Nvidia On Apple (And Others?): Additional Ruminations

Today, Nvidia predictably launched its GeForce 9400M, the core logic foundation of yesterday’s Apple new product unveilings (complete with yet another custom display connector, and no more FireWire on the MacBook…and whaddyaknow, software patches already!). As such, I thought I might flesh out the comments I made in yesterday’s writeup. Feedback, as always, is welcomed.

Nvidia’s new product has several advantages over Intel’s Montevina (aka Centrino 2):

• It’s a single chip, versus a multi-chip set, thereby delivering board space savings, and
• It delivers higher graphics performance than does Intel’s competitive integrated graphics core

But given the cooling relations between Intel and Nvidia in recent months, I doubt you’ll see a big rush from other Intel-loyal system OEMs to adopt Nvidia’s chipset in their products. And the graphics advantage is a bit of a moot point. Intel’s Mobile G45 chipset family variants also enable you to optionally mate them with a discrete GPU over PCI Express, as an alternative to the integrated X4500 graphics core. And Intel’s products’ integrated graphics performance is adequate for entry-level gaming…no matter what Nvidia’s demos, which artificially crank up the quality settings beyond what a typical user needs, might otherwise lead you to believe.

Ironically, I’m most excited about Nvidia’s product’s processing muscle (both in an absolute sense and relative to the Intel core logic alternative) in non-graphics apps…still image and video editing acceleration, for example, along with other GPGPU stuff. But from an Apple standpoint, such capabilities won’t get fully unlocked until Apple’s next OS iteration, ‘Snow Leopard’ OS 10.6, via the to-come OpenCL API. On Microsoft-powered platforms, we’ll need to wait for Vista-only DirectX 11. And as for today, as you’ll see more in-depth in a post to come from me soon, even Adobe’s latest Creative Suite 4 apps make underwhelming use of the GPU as an accelerator…for perfectly valid technical reasons which I promise I’ll explain then.

Speaking of moot points, the Intel Penryn CPU that Apple’s using in its new hardware is already on its way out the door, anyway. Intel’s next-generation Nehalem microarchitecture, in the initial product form of the Core i7 CPUs, will likely launch by year end. Nehalem uses a completely different interconnect scheme (QuickPath) between the CPU and chipset, along with integrating some functions formerly served by the chipset, such as a DRAM controller. And currently, nobody else (including Nvidia) has a license for that interconnect bus, so nobody else can sell chipsets for Nehalem. If I had to guess, I’d predict a complete Penryn phase-out in mobiles by the end of 2009, with Nehalem taking over all but the lowest-cost and lowest-power segements of the laptop business (which Atom follow-ons will alternatively target).

Granted, FTC antitrust pressure will probably compel Intel to sell a QuickPath license to someone else(s) sooner or later, but for now (and to be clear, initial Nehalem-based system design work is being done now)…

Followup: Looks like the MacBook Air updates were more substantial than one might believe at first glance. First-generation systems used a low voltage, low power, custom-packaged 65 nm "Merom"-generation CPU, but the latest MacBook Air design migrates to an "S" spin of Intel’s 45 nm "Penryn" processor. Hit this link, along with this earlier one, to garner an understanding of what the generational up-tick translates to in terms of feature set and predicted performance.

Followup II: Ars Technica has a nice summary of the new Nvidia chips (9300 and 9400), in both their desktop and mobile incarnations.