AMD (NASDAQ:AMD) just launched the company's new low powered cat core line, which is targeted primarily at mobile form factors.
In this article I will take a quick look at the performance enhancements made to the mobile APUs and show why I think the performance will not be what holds the chip back in market.
For the more technical sections I will provide a TL;DR (too long; didn't read) synopsis at the end of those particular sections.
Puma Is A Potent Refresh With Minimal Overhead
Author's Note: All AMD press images and benchmarks via HotHardware unless otherwise noted.
In a previous article I surmised that Beema and Mullins would be refined versions of Kabini and Temash, and explained why I felt the performance claims from AMD were highly plausible.
The guesstimates I made in the article proved to be fairly accurate, so now I will use some of the reviews of the chip to look at CPU and GPU performance compared to rival chips from Intel (NASDAQ:INTC).
Beema is aimed at notebook form factors, while Mullins is geared for tablet and 2-in-1 type usages. If you were to compare CPU and GPU clocks for the devices to their predecessors, you would find around a 20% or so (very rough estimate) higher clocks for the new offerings, along with a significant reduction in rated power consumption.
One importance to note is AMD is only launching 2 dual core chips; one for each segment. Regardless of whether we're referring to dual core AMD or Intel Bay Trail chips, Windows 8.1 is likely to struggle when only running on two cores, so more quad core chips will help come review time.
Specific Performance Enhancements
Based on AMD's provided figures, it appears the company was able to achieve better performance characteristics on GlobalFoundries' 28nm process. This allowed for reduced leakage current, helping the APUs achieve higher clock speeds at lower power levels.
AMD implemented a couple of unique features into these chips which allow the APUs to boost above base clocks for brief periods. This is a departure from Jaguar, which could only limit clock speeds to save power, but not boost clock speeds to give performance. AMD accomplishes this by a combination of monitoring the heat output of the processor and by actively targeting which applications should be boosted.
And because many of these changes didn't require a complete rework of the core architecture, the R&D associated with these designs was likely smaller compared to a completely new architecture.
Moreover, because the features implemented in Puma were done so at a higher level (meaning at a level above the core architecture), many of these advancements should be available for use in later product offerings.
TL;DR: AMD was able to release what appears, at least initially, to be a very strong portfolio of low power products. Many of these features were implemented in a manner that can be applied to future products. Because there were no significant changes to the actual CPU architecture, R&D spending was likely lower than designing a new core. And because the chips are made at GlobalFoundries, these orders will serve to satisfy WSA (wafer supply agreement) commitments.
To The Numbers
AMD's top of the line A10 Micro-6700T based on the initial benchmarks looks to offer comparable, if not slightly better, CPU performance compared to Bay Trail.
You'll notice AMD's tablet chip is essentially on parity with Bay Trail, offering slightly higher single threaded performance, but relatively the same multi-threaded performance.
In synthetic GPU tests, the AMD tablet chip nearly doubles Bay Trail's GPU score. Note the physics test more measures CPU performance, mimicking what we saw above in regards to a near performance parity on the CPU side.
Because the power consumption allowance is split between the CPU and GPU, it is often a good idea to look at real world performance scenarios as synthetic GPU tests, if well designed, should primarily stress only the GPU. Real world scenarios could tax both the CPU and GPU, which could limit GPU performance. However, above we see that Mullins offers significantly higher real world performance when compared against Bay Trail, and in 2 of the 3 tests offers double the frame rates.
TL;DR: AMD has a low powered CPU that is extremely competitive with Bay Trail, offering equal to slightly higher performance when compared against Bay Trail. The GPU performance of the Mullins chip is significantly faster than Bay Trail, backing up Dr. Lisa Su's claims regarding GPU performance made during CES 2014.
Specifically for tablets, power consumption down to the milliwatt is very important. AMD provided some power consumption numbers in different usage scenarios, but these numbers have not been independently verified.
Also, because AMD has implemented a boost function based on temperature, during prolonged usage scenarios (think gaming), performance may drop off after extended use.
However, for tablet usage, I can think of very few instances in which power consumption should stay peaked for prolonged periods, with the exception being gaming.
So after reading initial reviews, my biggest questions remain behavior during prolonged gaming sessions and how does the battery life compare to competitor products.
AMD vs. Intel In the Market Place
Intel is very vocal about going after the low end of the PC space, and tablets in particular. During the recent earnings call, AMD CEO Mr. Rory Read explicitly stated AMD had no intentions of going after the low end of the space at the expense of profitability.
Contrast this with Intel's ~$930M loss on ~$150M in mobile revenue and you can see the weight of Intel's claims. Intel really wants to cement itself in mobile computing, and will aggressively fight to do so.
In a previous article I discussed the value proposition of AMD's recently launched AM1 platform from an OEM's perspective. Here, thanks to BOM subsidies and cheap pricing from Intel, the roles are reversed and it makes more sense for OEMs to pursue Bay Trail products from an overall earnings standpoint.
And regardless of which company can put up the most favorable benchmark numbers, it's up to the OEMs to build and market devices, so focusing on pure performance in the tablet space may not be the most prudent, as the subsidies from Intel are likely enticing to OEMs.
The Silver Linings
Although I am not as optimistic as some that are bullish on AMD's tablet push, I do feel these chips will make extremely compelling products in larger form factors.
Intel has not announced any contra revenue subsidies for Bay Trail notebooks that I am aware of, and it wouldn't make sense to do so. Bay Trail is not a bad chip in the slightest, so over incentivizing OEMs to build with Bay Trail would eat into profit margin in regards to Intel's higher margin Core CPU line.
Slightly lower performance per watt matters far less in notebooks that utilize active APU cooling and bigger batteries. Although the initial PUMA reviews were all centered around the Mullins tablets, Beema powered notebooks are what I am most interested in. According to IDC, the notebook market is forecasted to remain a 150M+ unit market, which is very substantial.
Bay Trail notebook offerings are extremely efficient. But just because mopeds get good gas mileage doesn't mean they're fun to drive. Not that Bay Trail is a moped, but I'm rather simply pointing out that efficiency is only one piece of the puzzle.
Initial benchmarks of Mullins show that AMD can be competitive in regards to CPU performance, set the standard in GPU performance, and do so in a reasonable power consumption. Even if AMD were to double the rated TDP of the top end Mullins tablet chip, at that level of performance it would still be a great performance/watt proposition for notebooks.
Differences in the way Intel and AMD market and report power consumption make it difficult for direct comparisons to be made. The above two graphs together depict these differences. A low power Haswell chip consumes ~26W under load, despite being rated for 15W. Whereas an AMD chip could consume less than the rated power consumption, as AMD reports the max that a chip should consume. In the first graph, the Intel and AMD chips are respectively rated for 17W and 15W of consumption, but we can see at the platform level (meaning memory, hard drive, etc.) the entire AMD device consumes almost half the power of the Intel device.
I bring this up to point out that notebook form factors will give Beema more room to run, allowing higher clock speeds and higher performance. Because of this, I think Beema will likely have a stronger showing in notebooks than tablets, and I am not holding my breath for truly compelling Mullins tablet wins.
AMD has mentioned that most of the OEMs the company partnered with to launch Kabini and Temash products should be launching Beema and Mullins designs. Specifically, attention has been called to Lenovo (OTCPK:LNVGF) and Samsung (OTC:SSNLF) (source: PCWorld). And to quote TechReport on the Lenovo design win:
Happily, Beema may have an easier time working its way into the marketplace. Lenovo has already announced several laptops based on it: the Flex 2 and the B and G series. Those systems are coming in June, and they'll feature a special, A8-branded version of Beema that AMD offers "for select opportunities." (That model isn't part of the standard Beema lineup.)
Samsung initially launched a specialized version of the A6-1450 Temash APU in the Ativ 9 Lite. If AMD in fact launches a special version of Beema for the largest IHVs (hardware vendors), it's possible that we could see some better uptake with Beema than compared against Kabini.
From an investor's standpoint, Q2 for AMD runs through April, May, and June, and we should see some of these products come to market this quarter. During the Q1 earnings call, AMD stated they had begun volume shipments of these products to OEMs. The types and number of design wins could give a clue as to how much these chips have the potential to impact CS revenues. Also be mindful that if you're in the US, we actually get some of the more interesting designs later, so keep an eye out for international design wins.
Benchmark bar graphs can only tell someone so much information. While they have their place, I will venture out on a limb to say that OEM profit is a more influential factor, and Intel is obviously gunning for the tablet space. This is likely going to limit the success of AMD's Mullins tablet chip.
Unfortunately OEMs continue to flounder as they try and catch up to the changing PC market, which motivated me to call attention to how they are hamstringing themselves. In this article I break down some of the costs of various components and explain how an OEM could increase the BOM by around $50 to $100 and launch devices that are much more attractive. But instead we see the continual slew of 15"+ form factors with washed out screens and small batteries.
My hesitation with AMD's success for Mullins stems from the inability to gain traction with Temash. This video demonstrates how woefully inadequate Clover Trail was at running Windows 8. Windows is an albatross of an operating system and really overpowered Clover Trail. Although the dual core Temash would've struggled to run it well, it was much more suited for Windows than Clover Trail. AMD had a slew of reference designs based on Temash at CES 2013, and the chip was shipping well before back to school season. Rather than aggressively attacking the market, OEMs seem to have largely waited on Bay Trail and the resultant subsidies before branching out and attempting Windows tablets in earnest.
So on paper, it appears Mullins is a better design for tablets than Bay Trail. Tablets are highly visual devices, and I feel the graphics capabilities of Bay Trail have been one of the design factors holding the chip back. Mullins offers much better graphics performance, roughly equivalent CPU performance, and competitive, but likely lagging, power consumption. Performance comes at a power price, and ~2x the graphics performance of Bay Trail could very easily justify the extra power consumption depending on the user. Regarding CPU performance, anyone that wants to call attention to minor differences between AMD and Intel CPU performance, in my opinion, is doing nothing more than creating a straw man to attack. A several year old desktop i5-2500K is several times more powerful than Bay Trail. Show me anyone that wants to do real productivity on a Bay Trail tablet and I will show you someone that watches paint dry and water boil for fun. The CPU differences would be largely lost among the majority of the consumers, and most probably wouldn't even be able to tell you which CPU was in their tablet. When they walk into BestBuy (NYSE:BBY), unless they're tech savvy, they're more likely to be influenced by overall design, consumer reviews, and price.
If OEMs actually build an attractive Mullins device priced around $450 or under, I would expect that device to do quite well. AMD's Roy Taylor once jokingly referred to AMD as a starfish and Intel as a whale, suggesting whales are in trouble when the tides go out and the whale is too close to shore. Because Intel is the large player of the two in terms of market share, it takes substantial unit growth to see real revenue growth. But because of differences between tablet and desktop prices, it takes a disproportionately larger change in tablet chips to positively affect revenues for Intel. AMD, conversely, only needs a couple compelling design wins.
That being said, I am going to keep my tablet hopes shelved until I see such a device. Free Windows for tablets could indeed be a tailwind in terms of bringing a cheaper device to market, but I won't put the cart before the horse. If I see a compelling high volume product launch from a major OEM, I will gladly change my view.
Moving to the notebook space the pressure doesn't seem quite as high. The majority of Intel's Bay Trail chips, regardless of form factor, top out around 2.4 GHz. I do not believe any clock above 2.7 GHz save perhaps one or two.
This allows AMD to match or exceed Bay Trail's CPU performance in notebooks and handily beat the chip in GPU performance. A couple bucks for a larger battery from the OEM could easily negate any differences in power consumption between Intel and AMD, offsetting any performance/watt advantage for team blue.
But again, design wins will play a big factor. During Q2 and later we will see design wins for AMD's Beema chip, and perhaps Mullins. The number and types of design wins, as well as the timing of appearance, could be indicative of success and the ability to meaningfully impact revenues.
In an episode of The Simpsons, Lisa told Homer she had a rock that would keep bears away, and pointed out there were no bears. Homer then offered to buy that rock. Pointing out slight overall differences in performance between the two and citing these differences as reasons for AMD's trouble in obtaining good designs is no different, in my opinion, than Lisa's rock. Intel is several times larger than AMD, and likely has a marketing budget larger than AMD's R&D expenditures. When was the last time you saw an AMD tablet commercial on prime time TV?
AMD's success will rely more on compelling designs from OEMs leading to word of mouth advertising via positive consumer reviews on retail websites. AMD's low power portfolio is very competitive, but it's up to OEM implementation of these products, hence my fixation with design wins more than total performance.
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. I actively trade my AMD position and own both shares and options. I may add to or liquidate all or part of my position in either AMD or INTC at anytime.