AMD Exposes More Weakness At E3

| About: Advanced Micro (AMD)

Summary

At E3, AMD showed more cheap Polaris-based GPUs, but nothing for notebooks.

The low performance and price of AMD's Polaris chips indicate continuing process problems at Global Foundries.

Microsoft's Xbox One announcement also indicates that chips combining Zen and Polaris won't be available until late 2017.

Advanced Micro Devices (NYSE:AMD) CEO Lisa Su took the stage at E3 to highlight more graphics cards based on its new Polaris architecture. These cards, the RX 470 and RX 460, feature quaintly anachronistic single slot form factors, and their main selling feature is their higher performance per watt. In the name of "democratizing" PC gaming and VR, AMD is offering consumers a backward step in performance for a very low price. Meanwhile, the graphics chips that AMD keeps promising, low power notebook Polaris GPUs for which there is probably a reasonable market, were nowhere to be seen.

Click to enlarge

Source: PCGamer

True Vaporware

The hallmark of AMD's marketing strategy since the beginning of the year has been to highlight products that can't be bought by consumers. AMD has been very careful not to over-promise. It usually doesn't say when the products will actually be available, or what their performance will actually be. But it makes a lot of glowing generalizations that leaves fans and investors assuming that its products will outperform the competition.

For all the marketing hype, AMD has only made two very specific claims about its Zen CPUs and Polaris GPUs. AMD has said that Zen will feature 40% more instructions per clock cycle, due to implementing hyperthreading in the CPU cores. For Polaris, the claim has been higher performance per watt, in the range of 2-2.8X depending on the product.

And that's about it. When you sift through all the marketing hype, those are the main claims that withstand scrutiny. The only other specific performance information we have is the CrossFire performance that Koduri showed at Computex for a pair of RX 480 cards.

That limited information does not support the hypothesis that AMD's offerings will outperform its rivals' CPUs or GPUs, which is why we now are hearing the spin about "democratization". AMD knows it doesn't have the premium end in GPUs, so it is embarked in a race to the bottom to sell its products based on low price.

The new RX 470 and RX 460 are cases in point. It's clear that these cards will offer sub-par performance for even the previous generation of graphics cards. The only real improvement they offer is that they'll consume less electricity. Gamers looking to play games in 4K or VR will not be able to use these cards.

The important market for Polaris has seemed to be for sometime in notebook GPUs, where the power efficiency could be used to best advantage. AMD's press release that accompanied the E3 presentation contains a lengthy reference to the value of Polaris in notebooks:

Console-class GPU performance for thin and light notebooks - Gaming notebooks have traditionally been large and cumbersome or under-powered for today's gaming needs. The Radeon™ RX Series addresses this with flagship technology that effectively gives mobile users GPU performance that rivals that of consoles with exceptionally low power and low-z height to drive thin, light and high-performance gaming notebooks, and 1080p 60Hz gaming experiences for both eSports and AAA titles.

This is classic AMD marketing. It extols the virtues of Polaris in notebooks. AMD makes it seem as if the product is right around the corner. But there are no specifics. There's no model number. There's no launch date. There's not even a rough time frame for availability. This is true vaporware.

Bottom Feeding

Neither of the cards that Lisa Su held up at E3 would be suitable for use in a modern thin and light notebook or 2-in-1. The RX 460 will have a starting price of about $100 and performance of the R7 360, according to expectations of AnandTech. But even the 460 will not have a low enough power consumption (expected to be about 75 W) to go into a notebook.

It's important for investors to understand the technical reasons why AMD has been forced into bottom feeding with what should have been its premium GPUs. This is really driven by yield and maturity issues in Global Foundries' 14 nm FinFET process that AMD is using for Polaris, and expected to use for Zen as well.

When a foundry first starts a new process node, yields, in terms of percentage of chips on a wafer that are usable, are not very good. As the foundry gets more experienced with the new process, it figures out what it's doing wrong and yields improve.

One way to get around the yield problem in the near term is to make sure that the individual chips are small. AMD has had to keep Polaris small in order to help GloFo get decent yields. So the RX 480 actually has fewer stream processors (the basic unit of GPU processing for AMD) than the previous generation R9 390 (2304 vs. 2560 according to AnandTech). Other performance specs are expected to be in line with the R9 390.

The chip inside the low-end RX 460 is Polaris 11, and is assumed to be even smaller than Polaris 10 inside the RX 480. Smaller means better yield and more chips per wafer, so cheaper, but also lower performing.

AMD's next-generation Vega GPU architecture is expected to be more comparable with Nvidia's (NASDAQ:NVDA) Pascal in performance, but this is off in the future, probably some time in 2017. Once again, the fact that Vega has been put off to 2017 is indicative of process limitations. Vega, which will need to be much larger than Polaris, can't be made profitably until Global Foundries improves its process.

Process maturity also impacts the performance of parts that are considered acceptable. Polaris 11 was expected to go into notebooks, but the high power consumption probably indicates that it isn't performing as efficiently as expected. This is why we aren't seeing any notebook parts, nor are we being given a time frame for availability. Probably neither AMD nor Global Foundries have any idea when they'll be able to produce notebook parts.

Project Scorpio

E3 offered more evidence of production limitations for Global Foundries and AMD in the form of Microsoft's (NASDAQ:MSFT) announcement of Xbox One S and Project Scorpio. Microsoft announced the One S update, which features a 40% reduction in volume compared to the Xbox One. It also has a 4K Blu-ray disk player and the ability to play 4K video, but not 4K games. Technical details on the S remain sketchy, but its processor inside is likely to be a minor update of the current semi-custom processor from AMD. This means it's still a 28 nm part using previous gen CPU and GPU technology. No Zen or Polaris.

Part of the reason for inferring this is that Microsoft also previewed a dramatically improved "Project Scorpio" version of Xbox One. This will feature a 6 teraflop GPU and be capable of true 4K gaming, according to Microsoft.

Scorpio is exactly what one would expect from an upgraded Zen+Polaris (ZP) semi-custom chip. But it won't arrive until "Holiday 2017". This is a pretty good indicator of when AMD can be expected to have Zen and Polaris combined in a system on chip (SOC) or what AMD calls an Accelerated Processing Unit (APU).

This also points to the current state of the GloFo process that nothing like a ZP chip is possible in the near term. Once again, the main limitation is the physical size of the chip.

ZP APUs, when and if they arrive, can be expected to offer much better performance than the current generation of APUs that actually are shipping, Bristol Ridge. But Bristol Ridge (another 28 nm part) is basically dead on arrival, shipping in only one product, an undistinguished HP (NYSE:HPQ) Notebook. Bristol Ridge isn't at all competitive in performance with Intel (NASDAQ:INTC) Skylake or Kaby Lake. Once again, AMD is just bottom feeding.

While ZP APUs will offer improvements over Bristol Ridge, it's not obvious at all that they'll be better than Intel's parts that come out in the same time frame. By late next year, Intel will have moved on to its 10 nm process and Cannonlake.

Project Scorpio indicates that AMD's ZP APUs can be expected to help sustain console sales. They'll offer a significant performance boost compared to the 2013 era parts they will replace. In the area of PCs, the impact will be much less. I expect AMD to be forced into bottom feeding mode against Intel Cannonlake.

Investor Takeaway

I often receive comments and even direct messages asking why I'm so negative about AMD. It's nothing personal, and it's not based on emotion.

AMD just has so many problems. It's losing buckets of money, and its revenue is shrinking. Despite claims of "market share gains", these are just indicative of AMD's bottom-feeding status. AMD is fighting a two-front war against much stronger and profitable competitors in Intel and Nvidia.

But the big millstone around AMD's neck is Global Foundries. Because of Global Foundries, AMD is delivering products that are about a year (or more in the case of Bristol Ridge) behind its competitors, and there doesn't seem to be any hope of closing that gap.

The June quarter will be another terrible quarter with y/y declines in revenue and no hope for profitability in sight. Will this deter AMD investors? Probably not. AMD may even be able to claim once again growth in GPU market share, which appears to be all that's needed to sustain the hopes of its investors. I continue to rate AMD a short opportunity based on the unrealistic hope of AMD's investors.

Disclosure: I am/we are long NVDA.

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.