Recently benchmarks were leaked showing what is believed to be Intel's (INTC) Bay Trail-T running circles around the competition. Since the benchmarks have been released, two articles on Seeking Alpha have been published here and here, and a few others on various websites.
I would like to point out that making conclusions based on one benchmark is probably never the right answer. Let me demonstrate why. I will summarize all the technical jargon at the end of the next section for those who do not enjoy the details.
A Deeper Look into Various Benchmarks
Image taken from ArsTechnica.com
Image taken from here.
Image taken from here.
Intel's Bay Trail is supposed to have 2-3x the performance over Clover Trail, and based on the leaked benchmark showing the Bay Trail-T scoring roughly 43,000 in AnTuTu, the claimed performance increase is believable.
But AnTuTu is only one benchmark. From what we have seen, Intel's CPUs and SoCs seem to fare a little worse in GeekBench. For comparison, Intel's Clover Trail+ z2580 is beaten by Tegra 3, and has roughly half the performance of Qualcomm's (QCOM) Snapdragon 600 SoC, or roughly one-third the performance of Nvidia's (NVDA) Tegra 4 and Qualcomm's Snapdragon 800. So you can see that if Intel hits its metric of 2-3x performance gain, this will put Intel in line with the best the competition has to offer in GeekBench, but ahead in AnTuTu.
Also noteworthy, Intel's ULV Ivy Bridge i3 scores roughly 3600-4600 in GeekBench (depending on 32 bit vs 64 bit). So if Bay Trail does indeed offer 3x the performance over Clover Trail generation Atom, this will mean that Bay Trail performance should be on par with Intel's lower priced Core i3 CPUs. One of the biggest advantages of Haswell is battery life and power consumption. I would like to remind those reading that Bay Trail is targeted specifically at lower TDP designs - meaning smartphones and tablets. To make my point, look at the TDPs of Bay Trail and Haswell CPUs. The Core i5-4200u is a 15w TDP part. For comparison, Intel's Bay Trail-T will most likely be at 7w or below. So given comparable battery sizes, the frugal power consumption of Bay Trail may also make normalized battery life of Bay Trail competitive/better than with the more expensive Haswells. Now notice the price of the i5-4200u? It has a tray price of $287, but if you figure OEMs pay half of this price if they're buying in bulk, that corresponds to a price of $143.50. I do not believe Intel has published any pricing information for the i3 Haswell CPUs, so here I will assume OEMs pay roughly $100 for an i3 CPU. If Intel maintains prices between generations of Atom processors, and we ignore any bulk discounts OEMs may receive on Atom processors, this means Intel is offering a similarly performing processor, both in processing power and power consumption, to its Core i3 at roughly 40% of the price. The Core i3's advantage will most likely show up mainly in graphics performance, but this is a guess on my part.
Lastly, I would like to point out that based on AnTuTu, Clover Trail+ to Bay Trail performance increased by roughly 70% (26000 vs 43000). Applying a 70% increase to Clover Trail+ to come up with a GeekBench score yields 1426*1.70 = 2424. Adjusting by 1,200 points to account for the shift to a 64 bit environment yields roughly 3600, still behind Tegra 4 and Snapdragon 800. Here I'm not trying to contradict myself but rather calling to light 1 of 2 scenarios. If Bay Trail reaches its claimed performance increase, this brings credibility to the possible cannibalizing of higher end CPU sales. If Intel misses the performance mark with Bay Trail, it will not be a runaway victory over the competition.
Point - Bay Trail does indeed appear to come out ahead of the competition in the AnTuTu benchmark, but will most likely be on par or behind the competition in GeekBench. But the most important benchmarks for mobile applications are missing - battery life, power consumption and real world user experience. These chips are targeted at lower power devices, so comparing performance without power consumption is fairly pointless. And given the fact that Bay Trail is targeted for lower TDPs, its battery life will likely be on par with Intel's more expensive Haswell CPUs, and possibly at similar performance levels of Intel's Core i3 ULV parts. Most likely the only large advantage most users will care about will be graphics for the i3.
What if Intel Outperforms the Competition?
Recently an article published on Semiwiki made the case that the smartphone sector is seeing much of its growth with cheaper devices in developing countries. With these devices selling for between $50-$75, there is no room for OEMs to offer better performance by placing a $40 SoC in their products. If Intel does decide to go after this segment, they are going to be competing on price with highly integrated, cheaper ARM variants. And, in my mind at least, it does not make much sense to spend the budget on processor performance, given what most people use their smartphones for.
To differentiate its products, it is my belief that the money would be better spent on a better display, or slightly larger battery. For an idea, here is a cost breakdown of a few Samsung smartphones. Notice the price of the SoC is at or below $30? The price of the Snapdragon 600 is listed at only $20. These are not low cost smartphones. So you can see why I assume that Bay Trail will be relegated to more expensive devices. And for my final point, think about the components that go into making a phone. Bay Trail graphics will likely be on par with an iPad 4, meaning it doesn't make sense to pair Bay Trail with a cheaper screen. Regarding CPU performance, it would not make sense to pair Bay Trail with bargain RAM, creating a memory bottle neck. Same for storage. If OEMs use Bay Trail, they are more than likely going to have to use more expensive components elsewhere in the phone, driving the BOM even higher. Although I am focusing mainly on tablets, the smart phone discussion is still relevant for tablets.
While Apple has been at the forefront of the tablet revolution, the current market expansion has been increasingly fueled by low-cost Android devices. In 2013, the worldwide average selling price (ASP) for tablets is expected to decline -10.8% to $381.
As tablet prices drop, and given the fact that a Snapdragon 600 should offer somewhere around 60-80% of the performance of Bay Trail at half the price, this will put some pressure on Intel. Intel's biggest advantage in this scenario is the ability to run x86 software.
Lastly, the most recent numbers I have seen on PC sales have a decline of 7.9% YoY. Fellow contributor Ashraf Eassa calculated roughly an average ASP of $120 for Intel's PC chips. Also according to his article, Intel shipped 280M PC chips last year. If we assume that IDC was off, and PC Sales decline by 5% vice 7.9%, this would mean that Intel's PC division sees a decline of 14M units YoY. To make up this loss with Bay Trail, Intel would have to ship 42M Bay Trail units, assuming OEMs are paying roughly $40 per Intel's ARK website, which puts them at the high end of the price range. If Intel competes on price, and Bay Trail has an ASP of $30, this number balloons to 56M. If IDC numbers hold true and Bay Trail has an ASP of $30, the number is closer to 90M. For comparison, roughly 58M iPads were sold in FY 2012. Given the fact that we're already over halfway through 2013, and we have not seen any design wins in the realm of an iPad or an iPhone, I believe Intel has little chance of making up for lost revenue via smartphones and tablets, at least in 2013 if IDC numbers are to be believed.
Bay Trail vs. the Competition
I have shown why I feel that even if Bay Trail beats the competition in the AnTuTu benchmark I do not consider this a clear victory given the fact that performance will most likely be on par in GeekBench, we have not seen any real world battery life metrics, and that even if Bay Trail's lead holds in real world performance over the competition, that gives Intel roughly a 20% performance advantage over the competition. I will say the average user may not notice a 20% performance increase.
One of Intel's biggest advantages is its process lead, which aside from better power consumption allows them to pack more transistors per square mm.
Image taken from semiaccurate.com
Image taken from SiliconMadness.Blogspot.Com:
I believe the die shot of the Bobcat cores are made on a 40nm process. I have included the second screen capture to give you an approximation of size in regards to cache, cores and GPU.
Advanced Micro Devices' (AMD) Jaguar SoC takes up roughly 107mm^2. Keep this in mind when real world performance benchmarks (not GeekBench/AnTuTu benchmarks) are released, and when we have die size of Bay Trail. This will give you an idea of performance/mm^2. Also noteworthy is keeping in mind everything that is integrated on the chip.
So Intel packing more transistors in a given area could allow it to lower manufacturing cost, but given the fact we have not seen Bay Trail die size yet, this is not a given. Also, above I have shown that while Intel leads in the AnTuTu benchmark, it could lag in the GeekBench suite. Meaning Intel's process lead in this case may only give them a power consumption advantage, not a performance advantage. And we also have not seen the cost of Bay Trail.
Ashraf does great research, and I look forward to reading his articles when we have more real world information on Bay Trail to get a good analysis of performance, cost and die area.
With PC sales apparently still on the decline and Intel's current low power, low cost offering to compete being based on Clover Trail for yet a couple more months, Intel may be hard pressed to make up the revenue decline. Using Ashraf's ASP of $120, and a 5% decline of PC sales YoY, Intel's PC group could possibly lose $1.7B in revenue or more.
Looking at Intel's IA group's operation, we can see it actually operated at a loss of $1.3B last year.
Image and quote taken from INTC.com:
Net revenue for the Other IA operating segments decreased by $627 million, or 13%, in 2012 compared to 2011. The decrease was primarily due to lower IMC average selling prices and lower netbook platform volume. To a lesser extent, lower netbook platform average selling prices contributed to the decrease. These decreases were partially offset by higher ISG platform average selling prices.
So if Intel is competing for more than half the year with the older generation Atom, Intel will be hard pressed to make up for lost revenue this year with Clover Trail based Atoms. However, Clover Trail design wins help pave the inroad for Intel into mobile. Intel also has more than just mobile to make up for declining revenue, which you can read about here.
Intel faces more competition in the mobile sector than they're accustom to in the PC space. Intel needs a couple high volume (think iPad or similar) design wins. But my optimism for Intel lies with cheap Windows 8 tablets. For example, the Acer Iconia W3. According to the Microsoft Store's website, the W3 comes pre-loaded with Office Home and Student, and the 64 GB version sports a $399 price tag. Not terrible given the inclusion of Office. The downside to the W3 is that it only includes 2 GB of ddr2 memory, and it's powered by z2760, so graphics will not be very strong. Swap that z2760 for Bay Trail, and give me 4 GB ddr3, and I'd open my wallet in a heart beat. I would gladly pay an extra $30 to cover an upgraded SoC and better RAM in this instance too. The Office preload is a major step in shifting tablets from consumption to productivity. This could also greatly benefit Microsoft (MSFT) longs. We're seeing $400 Win 8 tabs with MS Office. Office Home and Student retails for $140.
If investors believe in Intel's ability to dominate the mobile sector with Bay Trail, this year could present several good entry points to initiate or add to long positions. If you're more skeptical, the declining PC market could provide the opportunity to be patient and watch from the sidelines and wait, without missing too much upside, especially if the big boy analysts keep swatting down the price for us little guys.
Additional disclosure: I may initiate a long position in INTC, QCOM, or MSFT at any time.