Advanced Micro Devices: Immense Challenges Ahead
Summary
- AMD hasn't been this healthy in many years.
- However, many large challenges are now emerging.
- These challenges will grow ever more intense over the next 1.5 years and beyond.
- Looking for a helping hand in the market? Members of Idea Generator get exclusive ideas and guidance to navigate any climate. Get started today »
Advanced Micro Devices (NASDAQ:AMD) hasn’t been in such a good position as today for a long time. Since 2003, with the launch of its Opteron server processors, it had mostly been downhill for AMD. Yet, today, AMD can boast a competitive CPU lineup across consumer (Ryzen) and server (EPYC) markets. At the same time, it can be said that AMD recovered some ground in the GPU business versus Nvidia (NVDA) as well.
Yet, at the same time, an incredible number of challenges have arrayed themselves against AMD, and at times also Intel (INTC) and Nvidia. This article will cover all of the different challenges coming up – most of which hit within the next 1.5 years.
Intel Counterattacks
Historically, Intel has been AMD’s main problem. Even when Intel dropped the ball, it didn’t take many years to catch it again and run away with it. Typically, Intel’s mistake had been at the CPU architectural level, while Intel always kept a fabrication process lead. However, recently Intel saw itself both with an architecture and process problem, though related.
Intel managed to see itself surpassed in terms of production process by Taiwan Semiconductor Manufacturing Company (TSM), to which AMD now outsources its production. This has meant that AMD now manages to ship 7nm CPUs even as Intel still struggles to make its 10nm process work. Intel went from being 1-1.5 production nodes ahead to being at least 0.5 production nodes behind. This in turn had an impact on:
- Cost. A better production process allows for a lower cost per CPU die, since for the same die design, each die takes less area on a silicon wafer.
- CPU speed. A better production process allows for higher performance for the same power consumption.
- CPU efficiency. A better production process allows for lower power consumption for the same performance.
At the same time, Intel has only taken baby steps at the architectural level for years, allowing itself to be caught by AMD’s much quicker progress with its Zen 1 and Zen 2 CPU cores. Another architectural difference, the AMD usage of chiplets versus a monolithic die, also allowed AMD to offer higher core counts. AMD thus mostly closed the single-threaded performance deficit it had to Intel, while also building a multi-threaded advantage. And it did this at a low cost to itself (due to the chiplet strategy).
However, all of this might arguably be set to change. The main reason is that Intel has started reacting on an architectural level. Although there was a level of reaction with the Sunny Cove core used in some Ice Lake products, and this is likely to continue with the Willow Cove core set to hit Intel products in 2020, the real challenge might come in 2021 (at the earliest).
That’s when the Golden Cove core will hit Intel products. Why is this so relevant? Because that might be the first core where Jim Keller’s influence should be substantial.
Jim Keller is the individual who helped put AMD back on track in 2003 and again with the Zen core. He was also at Apple (AAPL), where he is credited with leading the A4 and A5 designs, when Apple brought out its revolutionary A6 CPU. This was the first CPU that propelled Apple into the ARM performance lead since 2012 (the CPU launched in September 2012, Keller left in August 2012). It should be noticed that this is a pattern with Keller, and the same thing happened at AMD (twice), at Apple and at Tesla (TSLA). Keller left right before a significant breakthrough happened (on which he most likely was a large influence). It’s also not in dispute that Keller led chip design at PA Semi, a company Apple acquired right after it hired Keller, and whose influence on the A6 breakthrough is undisputed.
Hence, the first Intel CPU under Keller’s influence has to be respected. This CPU will likely arrive in late 2021 and likely initially use a fabrication process (Intel 10nm) near TSMC’s 7nm, which will be used to produce AMD’s next generation CPUs using Zen 3 cores. AMD’s own Zen 3 products should arrive in early 2021. Intel also expects to start using its 7nm process technology in 2021, though by then TSMC will start deploying 5nm.
A potential saving grace here. While Intel’s Golden Cove core promises to focus on single-threaded performance/IPC (Instructions Per Cycle) performance, it’s not a certainty that this will be the true new architecture already influenced by Keller. It’s possible that only the following generation bears his mark.
There are other considerations which we could make, like Intel also moving to a multi-die architecture, etc. However, those other considerations should for the moment be less relevant than Intel regaining the architectural crown.
This challenge, where Intel might again take the crown from AMD, is the biggest challenge AMD will face. Historically, it has been very punishing for AMD when it loses its (temporary) leadership to Intel.
ARM-Based Solutions Will Start Stealing Server Workloads
I have already covered this challenge in my article titled “Amazon's Graviton2 CPU: This Time, It's For Real”. The thesis is simple: ARM cores have gotten good enough to be adapted to the server room, and they’ve started being deployed as such. And ARM cores are evolving faster than x86 cores, gaining more efficiency and performance per each successive generation, so this challenge will only grow.
As the server room increasingly adopts such ARM solutions, computing workloads will shift from x86-based solutions to ARM-based solutions. This is going to at some point impact the demand for x86 server chips from Intel and AMD. From the Graviton2 CPU performance, we can expect that this point is close in time.
The Graviton2 is exclusive to Amazon.com (AMZN), but others can and will replicate its success. This is so because everyone can get equivalent IP from ARM.
Apple Will Start Leaving x86 In 2021
I’ve been talking about this for a while, and recently this has gotten a further push in media. Apple is now expected to launch the first ARM-based MacBooks in 2021.
Apple has long had enough performance in its A-Series CPUs to do this. Apple will have even more headroom in its CPUs once it targets laptops with higher power and thermal budgets. Apple will deploy CPUs targeted at higher TDPs (Thermal Design Power), and with more cores than typical for smartphone or tablet designs.
Although this challenge is expected to materialize in 2021, its announcement is likely to already take place in 2020.
This challenge will have at least 2 effects:
- It will reduce the size of the x86 market as Intel is displaced from Apple products (though not all at once). This loss will mean that Intel will have to become more aggressive in the rest of the market.
- It will lead other competitors to emulate Apple (though in this regard Microsoft has already started doing so – but with a solution inferior to Apple’s). This will further reduce the size of the x86 market for both Intel and AMD.
Apple Will Likely Also Start Using Its Own GPUs
Apple doesn’t just possess its own proprietary ARM-based CPU solution. It also has, and is using it in its A-series CPUs, its own GPU solution.
Hence, in its migration towards using its own ARM-based CPU solutions in laptops, it likely won’t limit itself to using its CPU technology. In all likelihood, it will also use some of its own GPU technology. This might displace some AMD parts.
Intel Is Entering The GPU Market
Intel is also entering the GPU market, both for consumers and for server rooms. This is set to start happening in 2020 and 2021.
This event will transform the GPU market, as it will no longer be a duopoly. At the very least, it will weaken economics for all players. At worst, it might create a credible competitor taking away visible market share from AMD and Nvidia, both in the consumer market and in the server room.
As an aside, there are rumors, but not nearly as solid as Intel entering the market (a certainty), that Huawei might also enter the server GPU market.
FPGA/ASIC Solutions Are Set To Take Away Some AI Server Workloads Away From GPUs
Many AI workloads lend themselves to fixed-function hardware (ASICs) or to more flexible hardware (FPGAs). Thus, like with bitcoin mining, it’s likely that many server-room AI workloads will migrate to such solutions and away from GPUs.
This has already started. Google (GOOG) (GOOGL) started fielding its own TPUs (Tensor Processing Units) back in 2015, and is now into its 3rd generation, while also starting to make its TPUs available for commercial workloads.
Source: Google
Other large cloud providers, like Amazon.com or Microsoft (MSFT) are going down the same road with their own custom hardware. And moreover, commercially available hardware is being made available from sources selling into the open market.
Obviously, Nvidia and AMD are also incorporating fixed-function AI acceleration into their own GPU chips. But this is of little comfort, as previously GPUs didn’t have to divide the available workloads with other competitors, and now they do.
In the end, the GPU server market will become smaller (or grow much slower) as a result of this trend.
Conclusion
We’re on the cusp of many structural changes to the x86 market which can potentially impact AMD (and Intel and Nvidia).
On top of those many changes, Intel might finally show its reaction to AMD’s improvement. Intel's efforts now have Jim Keller on their side. This is the man who gave AMD its two chances at beating Intel.
With these challenges dead ahead, AMD's stock looks particularly rich here, given its high valuation even based on the 2021 consensus (32.5x earnings).
Granted, not all is bad. Between now and 2021, AMD is still likely to expand its GPU server market share, its CPU PC/server/laptop market share, and it will also see the launch of a new console generation. But many of the changes highlighted are structural. Simply put, the markets AMD (and Intel, and Nvidia) sells into are going to turn structurally less interesting due to the incoming surge in competition.
Idea Generator is my subscription service. It's based on a unique philosophy (predicting the predictable) and seeks opportunities wherever they might be found, by taking into account both valuation (deeply undervalued situations) and a favorable thesis.
Idea Generator has beaten the S&P 500 by around 29% since inception (in May 2015). There is a no-risk, free, 14-day trial available for those wanting to check out the service.
This article was written by
Portuguese independent trader and analyst. I have worked for both sell side (brokerage) and buy side (fund management) institutions. I've been investing professionally for around 30 years.
I have a Marketplace service here on Seeking Alpha called Idea Generator that's focused on deep value, real-time actionable ideas based on valuation and catalysts. The Idea Generator portfolio has beaten the S&P 500 by more than 74% since inception (2015).
I can be reached at paulo.santosATthinkfn.com.
Analyst’s Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours. 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.
Seeking Alpha's Disclosure: Past performance is no guarantee of future results. No recommendation or advice is being given as to whether any investment is suitable for a particular investor. Any views or opinions expressed above may not reflect those of Seeking Alpha as a whole. Seeking Alpha is not a licensed securities dealer, broker or US investment adviser or investment bank. Our analysts are third party authors that include both professional investors and individual investors who may not be licensed or certified by any institute or regulatory body.
Recommended For You
Comments (164)
www.usnews.com/...
gizmodo.com/...For the GPU side, AMD largely ignored the discreet GPU market for the past 5-7 years as it concentrated on getting its CPU division back competitively with Zen. A company of AMD’s size could only really fight on 1 front really. Trying to do so on 2 fronts would have been catastrophic for AMD I think.A while back, AMD shifted ⅔ of the GPU team to semi-custom and Zen work to ensure a steady stream of revenue through consoles. What was built for customers was adapted for the consumers GPU market just to stay afloat there. A good way of maintaining mindshare in the customers market with Microsoft & Sony partially footing the R&D bills. The experience gained here now shows up in AMD's APU and HSA capabilities also which will soon be demonstrated with 7nm mobile and Frontier HPC.AMD did more in GPU strategically than that though. With Zen’s main plan of attack on Intel was raising CPU core counts, AMD also started the groundwork in GPU with the introduction of Mantle API in 2013, a low-level low-overhead rendering API with multi-threaded parallel CPU rendering support for at least 8 cores. Hrm….coincidence? I think not.Well, it took a while but we now have Vulkan and DirectX 12 in place which are derivatives of Mantle. We’ve already seen how well games can perform on AMD HW if they're are fully optimized for DirectX 12 like Forza. It levels the playing field as it minimizes drivers optimization advantages which NVDA has with Dx11 given its massive SW team. It’ll also benefit AMD’s CPU more with 8 cores being the new minimum set by consoles.The last hurdle here is for game developers to truly adopt to these 2 APIs, not just partially like we’ve been seeing. Well, the arrival of the new consoles in 2020 just might be that final push for it to happen. RDNA2 will fare better in this new landscape with AMD now capable of adding more focus back to its GPU division, thanks to the current success of its CPU counterpart. I now think it's really part of a master plan altogether as everything just sorta dovetail nicely.
It is truly impressive how little David (AMD) is taking on 2 Goliaths in Intel and Nvidia, 1 at a time of course.
I did have my doubts earlier on with my investment in AMD thinking that chiplets would be too far off of monolithic CPUs in single thread performance which could hold AMD back. Glad I was proven wrong.
Wintel failed to create a market in it. ARM did it.By the way, didn't Apple hire Jim Keller and acquire
ARM based CPU startup he was working for ?
And Use it for iphone4+ iPad 1 and 2 .
in this upcoming revolution in big data/high performance computing.
This is going to be crowded field but lots of money to be made.
at best could get to 250M and that too is questionable.
www.tweaktown.com/...
optimized for ARM architecture.
That almost ensure no quick wild fire, IMHO.Much smaller software rewrites for AMD x64 but still they still
have some software issue.
Even if AMD's APU has such large advantage in performance
(in data center), somebody should re-write the software to
take full advantage of them....
particularly in AI server, data center. :)
I'm thinking there could be some synergy. :)
But, just like Apple learnt the best "rules" from Jim Keller, and has not lost any mojo even after many years of JK's (and other "rock stars of chips") departure, so too will AMD. I feel there are 5 important considerations that counter your conclusions:
- AMD's Mark Papermaster has whatever it takes to continue AMD on its innovation path, be it in chip designs, or increasingly important, interconnect technologies;
- Lisa Su is a magnetic, fiercely focused, visionary and trustable leader both within AMD and to a vast eco-system of partners, customers, governments and IDMs both in USA (Google, Microsoft, Oracle, Cisco come to mind,among others) and Taiwan (ASUS, SMCI, Toshiba, MSI, etc). I feel it would be a mistake to "ignore" her in any assessment of AMD's prospects;
- TSMC is another key factor that would be a mistake to "ignore" in assessing AMD, as gleaning from what I heard at their last earnings conference call (and the one before that), making AMD enormously successful is what is going to propel TSMC ahead of everyone else, esp. Intel and Samsung;
- be it competition from Google, Microsoft, Intel, even Amazon, the total, combined cost "per 'equivalent' chip" from AMD is going to be a lot cheaper than anything any of the above can finally get out of their respective fabs. This is because AMD's and TSMC's heavy-duty engineers are in Taiwan, and they are a highly focused (and "culturally and nationally" paranoichally bent), integrated group, with much lower wage scales than any of the other players. It is worth the while to listen to TSMC's last earnings call, esp. towards the end, when an analyst asked them about either doing an M&A in USA or building a fab there. TSMC management declared what I too have always believed, i.e. Taiwan is the best place in the world to make any kind of chips, incl. design, process, test, packaging, etc; and finally,
- their recently announced rDNA architecture is a must watch, as if it is anything like what they have promised, AMD will have a powerful counter to any of the threats you've listed in your report.
In sum, there are many challenges for AMD that you have rightly raised, but it may be too early to declare their likely loss when the next 12 months will reveal many answers. I'm expecially watching whether they do "deliver" at the US Dept of Energy in the highly ambitious El Capitan project.
You think Amazon is running an unmodified ARM core? I don't know the answer, but I suspect not. Something will eventually displace x86 from servers, and I suppose it could even be ARM. But remember that AMD nearly went bankrupt by betting the farm on ARM a few years ago.