Belated Analysis: Intel Atom/Silverthorne
Wednesday 19th March 2008, 01:50:00 PM, written by Arun
EETimes published a nice piece on what media coverage got wrong regarding Intel's new Atom platform (based on Silverthorne) earlier, it's well worth a read - but other points may also be worth discussing, from Intel's roadmap to others benefiting such as Imagination Technologies (PowerVR) and Icera.
Regarding Intel's roadmap and ambitions, AnandTech's article has by far the most information: the Menlow platform (based on Silverthorne) will be sold in UMPCs, MIDs and some laptops in 2H08, while Moorestown in 2009 will no longer support PCI which is required for Windows, so it'll be targeted at MIDs and large 4-5" smartphones. Finally, in 2010/early 2011, a 32nm shrink will be aimed at iPhone-sized smartphones. And they don't mention anything for smaller smartphones, which will likely remain a large part of the market.
While these platforms will eventually become direct competitors to ARM's processor IP, there's still plenty of UK IP in them. The graphics core that is integrated in the Poulsbo chipset (and codenamed "GMA500" by Intel) seems to be none othe than the SGX 535 from Imagination Technologies. Exact specifications are unknown, but it does support OpenGL ES2.0 *and* DirectX9 at the very least. Many Silverthorne-based devices will also have 3G access, and Intel's preferred partner for that seems to be Option Wireless and their GTM501 module, which seems to be based on Icera's 90nm baseband.
EETimes did get one important thing wrong though: the die size comparison between Silverthorne and ARM's Cortex-A8. The former is 25mm² on 45nm but, as they indicate, the core size only seems to be 9mm² (the rest is mostly cache and interfaces). However, they then compare that with ARM's claimed <3mm² on 65nm; but a much fairer figure for a number of reasons is 8.55mm² (link: see Page 2) as that includes NEON, place-and-route, etc. - it's not clear whether that includes L1 and you could probably do a bit better with some custom logic like Silverthorne, but >7mm² on 65nm vs 9mm² on 45nm is a much better comparison overall.
The risk Intel has with the 45nm Moorestown is that 45nm ARM Cortex-A9 SoCs coming out in the same timeframe will likely be lower-power and lower-cost while delivering higher performance, especially in dual-core configurations. And the dynamics likely won't change when comparing Intel's 32nm shrink to 32nm SoCs using the Cortex-A9. As for the supposed advantages an x86 ISA brings, they will be long gone by the time Moorestown or its 32nm refresh are available in finished devices given how many applications for Mac OS X, Windows Mobile and Google's Android will have been written by then.
Tagging
Add your comment in the forums
Related intel News
Nehalem Article @ RWT + 3.2GHz samples(?)
Opinion: Silverthorne fails but PowerVR impresses (+Montalvo trouble)
Havok physics software on PC soon-to-be free for non-commercial use
Intel purchases young game development house Offset
Larrabee: Samples in Late 08, Products in 2H09/1H10
Intel results indicate consumer spending strength; investor ignorance
Quick Analysis: Nehalem CPUs & Sockets
Larrabee and Intel's acquisition of Neoptica
Intel launches Penryn-based QX9650
Guess not as it features the 945GSE which I do not think is a PowerVR part.