AMD Debuts New SDK, Tools and Libraries, for Heterogeneous Computing Developers

[Cristian_25H]

AMD kicked off its 2013 Developer Summit (APU13) today, announcing a new unified Software Development Kit (SDK,) an improved CodeXL tool suite with added features and support for the latest AMD hardware, and added heterogeneous acceleration in popular Open Source libraries. Together, these tools provide a substantial step forward in productivity and ease-of-use for developers wishing to harness the full power of modern heterogeneous platforms spanning form servers to PCs to handheld devices.

"Developers are essential to our mission of realizing the full potential of modern computing technologies," said Manju Hegde, corporate vice president, Heterogeneous Solutions, AMD. "Enriching the developer experience by harnessing these technologies is a critical part of AMD's mission to accelerate developer adoption."



To achieve this common mission, AMD is announcing improvements across four fronts:

• The unified SDK includes AMD APP SDK 2.9 and is the most user-friendly heterogeneous computing SDK yet. It provides improved ease-of-use and developer productivity by adding several new capabilities. This is AMD's first step toward providing developers with simple access to all the programmable components of the company's products. Some of the salient features are: a Web-based sample browser that makes it easy to find the right samples for a project; added support for CMake -- a popular make utility; improved OpenCL source editing with a plug-in to visual studio; and the addition of several new samples highlighting use of optimized open source libraries (OpenCV, Bolt) to get acceleration with minimal effort.
• The unified SDK also includes the Media SDK v1.0 beta release. This will be AMD's first public release of the Media SDK, which enables developers to leverage AMD's unique and differentiated multimedia capabilities. Some of the key features of the Media SDK are a GPU-accelerated video pre/post processing library and a library for low latency video encoding.
• The unified SDK also promotes new heterogeneous acceleration optimizations in several open source libraries with the goal of making it simple for developers to accelerate applications. These include: OpenCV (most popular computer vision library) now with many OpenCL accelerated functions; clMath with accelerated BLAS and FFT libraries accessible from Fortran, C and C++; and Bolt, a C++ template library providing GPU off-load for common data-parallel algorithms, now with cross-OS support as well as performance improvements and new functionality.
• The latest CodeXL tools suite (version 1.3,) AMD's comprehensive heterogeneous developer tool offering, now supports Java, the world's most popular programming language. To further AMD's efforts to provide the best integrated tool suite, AMD has incorporated static kernel analysis capabilities. Also included in this release are added support for remote debugging/profiling to enable server and embedded customers as well as support for the latest GPU products from AMD (GCN-based discrete GPUs and APUs.)

"AMD continues to deliver excellent heterogeneous programming developer tools for OpenCL -- the industry standard for heterogeneous programming," said Bill Richard, vice president of Software Development at Winzip. "AMD's tools have been instrumental in our efforts to deliver significant application performance improvements to our customers."

These new product releases represent another step forward toward AMD's goals of supporting cross-platform solutions (across OSes and vendors,) multiple programming languages and continued contributions to the Open Source community. This is part of AMD's ongoing commitment to make heterogeneous computing pervasive and mainstream.

AMD also announced today at APU13 details about "Kaveri," the third generation performance APU from AMD, during a keynote delivered by Dr. Lisa Su, senior vice president and general manager, Global Business Units, AMD.

"Kaveri" is the first APU with HSA features, AMD TrueAudio technology and AMD's Mantle API combining to bring the next level of graphics, compute and efficiency to desktops (FM2+), notebooks, embedded APUs and servers. FM2+ shipments to customers are slated to begin in late 2013 with initial availability in customer desktop offerings scheduled for Jan. 14, 2014. Further details will be announced at CES 2014.

In addition to the "Kaveri" announcement, Dr. Su highlighted AMD's leadership in APU technology and heterogeneous compute capabilities, and discussed the developer-centric future of AMD technology. With graphics, gaming, performance mobile clients and embedded devices growing in prevalence, AMD is at the epicenter of this convergence, and is equipped with the necessary IP to execute on this opportunity for our customers. To enable these growing ecosystems, Dr. Su made commitments to continue supporting programming tools and platforms like Heterogeneous System Architecture (HSA), Mantle, OpenCL, C++ AMP and Microsoft DirectX that are important to application and game developers.

View at TechPowerUp Main Site

 

[NeoXF]

This is great and all (and quite essential for the success of APUs), but where`s the post on Kaveri (A10-7850K) itself?

Bonus: APU13: AMD "Kaveri" vs Intel Core i7 4770K + NVIDI...

 

[Recus]

This is great and all (and quite essential for the success of APUs), but where`s the post on Kaveri (A10-7850K) itself?

Bonus: APU13: AMD "Kaveri" vs Intel Core i7 4770K + NVIDI...

i7 4770k bottlenecked by GT 630. Point?

 

[sunaiac]

I think the point might be that X$ of AMD >> to 3X$ of intel+nvidia.

 

[NeoXF]

i7 4770k bottlenecked by GT 630. Point?

WTF do you mean `point?` Also, there`s no such thing as a GPU bottlenecking a CPU. GPUs need CPUs, not the other way around.

It`s 2.5x+ times faster than a GT 630 w/ a beefy CPU (on stock D3D11, not Mantle), which in turn, makes it roughly close to HD 7750 in performance, which is just great.

However the other thing I look forward to is some HSA-enabled recompiled frequently used software, w/o OpenCL/DirectCompute specific paths, so `pure` x86. Like 7-zip or something.

 

[librin.so.1]

"Kaveri" is the first APU with HSA features, AMD TrueAudio technology and AMD's Mantle API combining to bring the next level of graphics, compute and efficiency to desktops (FM2+), notebooks, embedded APUs and servers. FM2+ shipments to customers are slated to begin in late 2013 with initial availability in customer desktop offerings scheduled for Jan. 14, 2014. Further details will be announced at CES 2014.

Whelp.
That pretty much confirms that "Imma need to get a new mobo"

 

[ice_v]

An AMD Press Release without powerpoint slides? let me fix that:

 

[john_]

i7 4770k bottlenecked by GT 630. Point?

You get from an APU two times better performance than with 630. Maybe the APU is close, equal, or better than a GT 640?
So the point here is that everything from 630 and down will almost cease to exist as an option for new systems in a year from now and maybe in two years there will be no reason to build a new system with a less than $100 card.

 

[Jorge]

AMD is leading the way in APUs. As we will soon see APUs will replace discrete CPU/GPU for 90+% of the PC industry because they cost less, perform better and consume less power. Even hard core gamers will be using APUs on desktop PCs as the performance just continues to ramp on AMD APUs. It's all good for consumers and AMD.

 

[RCoon]

Even hard core gamers will be using APUs on desktop PCs as the performance just continues to ramp on AMD APUs.

>Implying in the year 2013
>hardcore gamers
>using APU's instead of dedicated graphics
>shiggy diggy

Pass the crack pipe jorge.

 

[Frick]

An AMD Press Release without powerpoint slides? let me fix that:

http://www.tomshw.it/files/2013/11/immagini_contenuti/50721/amd-apu-13-01_t.jpg

http://www.tomshw.it/files/2013/11/immagini_contenuti/50721/amd-apu-13-02_t.jpg

http://www.tomshw.it/files/2013/11/immagini_contenuti/50721/amd-code-xl-13_t.jpg

http://www.tomshw.it/files/2013/11/immagini_contenuti/50721/amd-libraries_t.jpg

http://www.tomshw.it/files/2013/11/immagini_contenuti/50721/amd-unified-sdk_t.jpg

Thank you, I got worried for a bit there.

>Implying in the year 2013
>hardcore gamers
>using APU's instead of dedicated graphics
>shiggy diggy

Pass the crack pipe jorge.

To be fair he didn't specifiy when this will happen.

 

[RCoon]

Thank you, I got worried for a bit there.



To be fair he didn't specifiy when this will happen.

When a PC can handle 80w of CPU heat and 300w of GPU heat ALL IN ONE DIE, I'll carry on passing the crack pipe.

 

[librin.so.1]

in other words: "when chips start using something other than silicon"

 

[NeoXF]

^Yeah... IDK about you guys... but something like a iPhone is doing quite a good job of having decent graphics, decent resolution and all that at a micro-fraction of the power usage to above said hardware does. Sure, it's apples and exotic fruit plantations... (pun intended), but Moore's Law is THE Law, and it's shining mighty bright right now.

Keeping that I mind, I sometimes disgust myself when I think of all the over-revved PC hardware I've had/seen/tried/and-so-on, that aren't even the slightest bit as efficient as I'd want them to be. Looks a lot like a 6,4L american muscle car vs a Ariel Atom. Bigger is surely, not always better, diminishing returns (a lot of other factors like, tons of abstract layers) are some nasty motherf*****s...


Anyway, tomorrow or so, is AMD's Kaveri Day, right? Can't wait!

 

[alwayssts]

I think a thing to consider about that BF4 demo is ram.

I would be willing to bet they are using 3000mhz ddr3, or at least something similar.

While one could argue trinity was a bet on ddr3 2133-2400 that paid off (by ram prices dropping through the basement) one has to wonder if that will be the case with 3000+ stuff (or ram that can at least hit that mark overclocked). That is a big part of the equation, and something that could certainly be a value off-set.

Again, not to poo-poo on this as it is exciting, but amd promised this very chip to be over 1TF over and over again. Slice that any way you want (keeping power down to compete with intel, problems at GF, a later sku with higher clocks, etc) but this is quite a bit off the mark.

I personally expected a ~825-900mhz gpu, something similar to past products that were matched to completely saturate the memory bus at a certain high speed ddr3 spec (which granted made little sense considering the cpu and their typical market). While this is much more realistic on several fronts (power, bandwidth for the whole chip) and should be more efficient in some ways (although gpu clock is pretty darn low for 28nm and at least .85v), it is certainly under-delivering on what they promised.

If this is the route they are taking, and understandably so, I would have liked to see the same 3.7ghz cpu clock matched with a 7 CU design at a similar clock (720mhz) or 6 slightly higher (which granted may be slightly less power efficient). It would have made sense for 2666mhz ram, which is under $100 for 8GB, and on top of less power and a smaller, more efficient chip (especially if 8 rops and not 16) most people would likely never tell the difference.