Intel (INTC) has broken ARM Holdings' (ARMH) stock as a result of finally winning some real, recognizable design wins with its "Atom" system-on-chip line for tablets. ARM shares are now in an undeniable downtrend, and it seems that after commanding a premium valuation for so long, the hot air is finally being let out of the balloon. Let me make myself clear in that I believe ARM runs a great business, but the valuation is completely out of line with the firm's earnings power, and now the firm has extreme competitive pressure from Intel that will only get harder over time. The X86 myth is "busted," and in fact, recent research from ABI Research shows that not only is Intel on par with the best ARM designs around, but is in fact - with its older 32nm, 5 year old designs - better performing at the same or lower power.
ABI Research Gets Some Real Data
For a while, ARM and Intel have been in somewhat of a "powerpoint pissing match" in which ARM and Intel would both claim superiority in performance, power or some other relevant metric. However, for the first time, a credible research firm has run both performance and current measurements.
Now, a brief aside: current is not the same as power consumption. In fact, we need to keep in mind the following equation from physics:
Power = Current * Voltage
The numbers that are given in the comparison are current measurements, but according to Qualcomm, the Krait 200 hit 1.7GHz at 1.05V, so I would assume that at peak 1.9GHz frequency with micro-architectural enhancements, the voltage required at full-tilt is slightly higher. On the same token, Intel's official documents list the Clover Trail+ part as having a voltage range of 0.3V - 1.20V, so I would assume that at full load all of the ARM parts and the Intel part all use just about the same voltage.
This means that for the purposes of this comparison, it is not far-fetched to assume that voltages are roughly the same and that the current is really what drives the power consumption of these parts. Without further ado, here were the results from ABI Research's tests:
- Memory Performance on the Intel part was 8703 and average current draw was 0.55A with maximum of 0.9A. The quad core Cortex A15 "Exynos Octa" scored 3838 and drew average current of 1.27A and maximum of 1.56A. The Qualcomm quad core Krait 300 0 did better, scoring 4235 and drawing 0.56A on average and 1.08A at peak. Intel clearly wins this battle from a performance/watt standpoint by a landslide.
- In CPU performance, Intel's part scored 5547 and drew 0.85A while doing so on average with 1.05A at peak. The ARM A15 quad core part scored 5277 with average current draw of 1.38A and maximum of 1.71A. Qualcomm did better in performance scoring 5378 but drew 1.79A on average and 2.1A at peak.
So in terms of a CPU that goes into a smartphone, Intel's part performs better in both CPU and memory, while at the same time drawing less power, and does so with half the number of cores at the competition - all on its 32nm process technology.
No Wonder ARM Is Down
It's clear to me the ARM bulls who claim that Intel won the ASUS MeMo Tab, Samsung Galaxy Tab 3, and ASUS Fonepad Note on price is likely misguided. Intel brought Clover Trail+ to the table with both superior performance as well as superior power consumption with its current generation parts, which suggests that Intel's next generation Atom, which has 2-4 much more powerful cores than its current generation in the same power envelope, should be a show stopper for both tablets and smartphones.
The ARM vendors got a break as Intel still had to bring Android and its ecosystem up to speed for its X86 instruction set, and it also got an additional break as Intel did not have an LTE part to compete with in the developed markets. Well, with Intel having demonstrated LTE, and with Intel's applications processor performing this well on a 5 year old design that will be replaced this year, the device manufacturers - particularly the companies such as ASUS, Acer, Gigabyte and more who have been strong Intel partners for years in the PC space - will have no choice but to look to Intel's parts for the longest battery life and highest performance.
Intel is a sleeping giant that has awoken, and when the momentum is in full force, it will be very difficult to bet against Intel or, probably more importantly, on ARM at current valuations over the next 12 months. Just remember that every design socket that Intel wins (and they will probably be at the high end at first), ARM loses a "quad core" or even an "octa-core" big.LITTLE type of design, which cuts off parts not only with plenty of ARM CPU content, but also GPU IP content as well. The knife cuts both ways.
On any bounce this is a good short candidate, as I expect the buy side is implementing the "long Intel/short ARM" pair trade as we speak.
But hey, on the bright side, Intel is probably not going to design a Cortex M0 competitor anytime soon, so ARM's low royalty but high volume markets are still perfectly safe...well, from Intel anyway.
Additional disclosure: I am short ARMH