AMD summit today; Kaveri cuts out the middle man in Trinity.

[happysmiles]

http://www.extremetech.com/computin...i-will-fully-share-memory-between-cpu-and-gpu

AMD is hosting its Fusion Developer Summit this week, and the overarching theme is heterogeneous computing and the convergence of the CPU and GPU.

Pretty much Kaveri CPU and GPU share the same bandwidth instead of mirror copying the data (trinity) which means super speed.
Seems realistic and all but Trinity isn't even available! more typical AMD slides and promises.

it all seems good but I guess I can only hope for the best.

The industry is definitely moving towards a more blended processing environment, something that began with the rise of specialty GPGPU workstation programs and is now starting to integrate itself with consumer applications. Standards like C++ AMP, OpenCL, and Nvidia’s CUDA languages harness the graphics cards for certain tasks. More and more programs are using the GPU for certain tasks (even if it’s just drawing and managing the UI), and as developers jump on board it should accelerate even more towards using components to their fullest on the software side. On the hardware side of things, we are already seeing integration of GPUs into the CPU die and specialty application processors (at least in mobile SoCs). Such varied configurations are becoming common and are continuing to evolve in a combined architecture direction.
The mobile industry is a good example of HSA catching on with new system-on-a-chip processors coming out continuously and mobile operating systems that harness GPU horsepower to assist the ARM CPU cores. AMD isn’t just looking at low power devices, however — it’s pushing for “one (HSA) chip to rule them all” solutions that combine GPU cores with CPU cores (and even ARM cores!) that process what they are best at and give the best user experiences.

The overall transition of hardware and software that fully takes advantage of both processing types is still a ways off but we are getting closer everyday. Heterogeneous computing is the future, and assuming most software developers can be made to recognize the benefits and program to take advantage of the new chips, I’m all for it. When additional CPU cores and smaller process nodes stop making the cut, heterogeneous computing is where the industry will look for performance gains.

 

[pelov]

Slow down Benchpre... wait a minute!

AMD needs to really improve upon their IMC drastically in Kaveri. Getting spanked by Intel at 1333 while at 1866 is a damn shame for the company that introduced the IMC in the first place. They're certainly not going to be increasing cache sizes to the point that Intel will either so they've got no choice.

 

[bronxzv]

http://www.extremetech.com/computin...i-will-fully-share-memory-between-cpu-and-gpu

wow! what a slide, CPUs with ~ 180 GFLOPS in 2008 to ~220 GLOPS in 2013 and a perfectly linear scaling, it's even more lame than the infamous Power PC alliance era "RISC vs CISC" slides

 

[ShintaiDK]

Aint the IMC speed issue not mainly related to the NB clock? Its still 2.2Ghz. While everything in SB/IB runs at core speed.

 

[Homeles]

Aint the IMC speed issue not mainly related to the NB clock? Its still 2.2Ghz. While everything in SB/IB runs at core speed.

Don't get me started on AMD's garbage IMC... it's probably one of the biggest reasons why they're behind Intel right now.

 

[podspi]

Don't get me started on AMD's garbage IMC... it's probably one of the biggest reasons why they're behind Intel right now.

Seems kind of odd, given how important memory bandwidth is for APUs...


Also, I honestly do not understand why AVX2 always comes up in discussions of HSA/FSA/OpenCL. I see people continually framing the discussion as "AVX2 vs. heterogeneous computing" and I just do not think that is the case. We've seen Intel and AMD release OpenCL-compatible CPUs, and we will see both Intel and AMD introduce AVX2-compatible CPUs.

The main benefit of HSA, as far as I can tell, is making it easier for developers to extract processing power out of a heterogeneous system. While I agree that a homogeneous system will probably always be "the fastest", I think it makes a lot of sense for developers to be able to use as much processing power as possible, from as many sources as possible. Especially in mobile situations where you might not have an abundance of any one computing resource.

 

[piesquared]

HSA seems to be about using the best resource for the particular job in the queue. If an application, or part of an application, runs best on a massively parallel GPU then the code is handed of to the GPU. Other types of workloads would be handed off to parts of the APU that execute that code most efficiently. The end result is a big reduction in power consumption and a big gain in performance. It's easy to see why intel tries so hard to evangelize avx2 as the Holy Grail of all things computing, HSA is going to be a disruptive technology. And better yet, it can be inclusive of all current and future instructions.

 

[Haserath]

Isn't the memory bandwidth 'problem' due to the fact that the cores memory subsystem can't keep up in tests like sandra?

Isn't that why as core speed rises, the memory performance in those benchmarks also seem to rise?

 

[SocketF]

Isn't the memory bandwidth 'problem' due to the fact that the cores memory subsystem can't keep up in tests like sandra?

Yes, Sandra only runs on the CPU-Cores and didnt utilize the GPU-memory controller.
Just compare Sandra scores of the FX4xxx series to the FX8xxx series, the memory bandwidth doubles, however both have the very same memory controller type.
Conclusion: The memory controller is one of the best parts, no problems there ;-)

 

[AMD > INTEL]

Yes, Sandra only runs on the CPU-Cores and didnt utilize the GPU-memory controller.
Just compare Sandra scores of the FX4xxx series to the FX8xxx series, the memory bandwidth doubles, however both have the very same memory controller type.
Conclusion: The memory controller is one of the best parts, no problems there ;-)

this

 

[CHADBOGA]

HSA seems to be about using the best resource for the particular job in the queue. If an application, or part of an application, runs best on a massively parallel GPU then the code is handed of to the GPU. Other types of workloads would be handed off to parts of the APU that execute that code most efficiently. The end result is a big reduction in power consumption and a big gain in performance. It's easy to see why intel tries so hard to evangelize avx2 as the Holy Grail of all things computing, HSA is going to be a disruptive technology. And better yet, it can be inclusive of all current and future instructions.

HSA ain't gonna do squat.

It will be the next 3D-Now.

 

[ShintaiDK]

HSA ain't gonna do squat.

It will be the next 3D-Now.

And SSE5. Basicly died before it was born.

 

[Idontcare]

HSA ain't gonna do squat.

It will be the next 3D-Now.

And SSE5. Basicly died before it was born.

It may be true but at least if it comes to pass then it died trying rather than doing nothing like Via.

AMD is taking a gamble, but doing nothing is a gamble as well, so at least they are taking a gamble on something that stands a chance of getting off the ground simply for the fact that so many other businesses are willing to give it a go as well.

What is AMD's alternative?

 

[Zor Prime]

Some people think AMD released 3DNow! yesterday and are ditching it tomorrow.

The fact of the matter is that 3DNow! is older than some people that post on this forum. It had its run.

 

[ShintaiDK]

What is AMD's alternative?

Good question. I dont know, and I am not even sure there is an alternative.

VIA never had a chance like´other x86 makers such as IBM, NEC, TI, STM, Fujitsu, OKI, Siemens, Cyrix, Intersil, C&T, NextGen, UMC, RDC, ITT, National Semiconductor, ULSI, Weitek, Centaur, Rise, Transmeta etc. And it might simply be AMDs turn to join the flok. In terms of economics there just really aint room anymore.

I guess thats why AMD is changing focus completely.

 

[Arzachel]

HSA ain't gonna do squat.

It will be the next 3D-Now.

You do know that HSA isn't an ISA extension but more like a set of hardware and software guidlines, right? And that it isn't tied to x86, right?

 

[Dresdenboy]

You do know that HSA isn't an ISA extension but more like a set of hardware and software guidlines, right? And that it isn't tied to x86, right?

You shouldn't criticize or make fun of other people's simple models to understand the world. ;-)

OK, some pictures might help:

 

[bronxzv]

You shouldn't criticize or make fun of other people's simple models to understand the world. ;-)

OK, some pictures might help:

If I understand it well native CPU code (Intrinsics + TBB) is half the performance of CPU + GPU with HSA (Open CL)

In other words in this AMD cherry picked example and AMD weak CPU simply doubling the CPU cores will provide the same speed without all the pain of this incredibly complex HSA thing ? Do I get something wrong ?

 

[Homeles]

If I understand it well native CPU code (Intrinsics + TBB) is half the performance of CPU + GPU with HSA (Open CL)

In other words in this AMD cherry picked example and AMD weak CPU simply doubling the CPU cores will provide the same speed without all the pain of this incredibly complex HSA thing ? Do I get something wrong ?

Yeah, you're completely off. The complexity of the architecture is on AMD's side, and as they are an electrical engineering firm, it's not like they're not used to complexity -- it's what they do.

Did you even read the chart?

Yes, you can double the cores, but the complexity of writing the program goes up significantly. If you're coding for HSA though, as the chart very obviously points out (given that it's the entire damn point of the slide), it takes little programming work to see massive performance.

 

[bronxzv]

The complexity of the architecture is on AMD's side, and as they are an electrical engineering firm

EE ? there is 4 levels of software in this stack between the user app and the hw

Yes, you can double the cores, but the complexity of writing the program goes up significantly.

not really for existing parallel code bases, by far less than rewriting everything for OCL and all its clumsy limitations

 

[piesquared]

HSA ain't gonna do squat.

It will be the next 3D-Now.

Thanks for reviving the thread. HSA is something everyone needs to read about.

Given the announcment of the HSA foundation (ARM, Imagination, MediaTek and Texus Instruments [which IDC may hold dear to his heart]) and further announcments inbound in the near future, i'd say you are counting the proverbial chickens before they hatch. There are clear benefits to HSA, and ISV's/IHV's are coming on board because of it. Any idea why they would fully endorse the platform if they thought it was going to fail? Your assesment doesn't seem to make a whole lot of sense.

Here's a link to help you catch up:

http://www.inxpo.com/events/amd/afds-d

 

[CPUarchitect]

Any idea why they would fully endorse the platform if they thought it was going to fail?

Because AMD is offering them free beers?

 

[Khato]

Because AMD is offering them free beers?

And PR statements are cheaper than said beers last I checked, right?

 

[ShintaiDK]

It cost nothing to endorse a project. Look back in history on how many of such projects there has been and how few materialized into something. Alot of companies uses it for free PR, since their logo gets shown alot of places.

 

[piesquared]

lmao of at the level of denial.