Hot Chips 2005: Xbox 360 Exposed
One of the big highlights of last week's Hot Chips conference at Stanford University was Tuesday evening's (last presentation of the show) unveiling of a number of details about the inner workings of the Xbox 360 by Microsoft engineers Jeff Andrews and Nick Baker. I'm more than a little surprised that the mainstream tech online and print press (and blog community) hasn't yet written up the news….so I'm happy to oblige, and to share the info with you. I scribbled all over my copy of the materials, but the conference staff graciously shipped me a duplicate set, which just arrived. I've scanned and posted it as multiple PDF files:
- Foils 1 and 2
- Foils 3 and 4
- Foils 5 and 6
- Foils 7 and 8
- Foils 9 and 10
- Foils 11 and 12
- Foils 13 and 14
- Foils 15 and 16
- Foil 17
One other tech journalist who was there, Dean Takahashi (now of the San Jose Mercury News, and the author of the outstanding book Opening the Xbox on the development of Microsoft's first-generation console), did do a short writeup (and a nice one at that). I agree with Dean that the elegance of the platform is evident in the 'little' things, such as the ability for the unified shader-based graphics core to directly fetch information from the CPUs' shared 1 MByte L2 cache. The platform exhibits tremendous bus bandwidth and processing muscle potential; how efficiently software developers will be able to transform that potential into reality remains to be seen.
At the conclusion of Andrews and Baker's talk, I voiced two inquiries in the public Q&A session. You'll note that the presentation generically refers to 'DVD' support, so I asked them if they could comment on the rumours that Microsoft was planning on eventually including a HD DVD blue laser-based drive in the Xbox 360. They deferred, noting that these were 'product proliferation-specific' details that they weren't prepared to address.
I also pointed out that a few days earlier, graphics guru John Carmack from id Software had, in his QuakeCon keynote, expressed his disappointment that all of the next generation consoles included CPU cores that had eliminated support for out-of-order instruction execution, and his concern that this decision would lead to dramatically inferior MIPS efficiency compared to x86 CPUs. I pointed out, and Andrew and Baker agreed, that Carmack's primary to-date expertise was with the x86 architecture, and that this factor might have coloured his comments; Andrews and Baker also admitted that this decision wasn't an easy one and was the outcome of the necessity to balance architectural efficiency against per-core die size (and cost), power consumption, and other variables. Keep in mind that this issue is particularly relevent given Microsoft's stated desire to enable some or all first-generation Xbox (based on an Intel x86 CPU, along with an Nvidia-developed core logic chipset and graphics core) titles to run via emulation on Xbox 360 (based on three PowerPC cores, along with an ATI graphics core).
My issue #26 article will be on next-generation consoles, both plug-in-wall (the Xbox 360, Nintendo's Revolution and Sony's Playstation 3….and let's not forget the perpetually coming-soon Infinium Labs Phantom) and portable (Sony's PSP, Nintendo's DS and Game Boy Micro, Nokia's N-Gage series, the Gizmodo, etc). It'll be a followup to my popular December 2001 writeup that compared and contrasted the then-leading edge Microsoft Xbox, Nintendo GameCube and Sony Playstation 3. Hopefully, by the time I file my story the Xbox 360 will be shipping and I'll be able to give you some hands-on feedback, and more (and more solid) information on the Revolution and Playstation 3 will hopefully be available as well.
