CEVA Dolphin Weninar SemiWiki 800x100 260419 (1)

Is the Intel Cash Cow in Danger?

Is the Intel Cash Cow in Danger?
by Daniel Nenni on 06-14-2016 at 4:00 pm

There was an interesting panel at the Silicon Summit sponsored by the Global Semiconductor Alliance (GSA) on “Designing for the Cloud.” It was led by Linley Gwennap (The Linley Group) with Ivo Bolsons (Xilinx), Ian Ferguson (ARM), and Steve Pawloski (Micron). Missing of course was Intel which derives close to 30% of its revenue from their Data Center Group (DCG), also known as their cash cow. There were however quite a few Intel people in attendance who were willing to talk, off the record of course.

Unfortunately the presentation slides are behind a firewall on the GSA site but I have them in front of me so I will give you a quick summary. But first, let’s talk about the Intel DCG and the changing cloud landscape.

DCG is recognized as one of the primary Intel growth engines along with IoT (which is a joke that I will cover another time). Unfortunately growth is decelerating from 18% in 2014 to 11% in 2015 and in the first quarter of 2016 DCG revenue growth is down to 8%. DCG revenue growth should still hit double digits this year but after that it is a little bit cloudy.

ARM based servers are definitely a threat. Ian Ferguson admitted that it has been 8 years since ARM entered the server business with little success (1% market share). The same could be said about ARM entering the mobile business many years ago with little success but they of course prevailed (99% market share). ARM has publicly stated that they are still aiming for 25% market share by 2020 and Ian shared the ARM Server Core Beliefs:

[LIST=1]

  • We will compromise on single threaded performance for non-linear gains in power efficiency (architecture licensees may make different trade-offs).
  • Right balance between standardization and innovation (Enable an ecosystem to coalesce with enough “greenfield” areas to empower opportunities for differentiation).
  • It’s about the system stupid (Once the CPU is “good enough” most important areas are memory, I/O, and on-chip hardware accelerators).

    According to Ian, boxes are being deployed worldwide with China leading the effort. Why China you ask? Because China, the most populous country in the world, consumes almost half of the semiconductors made and will have the single largest cloud demand in the world, absolutely. And China wants control over the silicon for security reasons to “avoid the perils of reliance on American technology.” This alone will maim the Intel cash cow.

    Qualcomm is the largest ARM licensee to announce a China JV (joint venture with Guizhou Huaxintong Semi-Conductor Technology) which will start by selling Qualcomm’s own server designs this year. This will be followed by new versions designed by the JV specifically for the China cloud and finally a completely new chip aimed directly at Intel.

    In addition to the ARM threat there is also a joint development agreement between China and the IBM OpenPOWER foundation for server chips. IBM was at the Design Automation Conference last week openly recruiting people in Austin for the China JV.

    AMD also announced a joint venture with China which includes x86 processor technology for server chip development. Again, China wants to make their own chips, which is also why China is allowing TSMC, GlobalFoundries, and others to build fabs in China.

    Given all that, plus Intel losing the process lead at 10nm and 7nm to TSMC and Samsung, continued double digit growth for Intel’s primary growth engine (DCG) is very hard to believe, just my opinion of course.


  • Is the U.S. ready to adopt a new financial model to support microelectronics?

    Is the U.S. ready to adopt a new financial model to support microelectronics?
    by Tom Dillinger on 06-14-2016 at 12:00 pm

    Amidst all the active news about new process introductions at 16/14/10/7nm and the status of next-generation lithography development, there was a recent press release that could have as large an impact upon the microelectronics industry in the United States. A groundbreaking ceremony was recently held in Marcy, New York for a new fab, to be run by ams AG, formerly known as austriamicrosystems AG (link here).

    ams AG focuses on the design and manufacture of analog IC’s, with specific emphasis on sensors, power management, and wireless networking applications.

    This fab will initially focus on products at the 130nm process node, with plans for more advanced nodes in the future. The goal is to be operational by YE’2017, with high volume manufacturing commencing in early 2018.

    There are two aspects to this announcement that are particularly noteworthy, IMHO.

    First, there remain significant capacity constraints industry-wide at older process nodes.

    Second, the financial arrangement provided to support this new fab is extremely unique, at least in the United States. The land and 360,000 sq. ft. building complex located at the SUNY Polytechnic Institute will continue to be owned by SUNY, with the facility available to ams AG on a 20-year lease “at very attractive rates”.

    ams AG will invest ~$2B to ramp the new facility, with a commensurate number of new jobs, estimated at ~700. (Indirect jobs outside ams AG to support this facility will increase the total additional employment.)

    To be sure, there have been business incubator facilities established as extensions to public institutions, such as universities. The goal of these relatively small sites is to facilitate the spin-off of new entrepreneurial pursuits, usually based on advanced academic research. These university parks offer low overhead, attractive financial terms for these small firms.

    Yet, to my knowledge, this is the first example in the U.S. of a public institution establishing a build/lease financial model for an established microelectronics company to equip and maintain a new fab for high volume manufacturing.

    When the deal was first announced last year, the SUNY Poly site development team said, “At the direction of the Governor, SUNY Poly has entered into a strategic research, development, and manufacturing partnership with ams AG.”

    Governor Cuomo himself said, “This is a transformative moment, that will make a difference in peoples’ lives in the Mohawk Valley for generations to come.”

    New York, and specifically, Governor Cuomo, have been spearheading public-private partnerships in the microelectronics industry, as led by the Nanotech Initiative at SUNY-Albany. This latest deal takes the financial relationship for a production manufacturing fab to a new level.

    Is this new partnership model for microelectronics indeed something that public resources in the U.S. will increasingly support? Is the U.S. at a disadvantage compared to other countries due to its reticence to invest in these kinds of partnerships? Will other states and/or the federal government be willing to make similar economic investments to grow semiconductor manufacturing here?

    How do public taxpayers minimize the risks of financial loss in these ventures? (Although not in the microelectronics area, the economic stimulus provided by the federal government to Solyndra in 2009 comes to mind.)

    Time will tell, I suppose, whether the New York-ams AG partnership is indeed a public-private “win-win”. It is definitely a new, innovative approach toward revitalization of a region and of a domestic manufacturing industry.

    PS. It was reported that Governor Cuomo was at one time considering a Democratic Presidential campaign in 2016 – that would no doubt have resulted in some interesting economic development-related discussions among the candidates. 🙂

    -chipguy


    Climbing the Infinite Verification Mountain

    Climbing the Infinite Verification Mountain
    by Bernard Murphy on 06-14-2016 at 7:00 am

    Many years ago I read a great little book by Rudy Rucker called “Infinity and the Mind”. This book attempts to explain the many classes of mathematical infinity (cardinals) to non-specialists. As he gets to the more abstract levels of infinity, the author has to resort to an analogy to give a feel for extendible and other cardinal classes.

    He asks you to imagine climbing a mountain called Mount On (you’ll have to read the book to understand the name). You climb at infinite speed and, after a day’s climbing, you look up and see more mountain stretching ahead, looking much the same as the stretch you just climbed. Anyone who’s ever climbed a mountain or a hill knows exactly how this feels. The author’s point is that no matter how fast you climb, there’s always more mountain ahead – it seems like you never get close to the top.

    The reason I bring this up is that verification feels very similar, largely because the problem space continues to explode faster than we can contain. Cadence hosted a thought-provoking lunch at DAC this year (Seamlessly Connected Verification Engines? What Does It Take?) which I felt was very relevant to this topic. Jim Hogan opened and underlined the challenge. The amount of knowledge/data we have to deal with is increasing supra-exponentially. Today it’s doubling every 13 months. By 2020 when the IoT is in full swing, it is expected to double every 12 hours. The capabilities we will need to handle that volume with high quality are going to demand a (currently) almost inconceivable level of verification, touching almost everything in the stack. We’re going to have to seriously up our quality game, as Jim put it.

    On that daunting note, the panel started with where the tools play best today. Alex Starr (Fellow at AMD) felt that formal was primarily useful for IP and subsystem, emulation is good for system but (software-based) simulation is struggling everywhere. That said, simulation still has an advantage over emulation in being able to verify mixed-signal designs. But a better and longer-term AMS solution may be to model virtual prototyping together with analog models in order to get system-level coverage with mixed-signal. AMD has invested over several years to make this work and it’s now paying off.

    Narendra Konda (Head of Verification at NVIDIA) further emphasized the need for tools in the verification flow to play together, especially with virtual prototyping (VP). He pointed to the need for VP to play well with emulation and FPGA prototyping (FP), also for assertion-based verification (ABV) to play well in these flows. They are simulating 1B gates with a rapidly growing software stack across large banks of test-cases. They have to put a lot of this in VP to get practical run-times.

    The perennial question of team interoperability came up of course. You can make the tools completely interoperable but that doesn’t help as much as it could if design and verification teams stick to their silos of expertise. Alex agreed this could be a problem – a good example is in power management verification where you have to span all the way from software applications potentially to physical design. His view was that this takes education and of course improvement in the tools, like portability of test cases.

    Narendra took a harder line; designers don’t get to have comfort zones – they adapt or die. Fusing tools together is the bigger problem, but it is being solved by stages. For NVIDIA it took 1½ years working with Cadence to get VP, emulation and FP working well together. He views this collaboration as a necessary price for staying ahead. They had the same problem 2-3 years ago with ABV and the same approach to solving the problem – now this is also starting to work. Dealing with automotive safety and reliability requirements is a new frontier; there is no productive tool today in this area. Narendra expects this will take another 8-10 months to get to a solution.

    The panel wrapped up on current deficiencies in cross-platform methodologies. All emphasized more need for standards, especially for interoperability between platforms from different vendors. Some of that is happening with the Portable Stimulus standard, but more still needs to be done, for example in normalizing coverage metrics between vendors.

    In verification, performance is never where it needs to be. Narendra saw a need for something in-between the speed of emulation and prototyping (with emulation setup times and debug-ability of course). He felt doubling the current speed of emulation would help. Alex agreed and also said that for serious software regression, emulation and FP aren’t fast enough. There needs to be more of an emphasis on hybrid modeling between hardware platforms and VP, where it’s more feasible to get within range of real-time performance. This echoed Narendra’s earlier point about the need for VP hybrid modeling.

    I found continued references to VP here and in other meetings particularly interesting. Software-driven verification with bigger software stacks and more test-cases really do drive a need to model more of the system in VP-like platforms, dropping into FP and emulation where needed. This need can only grow. Perhaps VP is becoming the new emulation and emulation is becoming the new simulation.

    The vendors are doing a great job advancing what they do, and they’re clearly partnering effectively with customers to build those solutions, but the top of the verification mountain keeps receding into the clouds (there’s a pun in there somewhere) and probably always will. Meantime you can read more about NVIDIAs success with Cadence emulation HERE.

    More articles by Bernard…


    Top Ten #53DAC Highlights

    Top Ten #53DAC Highlights
    by Tom Dillinger on 06-13-2016 at 12:00 pm

    Here is a very subjective list of the Top 10 logistical and technical highlights from DAC’53.

    (10) With DAC attendance down from its peak days, the Austin Convention Center served as an excellent venue. There was good participation from companies with design centers in the “Silicon Hills”. And, I saw colleagues from Silicon Valley, SoCal, and overseas who made the trip, as well. Kudos to the DAC Committee and sponsoring companies for the free parking!

    (9) SWAG is actually an acronym, short for Souvenirs, Wearables, And Gifts.

    I’ve never really understood why exhibitors give away pens, stuffed animals, stress-relief squeeze balls, even drumsticks. Then again, since I’m mentioning it, I guess this form of advertising works.

    (8) The $99 Design + IP Track registration fee is a great deal.

    Although 15 minutes is perhaps a bit too short for many Design Track presentations, there is an opportunity to follow up with the presenter at the evening poster session. Now that the conference Proceedings are distributed electronically, it would be nice if the Design Track and IP Track slides were available for review prior to the conference, rather than posted afterwards.

    (7) The committee needs to establish better guidelines and standards for presentation charts – there are still simply too many slides that are illegible from the back of the room.

    (6) The keynote from Professor Ken Shepard at Columbia Univ. was fascinating. The potential to harvest (and store) electrical energy by controlling biological cell membrane ion transport to power electronics could open up new opportunities, despite the very low duty cycle of actual computation to energy charging.

    (5) As you might imagine, the growth of electronics in automotive applications was a prevalent theme of the conference. And, not all the focus was on autonomous driving. Monday’s keynote from Lars Reger at NXP provided several illustrations of advanced inter-vehicle and infrastructure communications that would also provide significant benefit.


    Take the NXP “RoadLINK” feature. For example, you’re likely aware that race cars draft behind a leader to save fuel – consider the fuel-saving benefits of a set of freight trucks linking into a “platoon” (or “convoy”, for you old-time CB’ers). A rather aggressive distance is maintained between the trucks, enabled by syncing up to the forward vision and braking controls of the lead truck.

    And, if the infrastructure were in place, stoplight timing in a city would be adjusted to green light the oncoming platoon. (Here’s a link for more info.)

    I concluded there are lots of technology options in the automotive area with significant ROI, but am also anxious about the tremendous coordination required in this country among political and industry organizations to address the corresponding costs and schedule. (Europe seems to be well ahead of the U.S. in this regard.)

    Coincidentally, this month’s issue of IEEE Spectrum has an enlightening article about the emerging concerns about vehicle manufacturer and driver liability in a world with increasing automotive automation (especially, relative to the vehicle software). I was optimistic that my car insurance rates would go down in the future – now, I’m not so sure. 🙂

    (4) IP security was also a major topic of discussion. Clearly, the IP providers are concerned about the potential loss of revenue due to IP theft, and the loss of traceability for edits to (configurable) IP variants.

    One paper I attended presented a proposal for collaboration between IP providers and tool developers, to support the use of encrypted IP models. The presenter reminded the audience that encryption will be needed across the gamut of models. He highlighted that encryption would not only apply to functional verification and synthesis, but also the full DFT tool suite. (I’m still trying to get my head around an encrypted fault diagnosis tool.)

    (3) On the EDA vendor exhibit floor, electrical analysis of SoC’s was the hot topic.

    It should be no surprise to regular Semiwiki readers that I*R power distribution voltage drop and power/signal ElectroMigration (EM) are critical sign-off steps in current fab processes.

    Specifically, to prevent long iterative loops between implementation and electrical analysis, the physical design platform needs to integrate “in-design” analysis engines, ideally with corrective recommendations provided to physical designers.

    (2) To address the burgeoning IoT application market, it should also be no surprise that power management and power optimization techniques were front-and-center. Several designs had adopted increasingly complex clock grid/tree distribution methods, with multi-level gating circuit topologies.

    Of note was that these clock networks had to be successfully managed across multiple hierarchical levels of the physical floorplan, and (ideally) with commonality across multiple instances of the same core IP.

    The handling of global signal/clock buffering and routing passing through and around block designs remains an intricate data management problem.

    (1) And, to me, perhaps the most exciting announcement at DAC could be found at Samsung’s exhibit, where they were demo’ing an SoC with an integrated magnetoresistive (MRAM) memory array. This very unique and attractive technology option will be available from Samsung Foundry in 2018 – look for another Semiwiki article on this announcement shortly.

    DAC has certainly evolved throughout the years. It remains the most important event for our industry, to learn of the newest developments – whether that be new foundry technologies, new EDA vendor tool features, or unique and interesting design approaches. The glimpse into active research and its potential future applications is pretty cool, too.

    See you at DAC’54 in Austin next June.

    -chipguy


    The Evolution of Emulation

    The Evolution of Emulation
    by Bernard Murphy on 06-13-2016 at 7:00 am

    Mentor hosted a panel on emulation in their booth at DAC this year. One thing I really liked about this panel is that it didn’t include anyone from Mentor. Not that I have anything against Mentor employees, who are a fine bunch of people from those I know, but I find panels most interesting when the discussion is purely among customers. Lauro Rizzati moderated, which is a bit of a cheat, since he consults for Mentor, but moderators don’t do much of the talking, so I’ll count it as a customer panel.

    Lauro opened with a quick history of emulation, starting with in-circuit-emulation (ICE) mode, later moving to more general application in simulation acceleration which then evolved to transaction-based emulation, followed by virtualization and then increasingly adding application areas like power modeling, network modeling and more. What he wanted to do was to explore how this broad range of usage is evolving among 3 of Mentor’s customers: Rick Leatherman (Director of developer tools at Imagination Technologies), Guy Hutchison (VP of hardware engineering at Cavium) and Alex Starr (Fellow at AMD) – left to right above, with Lauro rightmost.

    Alex said that AMD has been using emulation for many, many years. They started in ICE mode but have evolved to transaction-based and hybrid models, both at the IP and system levels. He added that software-driven verification increasingly demands use of emulation. Guy said that Cavium uses emulation all the way through the design cycle and they use it purely in virtual mode. ICE mode is not practical since they don’t feel there is any way to generate realistic traffic from hardware. Rick said Imagination/MIPS has been using emulation for many years, starting in ICE mode, now moving to transaction-based.

    Alex added that they still do a lot of simulation – both full-chip and IP. They do more emulation work at the platform level, as a part of the never ending effort to shift left. Software and firmware teams have been using emulation for a long time for this reason, and are increasingly using emulation in hybrid mode. Guy said Cavium only uses emulation for full-chip verification, which they break into 3 phases: performance verification, software bring-up and validation and post-silicon validation (back-porting silicon problems into emulation for debug). For Rick, bringing up software as fast as possible is very important. While most of us view Imagination as an IP company, increasingly they are providing more complete systems with software stacks for IoT markets, where system with software validation and power modeling become essential.

    On ICE versus virtual modeling, Alex felt these complement each other and hybrid modes continue to be relevant. He cited a hard disk device as an example of a component still best modeled in ICE. But he and others agreed that virtual mode fixes a lot of problems – reliability, debug, remote access, easy sharing of resources and saving and replaying state (for debug). From Guy’s perspective, only virtual mode is practical – again they don’t feel it is possible in their application to model realistic traffic through hardware.

    Lauro then asked about use of emulation in applications domains – power and DFT testing for example. Alex said they run both DFT and Design for Debug verification in emulation and have done for some time. Power analysis is becoming increasingly important, and the intersection between power and DFT – looking for peak power spikes in test mode – is a good example of of an area where emulation shines. Both Guy and Rick added that they are using emulation for power analysis.

    Where does emulation not help? Everyone agreed that analog/RF modeling was out of scope today. For example, verifying memory training software with hardware models for DDR is something for which AMD has had to build internal solutions. Of course if you can extract digital models from analog blocks, some cases might be amenable to emulation, but hand-creating models for emulation just moves the problem to validating the accuracy of those models.

    Overall an encouraging reality check on where emulation is at, where it’s headed and where there is still work to be done. Virtual is gaining ground fast, ICE still has its place and analog is still not a part of the solution. You can read more on Mentor’s view of the evolution of usage models in emulation HERE.

    More articles by Bernard…


    Six Reasons to Rethink Power Semiconductor Packaging

    Six Reasons to Rethink Power Semiconductor Packaging
    by Alex Lidow on 06-12-2016 at 8:00 pm

    In my 40 years’ experience in power semiconductors I have visited thousands of customers, big and small, on every continent except Antarctica. When the issue invariably turns to the packaging of the power semiconductor – transistor, diode, or integrated circuit – the requests for improvement fall into six categories:
    Continue reading “Six Reasons to Rethink Power Semiconductor Packaging”


    3 reasons why diode-based ESD protection ruins the IoT experience

    3 reasons why diode-based ESD protection ruins the IoT experience
    by bkeppens on 06-12-2016 at 4:00 pm

    The ‘Dual diode’ approach is one of the most used on-chip and off-chip concept for ESD protection of IO interfaces. It is simple to implement, smaller than any other IO/ESD concept, has a low parasitic capacitance and low leakage.
    Continue reading “3 reasons why diode-based ESD protection ruins the IoT experience”


    New NVIDIA Tesla M10 Could Drive Enterprise VDI Reassessment

    New NVIDIA Tesla M10 Could Drive Enterprise VDI Reassessment
    by Patrick Moorhead on 06-12-2016 at 12:00 pm

    NVIDIA is well known for its leadership in graphics processors (GPUs) for gaming, but their business is quickly diversifying with significant growth in other areas like their datacenter and automotive businesses. Within the datacenter, NVIDIA has been evangelizing a vision for a number of years about the benefits of GPUs for accelerating workloads in high-performance computing and web-scale datacenters. The company has recently started to see traction for GPU acceleration in the datacenter business for applications like deep learning and artificial intelligence.


    Graphics for all (Image credit: NVIDIA)

    While these application areas are certainly growing, the market opportunity is still a relative niche within the overall datacenter market. In parallel to these efforts, NVIDIA looks to capitalize on a vision to target mainstream enterprise datacenters with their NVIDIA GRID solution for virtualized desktop infrastructure (VDI) through partnerships with virtualization software vendors like Citrix Systems, Microsoft, and VMware. With NVIDIA’s new Tesla M10, they are doing just that.

    VDI is a virtualization technique that allows access to a virtualized desktop, which is hosted on a remote service over the Internet, by a client device. While this market has been around for a number of years, VDI could be a growth opportunity over the next few years driven by several key factors. Here are just a few of the important market drivers I see today.

    First, a massive movement by enterprise IT to private clouds means a move away from a traditional client server model to a model where heavy computation and data storage happen in a central resource accessible by many users rather than just one. With this in mind, enterprise IT may consider VDI as an extension of the cloud model for certain uses cases.

    Second, there is a direct relationship between acceptance of a VDI experience and the speed of the Internet. Look at what is happening with the 4G and LTE roll out. And also look at the backhaul improvements. Increased internet speeds equate to a better VDI experience, all things equal.

    And last but certainly not least, there is a cultural shift in the workplace that is driving IT to do things differently. Millennials have entered the workforce in droves and have climbed the corporate ladder into key management positions. This generation expects to use any device from any location to access corporate information, which has accelerated the BYOD movement. Also, millennials have grown up in a technology centric world and will work or not work for a company based on how good their IT services and equipment are. This means that users want to access to technologies that provide the best user experience for their applications.

    A few years ago, NVIDIA GRID was launched as a platform that allowed IT to bring virtualized graphics capability to more business users. When NVIDIA GRID was launched, graphics processing for business was a relative niche market for applications that required high-fidelity 3D graphics like EDA, CAD, simulation tools, and video editing workloads. But now, use of graphics acceleration is expanding to improve the experience of office productivity applications like Outlook, Office 2016, web browsers, Adobe Photoshop, and the Windows 10 operating system. And by some estimates, more than half of enterprise users access at least one graphics accelerated application. But providing client devices with dedicated GPU capability to large groups of knowledge workers does not make economic sense for enterprise IT.

    One of the key drivers for the use of GPU acceleration in a VDI environment will be giving mainstream IT organizations the right economic models to provide this capability to business users. With this in mind, NVIDIA GRID uses a “good, better, best” approach that allows IT to carve up the GPU resources based on need in a way that makes economic sense, with different offerings to target different user levels (knowledge workers, power users, designers). Today’s GRID announcement enhances this approach by bringing Tesla M10 as an offering in the GRID this fall, which will let IT provide more employees with virtualized graphics (up to 128 users per server) at a lower cost per user. In addition, NVIDIA is providing a new annual subscription pricing model for GRID that includes access to software upgrades for the life of the subscription. The annual subscription has the potential to provide a lower overall cost for IT over the traditional perpetual pricing model with starting costs as low as $2 per user per month for the basic service.

    I think the time is right for IT organizations to reassess their VDI strategy. And NVIDIA’s “graphics for all” strategy with the NVIDIA GRID makes sense to help IT organizations use low-risk economic models to provide GPU acceleration capability to a wider range of business users.

    More from Moor Insights and Strategy


    AMD Brings More Console Features To PC Gaming

    AMD Brings More Console Features To PC Gaming
    by Patrick Moorhead on 06-12-2016 at 7:00 am

    Advanced Micro Devices made a big press splash when they won all three major game console platforms. Their presence inside the Microsoft Xbox One, Sony PS4 and Nintendo Wii U has benefited the company financially at the operating income level and has governed how they design their future GPUs. As Advanced Micro Devices has remained with the GCN architecture, present inside of all three consoles, it would seem that Advanced Micro Devices would have had a built-in an advantage into their GPUs for the PC gaming space. You see, the X86 CPU instruction is the same and platforms also share APIs and tools more than ever. This advantage has been touted by the company for quite some time now, but they have been challenged to transfer the benefits of these design wins to the PC gaming space. With AMD’s new feature called “Shader Intrinsic Functions”, a new feature pulled directly from the console space, this could change.


    Image credit: AMD

    Mantle was the first step
    Advanced Micro Devices’s first attempt to harness this console advantage was the creation of Mantle which gave PC developers low-level access to Advanced Micro Devices hardware and enabled console-like direct hardware metal. While Mantle wasn’t perfect, it did get implanted in a few games before it was eventually integrated into an industry standard now known as Vulkan. I believe Mantle was also the timing driver for Microsoft DirectX 12. Vulkan takes many, but not all of Mantle’s features and ideas and implements them cross-platform allowing mobile GPUs and desktop GPUs to share the same APIs and improved, low-level access.

    Shader Intrinsic Functions
    As a result, Advanced Micro Devices is introducing many new features within their GPUOpen program to enable more low-level access to hardware in a console-like development environment. If you are unfamiliar with GPUOpen, I wrote on that here. AMD is introducing the support for what is called “Shader Intrinsic Functions” or built-in functions which allow the developer to directly access graphics hardware instructions. This access is granted to developers in situations where they would normally be abstracted by a higher level API, or not available at all. Advanced Micro Devices likens this to embedding optimized machine language code into higher-level code.

    Intrinsic functions are only useful if game developers know they can rely on them to exist in future hardware and that they will be able to gain direct access through hardware-specific code paths throughout the generations. Because Advanced Micro Devices has stuck with GCN, they make it possible for developers to invest in the use of built-in functions and improve performance. By enabling developers to utilize these new capabilities, Advanced Micro Devices is once again finding ways to enable their hardware to experience performance gains without any changes to their hardware whatsoever.

    In addition to improving performance on Advanced Micro Devices GPUs, the addition of intrinsic functions into GPUOpen also improves the compatibility of this feature. Because the addition of intrinsic functions is expected to run as a part of GPUOpen and exist on the PC platform, it will be supported in all the major PC APIs including Microsoft DirectX 11, DirectX 12 and even Vulkan.

    By having intrinsic functions work inside of all these graphics APIs, AMD is really making it an attractive play for developers that want to squeeze out that extra performance but maybe don’t want to be limited to a single set of APIs if they want to recycle or repurpose code. As far as we can tell, NVIDIA also has support for intrinsic functions inside of their GPUs as well, but they seem to be more focused on CUDA code and not for game development.

    Wrapping up
    Advanced Micro Devices’s introduction of intrinsic functions into GPUOpen is the start of the company really delivering on their promise of benefiting from all of the console design wins. It is also the concretization of the GPUOpen philosophy to enable console-like access to the PC platform. Sure, Advanced Micro Devices has made a lot of operating income from building the consoles, but when you consider how many console ports end up on the PC, there is no doubt that there should be some level of translation of that to AMD‘s benefit.

    There’s unfortunately no way for us to test these features since they have to be implemented by game developers first, so we will have to wait and see. We may finally start to see AMD’s benefits from the console design wins come to fruition with these new intrinsic functions and I expect that this isn’t the end of such features coming to GPUOpen and that we can expect more to come relatively soon. AMD has a video on the new feature if you want to check it out.

    I like this play from AMD and hope to see more like it in the future.

    More from Moor Insights and Strategy