AMD's Future In Scalable Solutions

| About: Advanced Micro (AMD)

To be successful, you must set achievable short-term goals that act as successive steps to a final, larger goal. I believe Advanced Micro Devices (NYSE:AMD) has a pretty lofty long-term goal in mind and the right shorter steps to get there.

What is the HSA Foundation?

The HSA is the collaborative efforts of those above to standardize programming an SoC. For the non-technical audience, an SoC is like a CPU (Central Processing Unit) but more complete. Think of it as having the CPU, graphics processor, and in/out functions (display, connections for storage, things like that) on one convenient chip.

The end goals of the HSA foundation is to make computer programs indifferent to the hardware they are running on, as well as make the programs run as efficiently as possible. This efficiency applies not only to how quickly programs will run, but how much energy is consumed. And with tablet and smartphone sales outpacing traditional PC sales, there are now more SoC style CPUs shipping each year than computer CPUs to consumers.

Comparing ARM and AMD

ARM Holdings (NASDAQ:ARMH) does not make the most powerful chips. That crown goes to Intel (NASDAQ:INTC) currently. Yet ARM based SoCs are in 95% of smart phones, and 40% of mobile computers which include both notebooks and tablets. The 40% would most likely be much higher if ARM broke out tablets as a separate category.

I believe this is because ARM gives vendors the choice of flexibility. ARM's offerings are so customizable that Nvidia, for example, is able to use ARM based central processing units, add their own graphics, and create a chip for tablets and smart phones. Nvidia's Tegra 3 was one of the reasons Nvidia (NASDAQ:NVDA) had a nice earnings surprise last year despite slumping PC sales to the tune of $1.2B in revenue in 2012, Q3. Note that Nvidia's SoC is struggling as of late, however.

Now consider AMD's contracts with both Sony (NYSE:SNE) and Microsoft (NASDAQ:MSFT). AMD offered both companies a complete solution for graphics and processing power on a single chip. But, in my mind, the most important part of the equation was allowing the companies to customize the chip for their specific desires.

AMD's Product Scalability

For my non-technical audience, the only part to pay attention to is the fact that this architecture allows AMD to more easily scale the amount of graphics processing power to fit various scenarios.

As is evidenced by the architecture in the Xbox and PS4, AMD has scalability in their low power Jaguar cores as well. For the CPU portion of the Xbox/PS4 hardware, AMD connected two separate quad-core Jaguar chips together for a total of 8 cores. AMD was able to offer Microsoft and Sony different GPU hardware to allow the companies' overall flexibility of processing and graphics power.

This puts AMD in a similar situation to ARM Holdings in that companies like Microsoft and Sony were able to pick and choose the amount of CPU power and GPU power to fulfill their respective needs. Even if Nvidia skipped out on the next gen consoles because of 'poor margins', I feel what they could have delivered would've been inadequate due to no fault of their own. Traditional PCs are based on x86 architecture, which means their ARM powered Tegra line of chips would not have fit the bill to allow easier portability from console to PC. The use of x86 architecture more easily allows developers to migrate games created on the consoles to the PC and vice versa. And AMD being an x86 chip vendor with customizable IP provides them a unique advantage. The beauty of AMD's Jaguar and GCN cores is their wide range of application. AMD is able to operate the cores at various frequencies and quantities to adjust for desired CPU/GPU power.

Anand has an interesting article on his website regarding Mr. Jim Keller. Mr. Keller rejoined AMD midway through last year. Raja Koduri also rejoined AMD after a stint at Apple (NASDAQ:AAPL). We could possibly see their efforts appear by the launch of "Beema", the successor to the Jaguar cores or in 'Kaveri' this fall.

Although AMD would not comment on Beema, they did respond to a question I had regarding Mr. Keller. They responded by stating, "Keller is leading the development of our processor cores, including next-generation cores. His team is responsible for developing the CPU cores across our various SoCs." For those unfamiliar with Mr. Keller, he was the lead architect on the K8, co-authored the Hyper Transport protocol, and helped develop Apple's mobile SoCs.

AMD's Small Steps

AMD is still at the mercy of a waning PC industry unfortunately. But one of their main focuses currently is to play to their strengths over Intel - graphics and customization. AMD is not trying to out-innovate Intel, but in the words of AMD's Mr. Feldman, "innovate right around them." Although Mr. Feldman is involved in networking so his quote does not directly apply to gaming, it still speaks to AMD's current vision (and the article I linked to is also worth a good read), which is "...committed to an ambidextrous approach and providing choice to our customers..." (This being a quote I received in response to questions I have asked of AMD)

AMD is making this approach possible by taking re-usable IP blocks and working with vendors to scale performance to fit specific needs. With the current paradigm shift of more energy efficient computing, a main driving force for AMD is performance per watt.

The above slide is no small feat. And it is verifiable in any independent review you read - no matter the tone of the article, almost every review I have read speaks favorably of Jaguar's power consumption (I, II, III, IV).

AMD's GCN 2.0 aims to further increase efficiency while minimizing power consumption. Bear in mind one of the strengths of Apple's A6 Swift was performance/watt. I believe this will be a focal point of Mr. Keller for the successor to Jaguar, and is why I mentioned him by name earlier.

AMD has added another keynote to AFDS13 to address their role in gaming specifically. Game console sales totaled $4B in 2012, and this was a down year for consoles at the end of their life cycles (excluding the Wii U). Numerous reports of the next generation consoles being out of pre-order availability point to strong demand, and AMD is making money on each of these pre-orders. Gaming will be a large source of revenue for AMD during their turnaround.

AMD generated $326M in graphics card sales last year. As an AMD long I am looking to see this number grow in light of console wins and AMD's Never Settle/Gaming Evolved programs. Although there may be signs of weakness this quarter as gamers wait for a potential release of AMD's next generation Radeon cards.

*POINT* AMD is focusing on improving performance/watt in their reusable IP blocks, while at the same time generating revenue from gamers.

AMD's End Goal: Heterogeneous Computing

As the main computing focus shifts away from traditional desktop PCs to mobile devices (I am including laptops and below in this category), AMD and other members of the HSA foundation are attempting to tap into the GPU to try and squeeze more performance per watt out of each chip, along with increasing portability of programs between different architectures (ARM RISC vs x86).

Image taken from

(Yes, I know these are Nvidia/Intel numbers - but they illustrate my point well)

This graph represents the progression of computational power in CPUs vs GPUs over time in non-integer workloads. HSA aims to take advantage of this extra computing power by making it easier for mainstream programmers to access the GPU. HSA also aims to increase efficiency by sharing system memory between the CPU and GPU. Gaming consoles and AMD's Kaveri will be the first time we see full implementation of this shared memory.

Intel's process advantage allows for more transistors to be packed in a smaller area, along with better power characteristics.

AMD's 7770 GPU built on TSMC's 28nm node packs in 10 GCN units in an area of ~120 mm^2. To get a very rough approximation of die area per GCN, simply divide, to come up with roughly 12 mm^2 per GCN core, or roughly 24mm^2 for the GPU in Jaguar. This should be conservative because things like cache memory are included in the die size of the 7770. Using some numbers from Anand, he calculates the size of 40 Execution Units (Intel's graphics cores) as 174mm^2. Scale this to 16 EU's gives us a rough die area of 16 EUs/40 EUs * 174mm^2 = ~70mm^2 die area for 16 EUs, if mine and Anand's envelope calculations hold up. 16 EUs is the number of graphics cores in HD4000, and the a4-5000 gives slightly less performance graphics wise as HD4000. Real world gaming benchmarks will favor Ivy Bridge however due to Intel's stronger CPU performance. To illustrate my point, HD4000 scores 20% higher on an Intel desktop CPU when compared to a mobile i3 IVB. To further demonstrate Intel's limitation of scaling graphics performance, compare HD4000 to HD5000.

Graphic and quote taken from

"The data ranges from a meager 2.3% advantage over Ivy Bridge ULV to as much as 40.8%. On average, Intel's HD 5000 offered a 15.3% performance advantage over Intel's HD 4000 graphics. Whether or not that's impressive really depends on your perspective. Given the sheer increase in transistor count, a 15% gain on average seems a bit underwhelming."

To keep this in context, HD5000 is giving higher performance in a lower total TDP. But the cost is 2.5x more die space in going from HD4000's 16 EUs to HD5000's 40 EU's for graphics. However, this scaling should improve with better power characteristics at 14nm for Broadwell.

Image taken from

Given a Jaguar core CPU size of 3 mm^2, you can see the estimate of 24 mm^2 does not appear far off. To put this in context, AMD uses around 35-40% of the die area to offer slightly less graphics power than Intel, based on some very rough math and screen shots.


The end-game in this scenario is to utilize heterogeneous computing to play to their strengths, neutralizing some of Intel's process lead by allowing the GPU to help accelerate computations where I have shown it appears AMD has a distinct efficiency advantage over Intel when looking at transistor budget.

As an AMD long, the catalysts I am looking for are regaining market share in the server space and discrete GPUs, and more embedded/semi-custom design wins to put AMD on track to divorce some of their revenue streams from traditional PC sales.

However, AMD's core business is still traditional PC sales, meaning a 7.9% decline this year could be painful. Luckily, Intel reports earnings the day before AMD, to give as best of a glimpse as AMD longs can see in traditional PC sales.

As long as the next earnings call is not abysmal from lost revenue of declining PC sales, I am much more interested in the tone and guidance that comes out on 18 July.

AMD will not be able to reproduce the efficiency magic between GCN and GCN 2.0 as they did between Bobcat and Jaguar (given the node shrink that occurred), but we should see a nice performance boost nonetheless, and if the rumors of the next graphics card launches hold, AMD will have new products potentially generating sales before the Q3 earnings call. And for a Q4 console launch, we should see the gaming console revenue by Q3 as well, if not as early as next week.

Earlier after the console wins were announced, EA's CTO stated that consoles were more powerful than PCs. Taken out of context, I will be the first to say this is completely false. But after a little more digging, I believe what EA was referring to was hUMA, and the gains seen from the new memory standard.

Quote taken from

" EA Sports boss Andrew Wilson says that one reason none of its next-gen sports games are coming to PC is because Microsoft and Sony's new game consoles are actually more powerful than many PCs in a very specific, subtle way: "How the CPU, GPU, and RAM work together in concert," Wilson told Polygon."

If this quote is taken for truth, any gamer wanting to play Madden 2014, for example, will be playing exclusively on AMD hardware by virtue of EA sports games being solely released for consoles.

AMD pushed back AFDS to November. Given the fact that the Fusion Developer Summit is typically a summer event, I believe this was done to coincide with the Kaveri launch. In November, we should get a glimpse as to how AMD's hUMA/HSA initiatives are faring, and how successful their turnaround is.

Disclosure: I am long AMD. 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 may initiate a long position in MSFT or INTC at any time.

About this article:

Author payment: $35 + $0.01/page view. Authors of PRO articles receive a minimum guaranteed payment of $150-500. Become a contributor »
Tagged: , , , Semiconductor - Broad Line,
Problem with this article? Please tell us. Disagree with this article? .