AMD's CPU/Semi-Custom Technology Roadmap And How It Relates To Investors

| About: Advanced Micro (AMD)

Summary

AMD's Zen will not contribute much to 2016 earnings, but should be significant as 2017 progresses.

Q3 of 2016 should be a profitable quarter for AMD, mainly due to console seasonality, and the new GPU architecture being sold in volumes.

Q3 2016 should be profitable for AMD, but Q3 2017 is more likely to be the breakout quarter for the company.

In my previous article, I went over the current situation for AMD (NASDAQ:AMD), and the GPU roadmap. To summarize, AMD is not currently selling technology that will allow it to stop losing money, but the new graphics processors released in the middle of the year will almost certainly gain market share for the company.

I would not expect the GPU release alone to be enough to return AMD to profitability, but I do expect Q3 of 2016 to be profitable for the company, while showing revenue growth. There are three factors that will contribute to profitability.

It's difficult to know just how much the GPU release will contribute to Q3 earnings. Some of it depends on when it is released, and some of it depends on how many products with it are released. Since 14nm FinFETs are not going to be very cheap per transistor in this timeframe, especially considering yields, it would not be surprising to see a staged release of products, with mobile likely being the first products the company focuses on. The benefit there is more, since the performance per watt will increase dramatically, and getting laptops to console levels of performance could be an attractive selling point. Also, AMD's demonstrations did not focus on high-end chips, but mid-range chips where the company could show the very low power use compared to the Nvidia (NASDAQ:NVDA) solution. Of course, desktop cards based on similar GPUs are likely as well.

Following the mobile products, I'd expect some higher end GPUs to be released. This is mainly because AMD has a confusing lineup, with the Fury line being limited to 4 GB of memory. This is not a terrible limitation, but it is a limitation nonetheless, and one which the slightly lower-end R9 390 cards do not have. If AMD considered 4 GB enough, then the release of 8 GB 390x cards would never have happened. Clearly, this is not ideal market segmentation, and the company does need to clarify it.

Q3 has two other important aspects. Console chips sales are highest in Q3, in preparation for the Q4 holiday season, and should be slightly higher than in 2015, as predicted by the company. It is also important to note that AMD's next semi-custom win starts contributing to revenue in Q3, and this could be significant since it also prefaces the holiday season, particularly if it is the next Nintendo (OTCPK:NTDOY) console. Put all of these together, and it seems very likely the company will show significant year-to-year revenue growth, particularly since there was still some minor channel inventory clearing in 2015. It also should be a profitable quarter for the company.

Regarding the semi-custom win that will contribute to Q3 revenue, AMD has been pretty quiet about the specifics. It seems very likely it is the Nintendo console, for a few reasons. It does make sense for Nintendo, since being closer to the other consoles would make it easier to port to it. Since Nintendo creates a large proportion of the games for its platform, it would not have to worry so much about its most attractive titles being moved to the PS4 or Xbox One. Also, considering the timing, it makes perfect sense, as it would it feed into the holiday sales cycle. It is also slightly after AMD's newest GPU architecture will be out. This asks quite a few questions though.

If we assume it is Nintendo, we have to piece together what this processor would be. Nothing quite works perfectly. Would Nintendo want a 28nm processor based on AMD's older GCN architecture? If so, would it also be using Jaguar/Puma or Carrizo? That really would not be ideal, since the cost reductions on 28nm would not be very good at this point; the process being quite mature. But, it is safer and probably cheaper short term than moving to 14nm, which will almost certainly have lower yields. However, 14nm would offer better performance/watt, and be much more likely to offer substantial cost savings in the future.

If it is 14nm, which processor would it be? AMD has shown no intention of moving Jaguar/Puma or the Bulldozer line to 14nm, so that would leave Zen, or an ARM derivative. Neither seems that likely, as ARM would not enable the cross-platform ease x86 would, and so far AMD only has produced A57-based silicon, which is not a particularly successful architecture. Zen is very unlikely. It would pre-date the general release of the processor, and would involve tying a GPU to it, which will not be in the PC market until 2017.

Although not ideal, given Nintendo's generally conservative hardware strategy (with respect to CPU/GPUs), it seems most likely this will be a 28nm part, based on a very similar architecture to the PS4 and Xbox One. This assumes this is the semi-custom win AMD has been talking about, which is not certain.

As we go from Q3 to Q4, we will see the expansion of the GPU lineup, and the introduction of the Zen processor. The general feeling I get from AMD is that Zen will not ship in anything even approaching significant numbers in Q4, and the company is very positive about the performance. It seems very likely the only processor from Zen in 2016 will be in the FX line, which will not have a GPU attached to it, and will sell into the higher-end of the desktop market. This will be followed by the server version in 2017, and presumably APUs and mobile devices after that. The order might indicate something, so we should consider some of the implications.

As mentioned earlier, the greatest advantage FinFETs have is the power use is so much less. It will also enable higher clock speeds, but performance per watt improvement is greater in magnitude. Couple that with the fact the mobile market is stronger, and the high-end desktop less significant in terms of numbers, and it does raise some questions. Then consider the slow uptake servers have, and that these are on the roadmap before APUs, or mobile, and the questions get even more insistent. Is this due to poor power characteristics?

Almost certainly, it was just simpler for AMD to make a standalone processor than an APU. This would explain clearly why mobile is coming after high-end desktop and server parts. Nothing else really ties all the parts together as well. Naturally, APUs will come later in 2017 as AMD moves Zen further into the product stack, but with a new GPU architecture coming out in mid-2016, it was probably too much for the company to manage all these moving parts and come out with Zen with the newest GPU architecture in Q4 of 2016. A mobile part without a GPU would be a non-starter, so I do not believe the later release in that segment is related to power draw.

Q4 should be sequentially lower than Q3 due to console chips seasonality, with Zen having only a minor impact. However, with the new GPUs gaining traction, particularly in mobile sales, and the new revenue stream from the newest semi-custom win, it should be higher year over year. Full-year revenue growth is also very likely, and almost certain.

Q1 of 2017 will be similar, being down sequentially, but up year to year. Zen will probably still be relegated only to the FX line, and although it will likely increase sales there, I would not expect them to be enough to offset the seasonality of the PC market, GPU market, and console market. If I had to guess, I would think late Q2 is when we will see server chips actually shipping for revenue, and late Q2 to early Q3 before Zen expands into the mobile and mid-range market, as an APU. Consequently, I would expect Q2 to be okay, but not strong, and AMD's third quarter of 2017 to be a breakout quarter, with Q4 going up sequentially due to the PC market overpowering the lower seasonal sales in the console market.

There are a few other tidbits we should look at that are of interest. Before Zen comes out in Q4, we will almost certainly see Carrizo on the desktop, as it will share the AM4 socket with Zen. Since AMD's APUs are notoriously bottle-necked by memory bandwidth, moving this platform to DDR4 could show a meaningful improvement in speed. This will also manifest itself in the move to the FP4 platform in mobile, which will also use DDR4.

The big question I have glossed over is Zen, and how it performs relative to Intel's processors. By the time Zen is out, Intel (NASDAQ:INTC) will have released Kaby Lake, which will almost certainly be Skylake with some non-architectural enhancements. This is not to say it will not be significant in some ways; most enthusiasts know that Broadwell is often a better gaming CPU than Skylake, despite the underlying architecture being older and running at a much lower clock speed. I would guess it will be more than Devil's Canyon, in terms of change, and marry the Skylake CPU with the L4 cache used in Broadwell, and likely a beefier GPU.

AMD has recently upped its assessment of Zen, saying it will offer MORE than 40% per core performance compared to Carizzo, whereas before it was 40%. Considering the move to FinFETs as well, clearly this processor will be much more competitive with anything Intel has, and have a significantly better GPU when the APUs come out. However, it's very difficult to see AMD duplicating Intel's L4 cache, since Intel has so much fab capacity that it needs to use, while AMD has to pay a foundry to make its CPUs. The impact of the L4 cache will make it almost impossible for AMD to offer a better gaming CPU than Intel, on most titles, although we may see outliers favor AMD's architecture. But it will significantly close the gap, and if priced properly could offer a superior solution to a large part of the PC market, considering cost. Keep in mind that L4 cache Intel uses on Broadwell is not exactly cheap, so will not proliferate the entire line, and stay mostly in the higher end desktop and high-end mobile parts.

What this means in servers is less clear, as large external (to the CPU die) caches are by no means an impossibility for AMD, given the price server parts can command. AMD will also be introducing Coherent Interconnect Fabric, which should find some use in servers, thus expectations for this need to be tempered, as this technology is by no means new. IBM's (NYSE:IBM) CAPI and Nvidia's NVLINK are both similar technologies, and both will be used in OpenPOWER. Coherent Interconnect Fabric has not been fully described, as the company has chosen not to reveal much about it, other than its very high bandwidth and low latency. By itself, this would fail compared to the aforementioned technologies, so probably also allows the GPU to read and update the processors cache, and share memory access. Anything less would not be competitive. In a nutshell, this technology will basically try to make an APU out of a pure CPU and discrete card, without the overhead now associated with using an external GPU. There's nothing to prevent an APU being used in this manner with a discrete card as well, so it will not necessarily be limited to pure CPUs.

As mentioned, this type of technology is already around, so is not going to give a competitive advantage by itself. However, at least it will eliminate a potential disadvantage, assuming what I have above. Given AMD's GPU architecture, there still may be advantages in this configuration, it just will not likely be the interconnect that is responsible, but the GPU architecture. We also need to keep in mind that the vast majority of algorithms will not run better on a GPU, so the scope of this has to be limited. Even so, servers are very profitable, and even a niche can be very helpful in terms of revenue, as well as margin accretive. Just try to keep it in perspective though; it is good the company is making it, but it will not impact the server market in an enormous way.

I do not believe ARM-based processors will contribute significantly to AMD's revenue in this timeframe, although certainly if K12 is a superior product, it could have an impact if the ARM infrastructure continues to develop. It may also be an attractive architecture for future semi-custom wins, but this is far from certain, and probably a bit further out than 2016/2017.

That is how I see things, from where we are now. Keeping in mind there is a lot of speculation in this article, it is important to remember not everything here will be correct. If you are not comfortable making a judgment now, I hope the reader will just keep these points in mind and watch how things unfold; what we cannot see now, we will in time. I will try to keep up with new articles when something interesting or unexpected happens as well. But, from where we are now, I do believe the company is poised for improvement, and the technology roadmap offers a lot of hope and potential, which more than justifies the risk at the current price.

Disclosure: I am/we are 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.

Editor's Note: This article discusses one or more securities that do not trade on a major U.S. exchange. Please be aware of the risks associated with these stocks.