Telirati Analysis #12 Intel and Rockchip: Why Intel Isn't Inside Your Phone

SoCs are mostly the same

One of the mysteries surrounding the question of why Intel hasn't gotten inside your phone is that systems-on-a-chip (SoCs) are all quite similar. Some CPU cores, a GPU, and i/o peripherals share a bus, all on one chip. If high-level architecture is so similar, and apparent barriers to entry low enough that numerous new entrants have flourished, what's up with Intel?

A combination of factors

If new entrants can get in the game, and SoCs are difficult to differentiate, it is unlikely that you can find one factor, or one problem to solve, that will unlock the mobile market for Intel. Here are some of the factors cited over time as keeping Intel out of your phone:

Power management: ARM CPUs were designed for low power applications and ARM licensees have all the expertise in integrating systems that result in low power SoCs. Intel has largely overcome this barrier, though it took more product generations than it should have, and Intel may still trail NVidia in bringing big GPUs to low-power SoCs.

Windows and netbooks: Netbooks are relatively higher margin products, Intel knows all the netbook OEMs, and netbooks have bigger batteries and can tolerate higher power consumption. The Intel instruction set architecture and Microsoft's Windows technology strategy kept ARM-based competitors out of the mainstream Windows 

GPUs: Intel hasn't brought their own GPUs into low-power SoCs. Currently, Intel uses Imagination Technologies (IMGTec) PowerVR GPUs in Atom chips. There may be an IP licensing price mismatch between IMGTec and the kinds of OEMs who might first bring an Intel SoC to market in their device. There is no open source driver for PowerVR GPUs, which limits the experience Linux developers have with that GPU. That said, PowerVR is in Apple's mobile products and many other ARM-based SoCs. There is nothing apparently deficient about that GPU architecture. IMGTec recently acquired MIPS and looks like they will get in the CPU licensing game.

Performance: Benchmarks for Intel Atom CPUs have always been competitive with ARM CPUs, even though the fastest ARM CPUs, like Apple's A7, outperform the top Atom CPUs. So raw performance numbers can't be the whole answer. The rise of modestly priced, medium-performance products like the Motorola G further reduces the importance of ultimate performance. Toolchain issues are also unlikely to holding Intel back, so while one can conclude that performance is part of why Intel is not in your phone it is not likely to be a major part. Performance may become a larger factor in combination with GPU performance if NVidia is successful with the K1 chip. General-purpose GPU computing is another factorin the performance equation. Renderscript is a little-known and little-used technology that 

Price: The question of price by itself embodies the complexity of why Intel has no significant mobile wins. ARM licenses their designs, as do mobile GPU designers. That has proven to be a powerful business model. Samsung and Apple are vertically integrated, integrating their own SoCs. ARM SoCs are often fabbed by contract manufacturers, except for Samsung, which operates their own fabs and also makes Apple's chips. And despite the vertical integration at the top of the market, there are low-cost fabless ARM SoC vendors. While prices are obscured by the difficulty of measuring transfer pricing in large users of CPUS like Samsung and Apple, you can conclude that diversity of supply means intense competition and that Intel has a difficult time convincing OEMs that they will be the low-cost vendor over the long haul.

Intel and Rockchip

It may seem hard to figure out why Intel entered into a relationship with Rockchip to produce SoCs. But it's a hard multidimensional problem they are trying to solve. Getting an experienced maker of low cost ARM-based SoCs that's got significant design wins involved is an approach that could solve more than one problem at a time. It's more alchemy than science, but Intel has much to gain and nothing in their history that would be get worse if this approach doesn't succeed.

One factor that isn't part of the answer is investment. Intel doesn't need Rockchip's money and Rockchip isn't taking any Intel money, expecting to make back their cost of development through selling this chip through Rockchip's own channels.

Both companies will sell the product, though it will be branded as an Intel product. This is significant because Rockchip has mobile handset customers and Intel doesn't. 

Another factor that can be eliminated from the answer is that Intel will license SoC CPU cores. This isn't an intellectual property license agreement in the same way that ARM licenses their designs, and it isn't being replicated with other SoC makers. Nor is it a foundry agreement. Rockchip is fabless.

IMGTec's venture into CPUs may also have motivated Intel to find a partner that can integrate GPUs other than PowerVR. Rockchip currently uses Vivante GPUs. But this does not necessarily augur a long-term relationship. Especially so if the next step on the road map is the integration of Intel's own GPUs into mobile chips.

Another indicator this may be just a bridging partnership is that the planned SoC will have a 3G radio. That may not be competitive is markets outside China and developing world markets, which are Rockchip's current target markets.

Not to say that Intel is exploiting Rockchip's customer base utterly without a road map for future collaboration, but it does appear that Intel sees this as an experimental way to acquire first customers in the mobile handset domain.

Strange bedfellows, and a limited relationship that leaves the future open for reconsideration. But this may be the way we get to see Intel-powered handsets in China. It may also be the way Intel learns to be competitive in a domain that has eluded them for complex reasons.


Popular Posts

5G: Hype vs Reality

A $99 Android Tablet That Doesn't Suck

The QUIC Brown Fox Jumped Over the Top of Carrier Messaging, or Allo, Duo, WebRTC, QUIC, Jibe, and RCS, Explained