A BRIEF PRIMER ON MANTLEMantle has been many years in the making by AMD, but we were not alone in this effort! Mantle was also directly shaped by the input we received from the greater game development community that has long sought a low-level graphics API for PCs.  We worked shoulder-to-shoulder with developers like DICE and Oxide Games to create Mantle in the image of their needs: a streamlined, robust, efficient API for modern graphics work. In fact, Mantle is the very first API designed directly by game developers for their modern craft!

 At the simplest level, Mantle is an Application Programming Interface (API), or a language that game developers can use to write code that creates the beautiful graphics on your screen. In its current iteration, the Mantle API uniquely leverages the hardware in the Graphics Core Next architecture (GCN) of modern AMD Radeon™ GPUs for peak performance.

 More broadly, Mantle is functionally similar to DirectX® and OpenGL, but Mantle is different in that it was purpose-built as a lower level API. By “lower level,” it’s meant that the language of Mantle more closely matches the way modern graphics architectures (like AMD’s own GCN) are designed to execute code. The primary benefit of a lower level API is a reduction in software bottlenecks, such as the time a GPU and CPU must spend translating/understanding/reorganizing code on-the-fly before it can be executed and presented to the user as graphics. Mantle comes in contrast to the “high level API,” which offers broader compatibility with multiple GPU architectures, but does so at the expense of lower performance and efficiency.

 Before entering into performance analysis, however, we wanted to provide some new insight on the design goals of the Mantle API.

 DESIGN PRINCIPLES OF MANTLEFirst and foremost, Mantle is primarily designed to improve performance in scenarios where the CPU is the limiting factor (so-called “CPU-bound” cases); CPU-bound scenarios are legion in gaming, as existing APIs typically have heavy validation overhead, along with difficulty scaling out to multiple CPU cores. In addressing this common problem, Mantle can enable a pronounced improvement for the majority of global PC gamers that have entry-level and mid-range processors. Some of the techniques to achieve this include:

• Low-overhead validation and processing of API commands
• Explicit command buffer control
• Close to linear performance scaling from recording command buffers onto multiple CPU cores
• Reduced runtime shader compilation overhead

 In turn, Mantle makes less of an impact in cases where high resolutions and “maximum detail” settings are used, as these settings are likely to be maximally taxing GPU resources in a manner that is more difficult to improve at the API level (so-called “GPU-bound” scenarios). While Mantle provides some built-in features to improve GPU-bound performance, gains in these cases are largely dependent on how well Mantle features and optimizations are being utilized by the developer. Some of those features include:

• Reduction of command buffers submissions
• Explicit control of resource compression, expands and synchronizations
• Asynchronous DMA queue for data uploads independent from the graphics engine
• Asynchronous compute queue for overlapping of compute and graphics workloads
• Data formats optimizations via flexible buffer/image access
• Advanced Anti-Aliasing features for MSAA/EQAA optimizations

 It’s also prudent to note that Mantle is still in the beta phase and may not reflect the full performance we might be able to achieve through the optimization time we’ll be investing in the months ahead. And, as developers are still familiarizing themselves with Mantle and its relationship to Graphics Core Next, they may not have capitalized on all available opportunities for optimizations—but that will come with time.

 One such optimization is the approach to multi-GPU performance scaling, which now rests in the hands of the game developer in the Mantle ecosystem. Developer control of multi-GPU performance empowers them to design an optimal multi-GPU codebase that perfectly matches the approach their rendering engine takes to graphics. Battlefield 4 is currently enabled with multi-GPU capabilities on Mantle, but the Oxide Games StarSwarm demo will be enabled with these capabilities in a later build.

 MANTLE ON BATTLEFIELD 4A game that needs no introduction, Battlefield 4 has captivated gamers with its intense multiplayer environments and, of course, the tremendous graphics courtesy of the game’s Frostbite 3™ engine. Augmented this week with Mantle, configuring Battlefield 4 to activate our API requires that you meet a few prerequisites:

• The latest game patch, available today from EA Origin™
• The latest graphics driver, AMD Catalyst™ 14.1 Beta
• A Graphics Core Next-based GPU
  
  

 MANTLE ON STAR SWARM
Star Swarm, meanwhile, is a technical demo developed by the incredible minds over at the new Oxide Games. Composed of industry veterans from Firaxis and Microsoft studios, the experienced artists and developers at Oxide have crafted the “Nitrous” engine to power a new generation of RTS titles, like the StarSwarm demo, with massive battlefields and a huge quantity of on-screen units—a perfect use case for Mantle!

 Star Swarm is important because it shows the impact Mantle can make in scenarios with a high number of “draw calls,” which are instances where the GPU and the CPU must communicate to display any object you see on screen. Traditional game engines based on DirectX® typically hit a limit somewhere around 5000 draw calls (or lower, depending on the user’s CPU), demonstrating severe performance degradation beyond that point.

 This performance degradation is due largely to the inability to efficiently utilize multiple CPU cores, which artificially limits the speed at which the CPU and GPU can communicate to do meaningful work. In contrast, the Star Swarm demo from the team at Oxide uses Mantle’s efficient multi-core scaling to raise that draw call limit into the neighborhood of up to 100,000 objects! With that in mind, tests performed on a range of CPUs demonstrate impressive performance improvements, even with significantly higher visual fidelity.
To test the Star Swarm demo from yourself, you’ll need to meet the following prerequisites:

• The latest build of the demo, available today on Steam™
• The latest graphics driver, AMD Catalyst™ 14.1 Beta
• A system with a Graphics Core Next-based GPU

 UPDATED FRAME PACING
Frame pacing is a technology that allows a multi-GPU configuration to deliver each frame of your game at consistent intervals. Even duration between each frame in your overall frame rate imparts a certain buttery smoothness that’s hard to describe in words, but luxurious to experience in-action in your favorite game.

 Last year, we developed and released our own implementation of this technology for customers of AMD Radeon™ products in the AMD Catalyst™ 13.8 Beta driver. In that update, we enabled frame pacing for Graphics Core Next-based GPUs on resolutions up to 2560x1600 in DirectX® 10 and 11 applications. With that release, we also promised a “phase 2” driver that would address additional configurations, and we’re pleased to say that today is the day: AMD Catalyst™ 14.1 Beta is the promised “phase 2” driver!

 AMD Catalyst™ 14.1 Beta brings new frame pacing support to the following AMD CrossFire™ or AMD Dual Graphics configurations when running DirectX® 10 and DirectX® 11 applications at resolutions higher than 2560x1600:

• AMD Radeon™ R9 280X Graphics
• AMD Radeon™ R9 270X Graphics
• AMD Radeon™ R9 270 Graphics
• AMD Radeon™ HD 7000 Series
• AMD Radeon™ HD 7000M Series
• AMD Radeon™ HD 8000 Series
• AMD Radeon™ HD 8000M Series
• These AMD Dual Graphics configurations
• AMD Dual Graphics configurations using the new “Kaveri” APU with the AMD Radeon™ R7 240 or R7 250 GPUs

 As a visual example of the benefits provided by frame pacing, we can plot out a real-world gaming scenario. The following diagram represents the time, in milliseconds, it took to render and display each frame to the user. To interpret the graph, look at the peaks and valleys: the line presenting fewer periods of frequent peaks and valleys, particularly fewer dramatic spikes, is representative of a smoother gaming experience.

 The below image illustrates an AMD Dual Graphics configuration: pairing the new “Kaveri” APU and the AMD Radeon™ R7 250 to run Tomb Raider™. As you can see, the time to present a frame is noticeably improved with the AMD Catalyst™ 14.1 graphics driver (red line), which is consistently and significantly less erratic than its predecessor (blue).

 THE ROAD AHEAD
Throughout the months that have followed our October 2013 unveiling of Mantle, you have been patient and kind to us as the Mantle consortium labored to make the first release the best it could possibly be. Concurrently, your enthusiastic support on Facebook, Twitter and at trade shows has been real and personal encouragement for every person working on the API. While we can never truly repay your kindness with a piece of software, we hope that it goes into the world with no uncertain amount of gratitude for us. We thank you so very deeply for your support, and vow that we will bring support to the full breadth of Graphics Core Next GPUs in the months ahead.

 More broadly, AMD Catalyst™ 14.1 can be a significant improvement for gamers with systems based on AMD CrossFire™ or Dual Graphics technologies, with the potential for vast enhancements to the overall fluidity of their DirectX® 10 and 11 gaming experiences.

 As this is a beta driver, we’d like to sign off by noting that your feedback is vital to the future of the AMD Catalyst™ driver. Please report any bugs or issues to our official reporting form!

 SUPPORTING RESOURCES

• A list of Mantle-enabled games
• Explore the official Mantle FAQ
• Read the “known issues” list for Mantle

 Robert Hallock is PR Manager for Gaming & Desktop Graphics at AMD.  His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites, and references to third party trademarks, are provided for convenience and illustrative purposes only.  Unless explicitly stated, AMD is not responsible for the contents of such links, and no third party endorsement of AMD or any of its products is implied.

SYSTEM CONFIGURATIONS FOR BATTLEFIELD 4™ and STARSWARM:

• Intel Core i7-4960X System: Intel X79 Chipset, 16GB DDR3-1600 RAM, Windows 7 x64, AMD Catalyst™ 14.1 graphics driver
• Intel Core i5-4670K System: Intel z87 Chipset, 16GB DDR3-1600 RAM, Windows 7 x64, AMD Catalyst™ 14.1 graphics driver
• AMD FX 8350 System: AMD 990FX chipset, 16GB DDR3-1600 RAM, Windows 7 x64, AMD Catalyst™ 14.1 graphics driver
• AMD A10-7700K System: AMD A88X chipset, 16GB DDR3-1600 RAM, Windows 7 x64, AMD Catalyst™ 14.1 graphics driver
  

• AMD A10-7850K, AMD Radeon™ R7 250, 16GB DDR3-1600 RAM, Windows 7 x64, AMD Catalyst™ 14.1 graphics driver, Resolution: 1920x1080, Preset: High Quality

 How about a vivid, practical example? Let's take a look at the remarkable Smart Sharpen feature. Smart Sharpen is a workhorse tool that lets users create crystal clear images by minimizing noise and halo effects.  To find the perfect image, the tool lets users choose the preferred algorithm and dynamically adjust the amount and radius of sharpening and noise reduction.
Early in 2013, AMD and Adobe collaborated to use OpenCL™ to help increase performance for just the “de-noise” portion of the Smart Sharpen feature. The results were good--we saw overall Smart Sharpen performance double simply by leveraging AMD graphics technology for this one portion of the Smart Sharpen pipeline¹.
With the newest release of Adobe Photoshop CC, AMD and Adobe developers looked deep into the code to help take maximum advantage of OpenCL™ across the entire Smart Sharpen pipeline and thereby fully leverage AMD GPU compute power. The results are nothing short of spectacular. Take a look at this video for comparison:

In fact, we found that the new AMD A10-7850K APU processed images up to 11 times faster with OpenCL On versus Off when we used it to process a 21 MP image with the new OpenCL™-optimized Adobe® Photoshop® CC Smart Sharpen feature². We also found up to 46 times faster performance³ on our top-of-line AMD FirePro™ W9000 card when compared with CPU-only processing, even with an image as large as 64 MP!  Yes, you read it correctly—up to 46 times faster!
What you are seeing is the impact of the close collaboration between AMD and Adobe to bring the promise of heterogeneous computing--where the system and software are designed to make the best, most intelligent use of all of the CPU and GPU resources on board--to the most demanding creative workflows. These speed boosts are critical to your workflow. They can mean the difference between having the freedom to adjust and tweak until you get the image exactly the way you want it or accepting a “good enough” outcome in the face of repeated processing delays and interruptions to your creative flow.
We are seeing great overall competitive performance on both benchmarks and application tests with our recently launched “Kaveri” products. This new lineup is topped by the AMD A10-7850K APU powered by 12 Compute Cores* featuring AMD Radeon ™ R7 Graphics for revolutionary, responsive computing and next generation gaming (*Visit AMD.com/computecores). 

    
Reviews so far include such accolades as the “Gold Award” from PC Perspectives, and “Dam Good Value,” and “Dam Good” awards from Hardware Canucks.  Joel Hruska from Extremetech wrote, “With HSA enabled, however, (Libre Office) performance skyrockets — AMD’s A10-7850K is more than 5x faster than the Core i5-4670. This, more than any other result, points to the potential of HSA.”

 Here are the results of our competitive testing for the new performance-enhanced Smart Sharpen now shipping with Adobe’s latest Photoshop CC): Here's a video showing the benchmark in action: The ultimate goal of all this is for the technology to keep pace with the user's creativity. We think this represents a dramatic step in the right direction.One of Adobe's most innovative and, frankly, remarkable new features for Photoshop CC is Perspective Warp, which lets the user adjust the perspective of a select part of an image while leaving other areas untouched to create the perfect image.

 Adobe developers have tapped OpenGL to take advantage of AMD GPUs to accelerate on-screen rendering so users can see and modify their results instantaneously. You will also find that same interactive, on-screen speed at work, with the outstanding new 3D beautify and 3D print preview features in the latest version of Adobe Photoshop CC.
The new release of Adobe Photoshop CC represents another important step forward in the development of intelligent application workflows that make full use of all the processing resources available in a system. Adobe continues to show real leadership in taking advantage of heterogeneous compute design. Working together, AMD and Adobe ensure that users’ creativity is never limited by performance.
Supporting Resources

• Press Release: AMD Delivers Peak Performance for Key Features and Creative Workflows in Latest Photoshop® CC
• Slides from AMD Developer Summit 2013 technical session: Smart Sharpen using OpenCL™ in Adobe Photoshop CC - Challenges and Achievements, by Joseph Hsieh at Adobe
• Smart Sharpen performance demo video from the AMD Developer Summit
• AMD Beats Intel with Photoshop CC video
  
• Adobe® Creative Cloud™ provides cross-platform acceleration for AMD graphics on both Windows and Apple computers, including the all-new Apple Mac Pro system with dual AMD FirePro™ graphics
• Details about AMD and Adobe
• For more information about AMD APUs
• For more information about AMD FirePro™ Graphics support for Adobe
• For more information about Adobe
• For more information about AMD

 1.AMD tests are performed on optimized AMD reference systems. PC manufacturers may vary configuration yielding different results. Test project used the new Smart Sharpen effect with Adobe Photoshop CC pre-release build, version 1.0.0.157 from 4/23/2013, test image: 5616x3744, 60.2MB, .PSD format. A notebook PC with AMD A10-5850M APU with AMD Radeon™ HD 8650G Graphics, 2x2GB DDR3-800 RAM, video driver 12.105.0.0 - 07-Apr-2013, SSD hard drive and Window 8 Professional (x64) build 9200 took 14.84 seconds with GPU acceleration on versus 31.03 seconds with GPU acceleration off. RIN-27

 2.The AMD A10-7850K APU can process images at up to 11 times faster with the  new OpenCL™- optimized version of the Adobe® Photoshop® CC Smart Sharpen feature. AMD tests are performed on optimized AMD reference systems. PC manufacturers may vary their configuration yielding different results. Test project used the Smart Sharpen feature in Adobe® Photoshop® CC, updated January 2014, test image: 5616x3744, 60.2MB, .PSD format.  A desktop PC with AMD A10-7850K APU with AMD Radeon™ R7 Series graphics, 2x4GB DDR3-2133 RAM, video driver 13.300.0.0 - 22-Dec-2013 (RC1), Window 8.1 build 9600, took 52.63 seconds with OpenCL off, 4.39 seconds with OpenCL on.  KVD-4

 3. The AMD FirePro™ W9000 graphics card can process images  up to 46 times faster with the 2013 preview of a new OpenCL™- optimized version of the Adobe® Photoshop® CC Smart Sharpen feature. AMD tests are performed on optimized AMD reference systems. PC manufacturers may vary their configuration yielding different results. Test project used December 2013 preview of a new OpenCL™- optimized version of the Adobe® Photoshop® CC Smart Sharpen feature, test image: 8000x8000, .PSD format.  A desktop workstation with Intel® Xeon® E5506 processor, AMD FirePro™ W9000 graphics card, 3GB DDR3-1600 RAM, video driver FirePro 13.152.4, Window 7-x64, took 529.5 seconds with OpenCL off, 11.2 seconds with OpenCL on.  FP-86

 4. The AMD A10-7850K APU can process images at up to 3.2X the speed of an Intel i5-4670K with the updated (January 2014) Adobe® Photoshop® CC Smart Sharpen feature. AMD tests are performed on optimized AMD reference systems. PC manufacturers may vary their configuration yielding different results. Test project used  the Adobe® Photoshop® CC Smart Sharpen feature, updated January 2014, test image: 5616x3744, 60.2MB, .PSD format.  A desktop PC with AMD A10-7850K APU with AMD Radeon™ R7 Series graphics, 2x4GB DDR3-2133 RAM, video driver 13.300.0.0 - 22-Dec-2013 (RC1), took 4.39 seconds with OpenCL on. A desktop PC configured with an Intel Core i5-4670K with Intel HD 4600 graphics, 2x4GB DDR3-1600, video driver 10.18.10.3379 - 17-Dec-2013, took 13.87 seconds with OpenCL™ on. Both systems used the same SSD hard drive and Window 8.1 build 9600.  KVD-3

 Back in the late 80’s I worked my way through my undergrad as a CNC machine operator for a manufacturer. Since then, I have only been a stone’s throw away from manufacturing.  I started as a systems engineer in a product company working with manufacturers to build chassis, populate motherboards and assemble systems for us.  Then I later became a product marketing manager working with ODMs to design devices into their products.  Now, in my current role, as the Strategic Marketing Manager focused on Industrial Control and Automation at AMD, one of my primary responsibilities is to track the existing and developing trends that are prevalent in Industrial Control and Automation applications.

 There are many different sources for information on these trends which include my personal experience and exposure to manufacturing operations, research studies, a variety of periodicals, numerous whitepapers, and webinars from a wide range of vendors. Taking all of this information and filtering out the individual opinions to drill down to the most relevant and reliable information can be a daunting task.  However, what I have learned is that there are nine existing and developing trends that seem to be having the most influence on manufacturing operations and industrial control and automation applications.  These trends include: efficiency, sustainability, connectivity, security, usability, safety, integration, scalability, and reliability.

 While there is certainly overlap in both the causes and the effects of these trends, and it can be argued whether these are good or bad trends, the bottom line is that these trends are driving the industry.  In some cases they are helping to drive resurgence of the American manufacturing industry.  In other cases they are helping to improve working conditions in factories.  In most cases, these trends are helping to make manufacturers more agile, eco friendly, and ultimately, more competitive.  Thus, I am working on figuring out what I can do to help drive these trends into the products that are being developed to serve industrial control and automation applications.  Whether you are directly involved in manufacturing, or just a stone’s throw away, take a look at my webinar to learn a little bit more about each of these trends and think about whether you should be considering them in the operations you control and products you design.

 Cameron Swen is a Strategic Marketing Manager for Industrial Control and Automation in the Embedded Solutions Business Unit at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites, and references to third party trademarks, are provided for convenience and illustrative purposes only. Unless explicitly stated, AMD is not responsible for the contents of such links, and no third party endorsement of AMD or any of its products is implied.

AMD's rich history in server-class silicon includes a number of notable firsts including the first 64-bit x86 architecture and true multi-core x86 processors. AMD adds to that history by announcing that its revolutionary AMD Opteron™ A-series 64-bit ARM processors, codenamed “Seattle,” will be sampling this quarter.

AMD Opteron A-Series processors combine AMD's expertise in delivering server-class silicon with ARM's trademark low-power architecture and contributing to the Open Source software ecosystem that is rapidly growing around the ARM 64-bit architecture. AMD Opteron A-Series processors make use of ARM's 64-bit ARMv8 architecture to provide true server-class features in a power efficient solution.

AMD plans for the AMD Opteron™ A1100 processors to be available in the second half of 2014 with four or eight ARM Cortex A57 cores, up to 4MB of shared Level 2 cache and 8MB of shared Level 3 cache. The AMD Opteron A-Series processor supports up to 128GB of DDR3 or DDR4 ECC memory as unbuffered DIMMs, registered DIMMs or SODIMMs.

The ARMv8 architecture is the first from ARM to have 64-bit support, something that AMD brought to the x86 market in 2003 with the AMD Opteron processor. Not only can the ARMv8-based Cortex A-57 architecture address large pools of memory, it has been designed from the ground up to provide the optimal balance of performance and power efficiency to address the broad spectrum of scale-out data center workloads.

With more than a decade of experience in designing server-class solutions silicon, AMD took the ARM Cortex A57 core, added a server-class memory controller, and included features resulting in a processor that meets the demands of scale-out workloads. A requirement of scale-out workloads is high performance connectivity, and the AMD Opteron A1100 processor has extensive integrated I/O, including eight PCI Express Gen 3 lanes, two 10 GB/s Ethernet and eight SATA 3 ports.

Scale-out workloads are becoming critical building blocks in today's data centers. These workloads scale over hundreds or thousands of servers, making power efficient performance critical in keeping total cost of ownership (TCO) low. The AMD Opteron A-Series meets the demand of these workloads through intelligent silicon design and by supporting a number of operating system and software projects.

As part of delivering a server-class solution, AMD has invested in the software ecosystem that will support AMD Opteron A-Series processors. AMD is a gold member of the Linux Foundation, the organisation that oversees the development of the Linux kernel, and is a member of Linaro, a significant contributor to the Linux kernel. Alongside collaboration with the Linux Foundation and Linaro, AMD itself is listed as a top 20 contributor to the Linux kernel. A number of operating system vendors have stated they will support the 64-bit ARM ecosystem, including Canonical, Red Hat and SUSE, while virtualization will be enabled through KVM and Xen.

Operating system support is supplemented with programming language support, with Oracle and the community-driven OpenJDK porting versions of Java onto the 64-bit ARM architecture. Other popular languages that will run on AMD Opteron A-Series processors include Perl, PHP, Python and Ruby. The extremely popular GNU C compiler and the critical GNU C Library have already been ported to the 64-bit ARM architecture.

Through the combination of kernel support and development tools such as libraries, compilers and debuggers, the foundation has been set for developers to port applications to a rapidly growing ecosystem.

As AMD Opteron A-Series processors are well suited to web hosting and big data workloads, AMD is a gold sponsor of the Apache Foundation, the organisation that manages the Hadoop and HTTP Server projects. Up and down the software stack, the ecosystem is ready for the data center revolution that will take place when AMD Opteron A-Series are deployed.

Soon, AMD's partners will start to realise what a true server-class 64-bit ARM processor can do. By using AMD's Opteron A-Series Development Kit, developers can contribute to the fast growing software ecosystem that already includes operating systems, compilers, hypervisors and applications. Combining AMD's rich history in designing server-class solutions with ARM's legendary low-power architecture, the Opteron A-Series ushers in the era of personalised performance.

 Lawrence Latif is the Manager of Technical Communications at AMD. His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites, and references to third party trademarks, are provided for convenience and illustrative purposes only. Unless explicitly stated, AMD is not responsible for the contents of such links, and no third party endorsement of AMD or any of its products is implied. This blog contains forward-looking statements concerning AMD, and features of AMD’s future products, the ability of AMD to win in traditional server segments with new Arm-based products in 2014, the benefits from AMD’s new technology partnerships and the timing of future products that incorporate AMD’s products, which are made pursuant to the safe harbor provisions of the Private Securities Litigation Reform Act of 1995. Forward-looking statements are commonly identified by words such as "would," "may," "expects," "believes," "plans," "intends," "projects," and other terms with similar meaning. Investors are cautioned that the forward-looking statements in this document are based on current beliefs, assumptions and expectations, speak only as of the date of this blog and involve risks and uncertainties that could cause actual results to differ materially from current expectations.