Advanced Micro Devices' Biggest Catalyst For 2014 Is The 'Little' Chip

| About: Advanced Micro (AMD)


AMD's "Jaguar" architecture is, in my opinion, one of the company's most competitive products.

The TAM (Total Addressable Market) for the Jaguar chip, based on number of units, is larger for Jaguar than most other products.

As Jaguar derivatives roll out for the desktop, AMD will also be refreshing the mobile versions of the chip.

This month, AMD is scheduled to roll out socketed versions of the company's Jaguar PC chips, aimed at the budget end of the desktop PC space. Also, during the first half of 2014, AMD is scheduled to release updates to the Jaguar architecture, codenamed Beema and Mullins, aimed at mobile solutions.

The combination of these two events will represent a larger TAM than other near-term impending product launches or developments. As proof, consider the PC market, which is roughly ~80M units per quarter, compared to the discrete GPU market, which is around 10M-15M units per quarter (source: IDC, Jon Peddie Research). Concerning consoles, just using a simple average starting late 2004 yields about 18M or so consoles sold annually (source: VGChartz).

The PC market represents AMD's biggest opportunity near term, so opportunities in this space warrant a closer look.

The Stability of Jaguar

David Kanter of Real World Tech recently published a technical, detailed write-up of the Jaguar architecture. The article is definitely worth a read, but for the purposes of my write-up, I will include his concluding paragraph:

"One of the interesting side-effects of AMD's success with Jaguar for consoles is the future trajectory of the core. Normally, an x86 microarchitecture stays relevant for 2-3 years before it is obsoleted by derivatives - but there are reasons to believe this will not be the case for Jaguar. Since all the high-performance consoles will rely on Jaguar, game developers will inevitably optimize their code to extract the maximum performance. The lifetime of these consoles is roughly 5-7 years, which means that game developers will be producing code optimized for Jaguar for the better part of the next decade. This gives AMD tremendous incentive to keep on shrinking and optimizing the Jaguar core, with minor revisions. In theory, this means that Jaguar derivatives should eventually break the 1W barrier and become suitable for extremely low power applications, but only time will tell."

If you divide up the tasks that typical PC users perform on a given microarchitecture, you are likely to come up with a list that sounds something like internet browsing, productivity, and gaming. The most taxing of these, depending on the user, is either gaming or productivity.

Because the Jaguar microarchitecture is at the heart of the next generation consoles, it is the target platform for a huge percentage of games for the foreseeable future (think 3-5 years or longer, barring a massive shake-up in the gaming industry).

Most games currently like single-threaded performance, and this is one metric in which AMD has a hard time competing with Intel (NASDAQ:INTC). Because the consoles are designed using 8 smaller cores, developers must ensure games and game engines are designed to more readily spread the workload, rather than relying heavily on a couple of threads (more on this later).

Source: AMD's CES 2014 Presentation via Slideshare

Market share can be measured in various ways, and the one I often see used to support arguments against AMD is to not include console APUs in x86 chip sales. However, by not including console chips, I feel a larger shift is missed.

Let me demonstrate by picking apart one argument against AMD's Mantle API: How could Mantle be successful given that AMD is the minority in both PC market share and GPU market share? Why would developers want to code software geared for AMD hardware given AMD hardware represents the smallest audience?

To be fair, on one hand I feel this is a perfectly valid argument, particularly in the near term. Nvidia (NASDAQ:NVDA) has a larger GPU market share, and Intel has a larger PC market share. Mantle's largest gains are shown when pairing a relatively weaker CPU core with a stronger GPU by reducing CPU overhead and allowing the GPU to "stretch its legs." Add to this that Mantle only works with GCN powered cards, and we arrive at a small sub-section of the GPU market.

However, by excluding console APU sales from x86 PC chips, or by only focusing on quarterly data, I feel the real point is missed: the Jaguar architecture is the target platform for the majority of new major game titles, and the installed user base is growing rapidly.

Games are not typically designed from scratch each time, but are rather made using a game engine. Studios such as DICE (Frostbite engine) or Epic Games (Unreal Engine) build these engines, and game developers then use these engines to create the games. These game engines are then used to develop both console games and their respective PC ports.

Meaning regardless of how developers accomplish it, it behooves them to try and extract as much performance from their respective game engines as possible. This is accomplished by ensuring more equal thread utilization across each CPU core, making single-threaded performance less relevant. PC market share is largely irrelevant in this scenario as well, as the consoles (powered by AMD) are the target platforms for many games developed. Consoles are based on 8 Jaguar CPU cores.

Enter Mantle *or* DX12 for the PC Space

As a quick aside, there has been much recent press surrounding AMD's Mantle API, along with Nvidia performance gains in DX11 and Microsoft's (NASDAQ:MSFT) incoming DX12 API.

Source: Dice/EA Presentation via SlideShare

Source: AnandTech

The graphics above, respectively, show Mantle's stated design goals and thread utilization for DX12.

Above, I mentioned that developers would have to ensure games are designed to fully utilize the 8 smaller cores in consoles. Note that both Mantle and DirectX 12 stress the importance of reducing CPU overhead.

Regardless of which API "wins", both of these APIs will try and shift the performance bottleneck to the GPU and make the CPU a less important part of the equation.

To tie this back into the PC space, realize that the Jaguar CPU architecture used in consoles is also used in AMD's lower-powered chips designed for PCs.

Note that the console chips are orders of magnitude more powerful than their cousins destined for laptops, so I am not suggesting that AMD's Jaguar chips are going to be able to play the latest titles at max resolution and settings fluidly.

What I am saying is that the gaming industry is focusing on software innovation to extract the most performance from hardware.

Source: Mantle Oxide Demo via YouTube (slide appears at ~40 minute point)

In the video linked to above, one of AMD's partners for Mantle is giving a demo of the company's game running on a heavily underclocked AMD 8 core CPU and discrete GPU. Testing reveals that the game is still GPU limited (meaning the CPU isn't as important), even when the CPU is running at about half speed.

Industry shifts are making single-threaded performance in PCs less important, largely thanks to the need to reduce overhead to extract performance from consoles. Note the Mantle slide above directly references leveraging "optimization work from next-gen game consoles to PC." This work in reducing CPU overhead negates some of Intel's single-threaded performance lead over AMD, making AMD a more attractive choice for consumers from a pricing standpoint, as AMD chips are typically cheaper. This situation represents improved performance from an AMD chip without increasing the price.

Socketed Chips Mean Longer Lasting And Cheaper Solutions for Builders

The release of the socketed Jaguar chips has seemed largely unnoticed, which is unfortunate, because this will likely have much more of an impact to AMD financially than other initiatives like Mantle or HSA during 2014.

To understand why I feel this is such a big deal, let me explain basic system building to show how this translates to savings for both DIY system builders and, in a later section, OEMs.

PC chips can either be attached directly to the motherboard in such a way the chips cannot be easily removed or replaced (soldered to the motherboard, known as "BGA"), or motherboards include "sockets", which the chips slot into, similar to putting a plug in an outlet.

Source: AnandTech

By releasing socketed versions of Jaguar, AMD is able to offer best-in-class graphics performance (comparing the GPU paired with Jaguar to Bay Trail) and very good CPU performance at extremely competitive pricing. Because the "cat cores" are aimed at mobile devices, these chips are much smaller than AMD's "big core" APUs, meaning the company can sell them profitably at much lower price points.

The average retail price of these chips will likely be in the neighborhood of $40 or so, with ~$30 at the low end and ~$60 at the high end (source: CPU World).

By introducing socketed low-cost chips, AMD is expanding the TAM for Jaguar to include the DIY community and budget system builders.

How Much Lower In Terms of Overall Build?

To build a PC, you need the following components:

  • Case
  • Motherboard
  • PSU (Power supply unit)
  • CPU
  • Memory
  • GPU (Graphics card, optional)
  • Hard Drive
  • Disk Drive (optional)
  • OS (operating system, some are free)
  • Peripherals

Because the Jaguar APU has all the necessary I/O functionality built in, fewer supporting chips are required on the motherboard, allowing for cheaper motherboards.

Looking at motherboards, you'll also notice that some actually do not require a separate PSU. The motherboard comes with an AC to DC plug, similar to what we find in laptops.

Source: Asrock

So these options not only lower motherboard costs, but in some instances, eliminate the need for an external power supply. The latter has two benefits, namely cheaper build cost and reduced cable clutter.

The chip supports up to 1600 MHz RAM, but only in single channel, so RAM costs will be minimal. Finally, the chip supports a PCIe 2.0 slot, so the consumer could add in a cheap discrete GPU if they desired to punch up gaming capability.

This chip brings decent performance to very low price points, which could very easily appeal to the cost-conscious consumers of the world.

Socketed Jaguar Chips In the Desktop Market Represent Incremental Revenue With Little Overhead

Read here (I, II) for a more in-depth analysis of AMD's re-usable IP.

As AMD is focusing on maintaining a tight control over OPEX, the company has designed most of the IP to be reusable and portable. For example, AMD's GCN graphics architecture is featured in both the Kaveri and Jaguar APUs, both game consoles, and in the company's GPU line.

AMD's Jaguar architecture is currently used in consoles and portable PCs. Minimal R&D should be needed to introduce socketed versions of the chip for the desktop market, so this represents an incremental opportunity with minimal overhead.

Beema and Mullins

While AMD is rolling out socketed versions of the Jaguar chip for desktop applications, the company is also slated to soon release the Beema and Mullins APUs.

Although good all around chips, AMD's Jaguar iterations of the cat core family were missing a turbo core feature, which I believe, will be implemented in the upcoming Puma core (source: Bjorn3d).

Source: AnandTech

The above table based on AMD's lab results for the new chips demonstrates an ~25% performance increase for both the CPU and GPU, while simultaneously reducing power. Although no independent reviews have been completed, the above is very plausible if AMD implemented an aggressive turbo and power management profiles.

Whereas the socketed Jaguar chips are aimed at the desktop space, these new chips will be exclusive to mobile computing for the time being.

Jaguar chips were already very well-designed. If my speculation is correct and most of the performance increases and power reduction come from an aggressive turbo and power management, the R&D cost here should be minimal. These chips are 28nm and feature GCN architecture, so AMD should be introducing a nice bump in performance at reduced power consumption with minimal R&D spending.

PCs versus Other Revenue Streams

Prior to the release of consoles, AMD's CS (computing solutions, where CPU and APU revenue are tracked) revenues were between 2 and 3 times that of AMD's GVS (graphics and visual solutions revenues).

Nvidia's GPU market share is around 2x that of AMD's (source: JPR). Combining revenues from both companies yields a total revenue of ~$1B per quarter from consumer-grade GPUs.

Source: Intel Investor Relations

Intel currently enjoys a near-monopoly in the data center, but still generates nearly 3x as much revenues in the PC space when compared against DCG.

AMD has stated the company is trying to diversify revenue away from the PC space, but the PC space remains very important to the company. This is easily seen by looking at the magnitudes of revenue streams to AMD.

Console wins are a big win for the company, but the semi-custom model relies on AMD's reusable IP. This is similar to Intel's relationship in the PC group and DCG. IP designed for the PC group is then reused for the data center chips, allowing Intel's DCG to post a higher operating profit.

Put another way, AMD's semi-custom model only works if operations from the other divisions funds the R&D for the associated IP. Even as AMD diversifies away from the PC space, it will remain an important segment for the company. Also, of all the markets AMD operates in, it is the largest overall in terms of TAM dollars.

AMD's Cat Cores Fit Perfectly Into The Trajectory of the PC Space

The Guardian has an interesting write-up concerning PC ASPs and profits that OEMs see on PC sales.

With profits and ASPs declining, lower-cost x86 chips offer an avenue for OEMs to build PCs more profitably. Looking at any BOM for a smart device will reveal that the processor is one of the cheaper components. Contrast this with PCs, where the price of a processor can easily exceed $100.

AMD's cat core family, or Intel's new line of mobile processors address this divergent pricing. However, as Intel is the dominant player in the PC space, if Intel makes the mobile processors too powerful or prices them too aggressively, the company runs the risk of cannibalizing Core CPU sales.

This means AMD has the chance to capitalize with a lower-priced processor, without as much risk of eroding profitability by cannibalizing the sales of bigger cores.

You may think I'm being petty when discussing a few dollars in savings from eliminating a couple of chips off the motherboard, or even eliminating the need for a separate PSU completely. The Guardian pegs profit margins on PC sales at about 2.5% and roughly $15 profit per unit sold.

Tech writer Joel Hruska published an article regarding the state of the PC market from 2012, but it has a couple relevant points I would like to touch on.

One of the ways OEMs have made money is through what we lovingly refer to as bloatware. OEMs are willing to install software that make most tech-savvy consumers cringe in the name of profits. Saving a few dollars from eliminating a separate PSU or cheaper motherboard could translate into higher profits for OEMS, as these are changes that are transparent to the consumer. Meaning OEMs could save this money, but not necessarily reduce the price by a commensurate amount.

AMD's OEM partners have ample reason to offer these cheaper platforms.


Although initiatives like Mantle or HSA are fun to debate, the revenue opportunity from these initiatives is likely much smaller than the release of socketed Jaguar chips and the Beema/Mullins launches. HSA and Mantle will also require time to play out, and only have second-order effects on revenue if these initiatives translate into increased market share.

AMD's ARM server push is more likely a story for the end of this year at the earliest, with 2015 or 2016 being the time frame where we will see how it actually plays out.

During 2014, in the PC space, we have the release of mobile Kaveri, socketed Jaguar chips, and the Beema/Mullins launch. Because of the TAM of the PC space, this is the largest opportunity for AMD near-term.

To draw parallels to the initial launch of Jaguar chips, AMD stated during the Q1 call last year the company had started volume shipments of the Kabini platform. This was largely before we knew of any design wins for the chip, save maybe one or two.

Fast-forwarding to Q2, AMD managed a significant increase in CS revenues, despite an overall declining market, which could be contributed to what CSO Mr. John Byrne called AMD's "largest commercial tender" for 2013 of 1.5 million units. HP (NYSE:HPQ) sold the units in India.

Contributor Russ Fischer recently published an article based on information from HP demonstrating how the company managed growth during Q1, and attributed the growth specifically to the commercial sector.

To quote Mr. Byrne:

If you look at future markets, it's an industry standard benchmark using a PCMark, 3DMark, BaseMark with open CL, AMD does very, very well. So, that win has opened the eyes up to our biggest customers. If I tell you the commercial PC market, Dell, HP and Lenovo own 77% of that market. And our platforms in 2014 are 3x what they were in 2013.

The cat core family is ideal for OEMs looking to offer a competent solution at an affordable price in either the consumer or commercial space. Socketed Jaguar chips and Puma chips for mobile computing give AMD the opportunity to capitalize in both the desktop and notebook space.

The issue here is timing and marketing. Socketed Jaguar chips are showing up at the beginning of April, so depending on sale through into the channel, we may not see much of an impact on financials. Also, expect to see various OEMs, not AMD, announcing design wins based on new platforms. We geeks also focus on the benchmark aspects, forgetting about marketing and brand recognition.

Based on leaked information (which could always be wrong), I believe we may start seeing some new designs appear this month.

Source: TabTec

Interestingly, if the above turns out to be true (again, based on leaked information, which may be incorrect), it marks a departure from the norm for Dell. Looking through Dell's website, you will be hard pressed to find anything AMD-powered, unless you dig hard enough.

In summary, initiatives like Mantle are important, but it is important to keep things in perspective. There is a reason bears focus on the PC industry; it's still a very large part of AMD's operations. Where I diverge with the bear mentality is when all emphasis is placed in the PC sector, ignoring any other positives. Although I may disagree with it, it is still important to understand the focus on PCs.

Something like Mantle represents a much smaller opportunity for AMD, when compared against the product launches mentioned in this article.

There have been no official announced design wins for Beema, Mullins, socketed Jaguar chips, or mobile Kaveri this quarter that could give a sense of potential impact to Q1 financials. Also, AMD has sold off after each earnings report, so buying in now carries more risk.

Based on the timing of product releases, I am not expecting a beat in PCs during Q1. What I am hoping for is verbal clues during this upcoming ER to give us a sense of AMD's efforts in the PC space, and if Intel is having any success in coming after the low end of the PC space, or if AMD's new design is helping the company gain traction.

Disclosure: I am long AMD, INTC. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

Additional disclosure: I own both shares and options in AMD and actively trade my position. I may add or liquidate part or all of my position at anytime, or initiate a hedge via puts prior to the upcoming earnings call.