division lost 7mio$ in Q4, and lost 22M$ on the year. And revenue dropped from 4005M$ to 3104M$ YoY.

Revenue is expected to drop 16% in Q1 2014.

Is that your definition of success?

 

[AtenRa]

The "Computing solutions" aka PC group is a loss with a continual drop in revenue. Thats anything but a success.

You do know that SoCs for the PS4 and XBOX are APUs. APUs comprise more than 60-70% of the total AMD Revenue, hardly a failure.

 

[AtenRa]

I guess you didnt check their financials. The PC division lost 7mio$ in Q4, and lost 22M$ on the year. And revenue dropped from 4005M$ to 3104M$ YoY.

Revenue is expected to drop 16% in Q1 2015.

Is that your definition of success?

16% is Quarter to Quarter, it will be way higher YoY.

 

[BSim500]

Cue HSA / OpenCL benchmarks, showing Kaveri outperforming a i7-4770K in LibreOffice spreadsheets.

Problem is, not many people who use giant complex spreadsheets large enough to benefit are using it due to poor formatting compatibility with MS Office. I tried Open/Libre Office several times in the past and it's still way short for being able to open a Word / Excel document, edit it, save it, then open it in Word / Excel without losing formatting in one way or another. Its open/save filters are still way behind, eg, SoftMaker Office. And some other "HSA" apps, eg, Corel Aftershot Pro, are no faster at all:-
http://www.extremetech.com/wp-content/uploads/2014/01/HSA-Corel.png

Likewise, LibreOffice programmers themselves admit much of the "slowness" of CPU-only and speedup of rewritten HSA is due to rewriting prior bad coding & design decisions rather than the actual HSA itself:-

"The reason that LibreOffice Calc is slow isn't because it lacks a GPU; it is because it is implemented in a very inefficient way. It is claimed that what is responsible for the sluggish performance is the object-oriented design that resulted in each cell being defined as an object."
http://www.i-programmer.info/news/202-number-crunching/6073-libreoffice-calc-to-get-gpu-support.html

In reality, you could double the speed of LibraOffice on pretty much every PC if you rewrote it properly - even without HSA. Last time I looked at the source code, half of the code comments were still in German - a leftover relic from Sun's acquisition of "StarOffice" dating back to 1985 StarWriter by StarDivision based in Lüneburg, Germany, which turned into OpenOffice, which turned into LibreOffice, etc. I admire coders ability to add new features and keep it modern, but it's quite possibly one of the slowest & most bloated pieces of software even written, and much of the "HSA speed-up" is "ditching sh*tty code" speed-up according to their own devs.

 

[norseamd]

There is a standard for GDDR5M on SODIMMs, but nobody seems to have actually implemented it in a product. To be honest we probably won't see the bandwidth we need for better PC APUs until stacked memory takes off. Thankfully, AMD is working on that: http://www.microarch.org/micro46/files/keynote1.pdf

with all the heat problems we have with current chips how are we going to make 3d stacked apus

can you integrate heat spreaders into the stacking of 3d processors

problem: this slide is located to the right of the 2020s. how long are they predicting this will take

Die Stacking will
reduce cost by
sharing at the die
level and driving
customization

 

[NTMBK]

with all the heat problems we have with current chips how are we going to make 3d stacked apus

can you integrate heat spreaders into the stacking of 3d processors

problem: this slide is located to the right of the 2020s. how long are they predicting this will take

Not necessarily 3D stacks- "2.5D". A stack of high-bandwidth memory, a CPU and a GPU all on an interposer.

And remember, AMD have sold 220W socketed CPUs before.

 

[norseamd]

why not just connect the memory and apu horizontally

 

[R0H1T]

Problem is, not many people who use giant complex spreadsheets large enough to benefit are using it due to poor formatting compatibility with MS Office. I tried Open/Libre Office several times in the past and it's still way short for being able to open a Word / Excel document, edit it, save it, then open it in Word / Excel without losing formatting in one way or another. Its open/save filters are still way behind, eg, SoftMaker Office. And some other "HSA" apps, eg, Corel Aftershot Pro, are no faster at all:-
http://www.extremetech.com/wp-content/uploads/2014/01/HSA-Corel.png

Likewise, LibreOffice programmers themselves admit much of the "slowness" of CPU-only and speedup of rewritten HSA is due to rewriting prior bad coding & design decisions rather than the actual HSA itself:-

"The reason that LibreOffice Calc is slow isn't because it lacks a GPU; it is because it is implemented in a very inefficient way. It is claimed that what is responsible for the sluggish performance is the object-oriented design that resulted in each cell being defined as an object."
http://www.i-programmer.info/news/202-number-crunching/6073-libreoffice-calc-to-get-gpu-support.html

In reality, you could double the speed of LibraOffice on pretty much every PC if you rewrote it properly - even without HSA. Last time I looked at the source code, half of the code comments were still in German - a leftover relic from Sun's acquisition of "StarOffice" dating back to 1985 StarWriter by StarDivision based in Lüneburg, Germany, which turned into OpenOffice, which turned into LibreOffice, etc. I admire coders ability to add new features and keep it modern, but it's quite possibly one of the slowest & most bloated pieces of software even written, and much of the "HSA speed-up" is "ditching sh*tty code" speed-up according to their own devs.

If any software can make use of OpenCL &/or the igp for compute oriented tasks it will benefit tremendously with the speedup in performance, like Winzip for instance, plus the fact that AMD has better iGPU at every price point will only make their products better, & not just the promise of HSA/HUMA which could also be implemented by Intel some years down the line & like Nvidia is trying to do come 2016.

 

[ShintaiDK]

If any software can make use of OpenCL &/or the igp for compute oriented tasks it will benefit tremendously with the speedup in performance, like Winzip for instance, plus the fact that AMD has better iGPU at every price point will only make their products better, & not just the promise of HSA/HUMA which could also be implemented by Intel some years down the line & like Nvidia is trying to do come 2016.

If is the key issue. We had GPU compute for ages now. CUDA, OpenCL, HSA, DirectCompute etc. And the amount of apps is at best counted in a couple of 100s.

And its always the next GPU compute that will be the saviour of all things. CUDA failed, OpenCL failed, DirectCompute failed and somehow HSA is now to succeed where all the others failed. Problem is there will never be a big market, its the hunt for a phantom ghost. And its done with resources that is not there.

 

[CHADBOGA]

If HSA was ever going to amount to anything, there would be a lot more buzz around it.

Hell, I've seen way more excitement for other technologies that failed, so to me that doesn't bode way at all, for believing that HSA will be some kind of game changer, let alone a game changer that gives AMD any kind of unique advantage over Intel.

 

[norseamd]

well open cl has a lot of potential also and there is a lot of trouble with adoption of that also

hsa or not heterogenous computing has a lot of potential

 

[R0H1T]

If HSA was ever going to amount to anything, there would be a lot more buzz around it.

Hell, I've seen way more excitement for other technologies that failed, so to me that doesn't bode way at all, for believing that HSA will be some kind of game changer, let alone a game changer that gives AMD any kind of unique advantage over Intel.

HSA is coming to desktops & servers there's no point denying that, the fact that Nvidia has already introduced unified memory (software only) for CUDA 6 proves this, but the potential of OpenCL is what is a major plus for AMD in all of this. Intel will follow suit now whether it's next year or 5 years down the line is their prerogative but AMD has yet again set the ball rolling, just like the first multicore CPU or even APU, & it's an innovation that only some of us will find any interest in downplaying, but its potential benefits are simply enormous.

 

[CHADBOGA]

HSA is coming to desktops & servers there's no point denying that, the fact that Nvidia has already introduced unified memory (software only) for CUDA 6 proves this, but the potential of OpenCL is what is a major plus for AMD in all of this. Intel will follow suit now whether it's next year or 5 years down the line is their prerogative but AMD has yet again set the ball rolling, just like the first multicore CPU or even APU, & it's an innovation that only some of us will find any interest in downplaying, but its potential benefits are simply enormous.

So will HSA be a bigger or lesser yawn than AVX or every other supposed ground breaking technology which amounted to literally nothing and certainly did nothing to shake up the current order of things.

 

[norseamd]

HSA is coming to desktops & servers there's no point denying that, the fact that Nvidia has already introduced unified memory (software only) for CUDA 6 proves this, but the potential of OpenCL is what is a major plus for AMD in all of this. Intel will follow suit now whether it's next year or 5 years down the line is their prerogative but AMD has yet again set the ball rolling, just like the first multicore CPU or even APU, & it's an innovation that only some of us will find any interest in downplaying, but its potential benefits are simply enormous.

the industry and the particular companies and investors involved have a influence on this as well

so no a particular product will not just sell because it is good

considering the way some of the old folks around here talk about the alpha processors why are they still not around

from the way they talk sounds like i might wish they were

 

[norseamd]

So will HSA be a bigger or lesser yawn than AVX or every other supposed ground breaking technology which amounted to literally nothing and certainly did nothing to shake up the current order of things.

and how much more work could software programmers do with any of these technologies?

 

[R0H1T]

So will HSA be a bigger or lesser yawn than AVX or every other supposed ground breaking technology which amounted to literally nothing and certainly did nothing to shake up the current order of things.

So then tell me the number of major apps that support (but more importantly use) AVX or AVX2 & Quick Sync, then compare that number with the ones enabling OpenCL acceleration. Somehow I feel you're deliberately downplaying this thing because AMD has it (HSA/HUMA) & Intel doesn't but the day they enable something like this on their CPU's it'll be the next best thing since sliced bread for you!

the industry and the particular companies and investors involved have a influence on this as well

so no a particular product will not just sell because it is good

considering the way some of the old folks around here talk about the alpha processors why are they still not around

from the way they talk sounds like i might wish they were

I'm not sure I follow what you're trying to say?

 

[norseamd]

I'm not sure I follow what you're trying to say?

he was saying that if the technology of a product is good enough then it will sell but i am saying is that what the companies and investors are thinking has practicly more of a influence

there is a lot of technology that could be made but we get what companies more or less decide to make

 

[R0H1T]

he was saying that if the technology of a product is good enough then it will sell but i am saying is that what the companies and investors are thinking has practicly more of a influence

there is a lot of technology that could be made but we get what companies more or less decide to make

That is true but what I'm saying is that we're reaching a point where most (if not all) advancements in the field of MT/multi core processor based computing are behind us & we're gonna get more/better performance only by ditching the traditional approach. Enter OpenCL & HSA/HUMA which IMO are the natural evolution in the process of computing & it's only a matter of time when people (software developers) realize that GPU's are their best friends in this era of cut throat competition & they simply cannot ignore the scale & efficiency that modern GPU's have developed over a relatively small period of time.

 

[norseamd]

That is true but what I'm saying is that we're reaching a point where most (if not all) advancements in the field of MT/multi core processor based computing are behind us & we're gonna get more/better performance only by ditching the traditional approach. Enter OpenCL & HSA/HUMA which IMO are the natural evolution in the process of computing & it's only a matter of time when people (software developers) realize that GPU's are their best friends in this era of cut throat competition & they simply cannot ignore the scale & efficiency that modern GPU's have developed over a relatively small period of time.

agree

 

[mrmt]

That is true but what I'm saying is that we're reaching a point where most (if not all) advancements in the field of MT/multi core processor based computing are behind us & we're gonna get more/better performance only by ditching the traditional approach. Enter OpenCL & HSA/HUMA which IMO are the natural evolution in the process of computing & it's only a matter of time when people (software developers) realize that GPU's are their best friends in this era of cut throat competition & they simply cannot ignore the scale & efficiency that modern GPU's have developed over a relatively small period of time.

I think you are forgetting something: Execution.

It doesn't matter whether AMD, Nvidia or Intel develops this or that concept, what matters is how well the companies execute their roadmaps.

AMD indeed fielded the first commercial x86 CPU on the consumer market, but it was Intel that made multicore computing cheap with Conroe. AMD was the first IMC on the x86 market, but Intel with FSB was more than a match for them in most workloads, and once they fielded an IMC, AMD was left eating dust. AMD is the first to field something but Intel gets all the money, not a good trade off IMO.

As much as HSA and HUMA could be clever concepts, it's AMD ability to execute them in meaningful products, e.g., monetize these ideas, that matters. I'd go further and say that it's the only victory that matters. And do you believe AMD is the company with the ability to execute *AND* disseminate these concepts to the market, or do you think it takes another kind of company get that?

 

[CHADBOGA]

So then tell me the number of major apps that support (but more importantly use) AVX or AVX2 & Quick Sync, then compare that number with the ones enabling OpenCL acceleration. Somehow I feel you're deliberately downplaying this thing because AMD has it (HSA/HUMA) & Intel doesn't but the day they enable something like this on their CPU's it'll be the next best thing since sliced bread for you!

I am a sceptic of all these extensions being that groundbreaking.

It's not me going around saying how HSA or any extension, is going to be a game changer.

If & when Intel implements HSA, I'll still be yawning.

Somehow me not buying into the over the top AMD hype, upsets you?

Why?

 

[R0H1T]

I think you are forgetting something: Execution.

It doesn't matter whether AMD, Nvidia or Intel develops this or that concept, what matters is how well the companies execute their roadmaps.

AMD indeed fielded the first commercial x86 CPU on the consumer market, but it was Intel that made multicore computing cheap with Conroe. AMD was the first IMC on the x86 market, but Intel with FSB was more than a match for them in most workloads, and once they fielded an IMC, AMD was left eating dust. AMD is the first to field something but Intel gets all the money, not a good trade off IMO.

As much as HSA and HUMA could be clever concepts, it's AMD ability to execute them in meaningful products, e.g., monetize these ideas, that matters. I'd go further and say that it's the only victory that matters. And do you believe AMD is the company with the ability to execute *AND* disseminate these concepts to the market, or do you think it takes another kind of company get that?

Honestly their CPU's are somewhat underwhelming, let's not get into the debate of VFM for now, but their (i)GPU's are top notch & this is where I think (& hope) that HSA/HUMA plus OpenCL will put them in a decent position provided the software developers do their job properly.

 

[R0H1T]

I am a sceptic of all these extensions being that groundbreaking.

It's not me going around saying how HSA or any extension, is going to be a game changer.

If & when Intel implements HSA, I'll still be yawning.

Somehow me not buying into the over the top AMD hype, upsets you?

Why?

Why would I be upset if you continue to downplay the path AMD has taken? Maybe it's because I do take objection to your painting of this road as something less than pedestrian, sure you may yawn when we have HBM on basilisk but to say it doesn't matter, when in fact it's just your opinion, to the rest of us (or even a large majority of people some years down the line) is just being obtuse!

 

[CHADBOGA]

when in fact it's just your opinion, to the rest of us (or even a large majority of people some years down the line) is just being obtuse!

HSA will be one of hundreds of extensions, years down the line.

Why will it stand tall amongst them all?