Preliminary benchmarks had shown that Intel (INTC) outperformed the competition in AnTuTu, while the results I had pulled from GeekBench showed Intel struggling to compete with Qualcomm's (QCOM) Snapdragon line of processors. Today while browsing the benchmark results, I came across GeekBench scores for a newer revision of Intel's Atom.
A Newer Version of Intel's Z3770 Appears
This result has 2 implications: that the CPUs used in the original leaked benchmarks could have been earlier revisions of the processor and AnTuTu could show improvements in scores, and it gives 2 data points of Intel having a very competitive solution.
The only reason an investor should care about benchmarks is it shows the potential of a product. Given that Bay Trail jumped ~50% between the "stepping 2" and "stepping 3" revisions, it is possible that AnTuTu could see a nice performance jump over the previous score. Sometimes GeekBench can spit out some outlier data as well, so I do not take the 50% performance bump as ground truth, but submit it as a potential increase in performance.
After recompiling the AnTuTu benchmark to prevent Intel's compiler from "optimizing out" steps it felt were erroneous (rendering portions of the test essentially useless), Intel's SoCs took roughly a ~20% performance hit.
Without doing any hard math (as I feel at this point it is kind of pointless), I will point out that based on the dates these leaked benchmark scores appeared, it's entirely possible that the leaked AnTuTu benchmarks showing Intel beating ARM based chips (based on the erroneous data) were based on the earlier revision of the processor, and the scores are likely higher based on the current revision of the Z3770.
I will say here that benchmarks do not guarantee performance, especially as mobile benchmarks have seen much scrutiny lately. Also both power consumption and the data regarding the platforms are still missing, so there are other difference that could affect performance that are unaccounted for.
Spoofing benchmarks is not a new practice, and is one of the reasons I say I never solely rely on them. In the end, in my opinion the most important benchmark is the number of stars consumers rate the end device. However, this new GeekBench score and the possibility that the leaked AnTuTu scores were based on an earlier revision of Intel's new Atom provide 2 data points demonstrating the possibility of Bay Trail-T being a very competitive solution.
Several leaks have appeared regarding Bay Trail's power consumption. None of the information that I had seen was jived with information available via Intel's official PR and on the website, so I have been waiting on data that matches more closely before writing this section.
This week, leaks from VR-Zone appeared showing different variations of Bay Trail-T at different SDPs, all around the 2W range. You'll notice that the chips run at "up to" 2.4 GHz, but the benchmarks I have posted above all point to a 1.46 GHz operating frequency -- meaning the chip likely operates at 1.46 GHz, but can "boost" to 2.4 GHz for short periods, provided average power consumption is maintained within the "SDP" numbers.
While there have been no power consumption comparisons made yet, if the leaks above are indeed true, Intel has a very compelling solution. It is very probable that Intel has doubled the performance in roughly the same power consumption, and at these low power consumption levels it can fit in sleek, thin tablets without a hitch.
Previously I had stated that I feel the most important reviews are consumer sentiment of end devices. If you look at the reviews of the previous Clover Trail generation tablets, other than OEMs putting them in unattractive devices, another common complaint was performance. Doubling the performance, along with supporting faster DDR3, should eliminate some of these sentiments.
I had planned on refraining from writing further on this subject until after the official launch of Bay Trail, but after seeing this I wanted to submit an update to my previous article based on this new data. Official data regarding Bay Trail should be released at IDF 2013.
Intel's upcoming 22nm SoCs for smart phones and tablets are an important milestone. First, they will be the first earnest attempt at a high-performance mobile solution from the ground up. This will provide a source of revenue to offset some negative pressures felt from the PC market. I have made the point prior that I do not feel tablets are nearly as important as smart phones. Intel currently has several design wins based on the Clover Trail Atom, and a common thought is that tablet sales have been chipping away at the PC market.
But as Intel is featured in very few smart phones, and the chances of consumers opting for only a smart phone over a PC in my mind are smaller, I feel revenue from SoCs placed in smart phones should be incremental to Intel's top line with fewer higher order affects that tablets can generate (i.e., tablets cannibalizing PC sales).
Lastly, the smart phone SoCs represent a larger volume market. If Intel is able to successfully penetrate the smart phone market, this represents a large volume of potential shipments, increasing fab utilization, and decreasing the excess capacity charges for not having fab facilities running at capacity.