This is not my article, nor is it necessarily a representation of my beliefs. It is just some topics to ponder.
http://www.extremetech.com/article2/0,2845,2341005,00.asp
4 Page full article, follow the link.
The Next Xbox--What to Expect
A typical five-year span before the next generation would mean a launch of what I'll call "Xbox Next" late in 2010, but with the PS3 and Wii hitting the market late in 2006, one could safely push that to 2011. What's more, games of the current generation are bigger and more resource-intensive than before, taking longer to bring to market and requiring longer-term sales to be profitable.
When will the next generation begin? When will the next Xbox arrive? Only a small team inside Microsoft, surely working on the specs and design of the Xbox Next, have any idea. And they're not talking. If you ask me, I would say a launch in time for the holiday season of 2011 is reasonable, and a launch a year later in 2012 is equally possible, but anything beyond that is unlikely.
So, given that launch timeframe, what will the hardware be that powers the future Xbox, and why? I don't have any inside information, but I do have a few ideas that make sense. Who knows, maybe Microsoft is listening
Microsoft's Priorities:
Reliability, affordability, and next-gen power.
Storage: Blu-Ray and SSD
On Blu-Ray: I think Microsoft will go with a Blu-ray drive. Right now they're sort of expensive, but by late 2011 they'll be pretty cheap, even in a 4x or 6x version that will have the desirable data transfer speeds (data transfer speeds with the PS3's 2x drive are one of its drawbacks, and future games will be reading a lot more data off the disc).
On SSD: First, the cost curve of hard drives stinks. Yes, the price-per-GB goes down dramatically, but the base cost of the drive itself hardly ever changes. It's a sensitive mechanical device with a rotating platter (or two), and it costs X amount to manufacture regardless of the data that platter holds. While SSDs will be expensive in late 2011 or 2012 relative to a smallish hard drive, their base cost will continue to fall throughout the life of the console at the exponential rate of Moore's Law. Eventually, the storage could be down on the motherboard in just a couple of flash chips.
SSDs are best at what they're needed for most in a console system—bolstering limited RAM by caching data from the disc to be streamed in. Nearly every game does this now with the hard drive (Microsoft reserves some HD space for game caching) but the ability for SSDs to read little random bits of data from any memory location 10–100 times faster than a hard drive can really open up possibilities for developers.
The CPU: A bigger Xenon
Without too much design work, the current cores could be upgraded for improved performance. Increasing instruction fetch and decode capability, and doubling the SIMD unit from 128-bits to 256-bits, would dramatically increase the performance of each core without totally changing the programming model. This would make each core bigger, perhaps driving the transistor budget up to 250 million transistors for a similar 3-core CPU. But you wouldn't want three cores, you'd want 9–12 cores. And you'd need more L2 cache to keep those cores fed, perhaps 4MB.
The biggest advantage of this approach is that developers would go into the Xbox Next with effectively years of experience in optimizing for its CPU. The cores would be higher performance, and the SIMD execution would be wider, but that's an easy adjustment. Already developers target a six-thread processor with tools designed to scale to "n-threads," so scaling to four times as many threads would be simple, compared to learning the multithreaded, in-order execution tricks required to optimally use the Xbox 360. Peak CPU performance would be at least 8x, at worst 4x, and the tool chain for development would be the very definition of mature and reliable from day one. Oh, and backward compatibility would be a cinch, from the CPU side.
Memory:
Shifting to GDDR5 probably makes the most sense. Of the available standards, GDDR5 will be the one that is at the appropriate level of maturity to be very cost effective by late 2011. It doubles the memory bandwidth at the same clock speeds and bus widths as GDDR3, but most importantly it allows for very high clock speeds at very reasonable power levels. If Microsoft tries to keep the motherboard down to a low number of layers and trim costs a bit by sticking to a 128-bit bus, it could achieve 76.8 GB/sec by using 1.2GHz GDDR5 RAM. But that's not really enough. I think the smarter way to get the best overall system performance would be to go for a 256-bit bus. This makes the motherboard require more layers and costs a bit more, but it lets you use slower RAM while getting a lot more memory bandwidth. Because slower RAM is cheaper, it also means you can have more of it at the same cost, or offset the cost of the wider bus. I think Microsoft should aim for six times the memory bandwidth in the next Xbox, or 134.4 GB/sec.
Controller
I humbly suggest that the Xbox Next's controller really should be two controllers. That's right, I want Microsoft to package two controllers in every box. One will be similar to the current controller—the dual-stick design with four colored face buttons, triggers, and shoulder buttons is a fantastic layout for a great many extremely popular games. It's critical if backwards compatibility is a concern. This new Xbox gamepad would be somewhat different, thoug.
What about this second controller? That should be the motion controller. Yes, motion control has a place in games, and not just Wii-style casual things. In fact, the new generation of motion sensing technology is really quite fantastic. Where motion control falls apart is when it's an afterthought as it is on the PS3, or when you try to do it with cameras looking at your hands or face or whatever (this never works right once you get it into the living room with strange lighting, it's imprecise, and it's hard to fully integrate into games to control menus, pausing the game, and so on).
Sounds like some very smart and logical updates to the hardware. He also commented on graphics and on the possibilty of using the Larrabee GPU, but this bit was long enough as it is so if you want to read his thoughts on the GPU hit the link at the top.
I think the most interesting and smart decision is to add more cores, and to update their Xenon CPU structure. This makes it so the next Xbox is still the easiest to program for, while increasing the performance levels by as much as 8x the level of the current CPU. That and adding an SSD would be perfect if they do leave the "Install to Hard Drive" option as it would make for insanely quick loading times. What do you think? (Sorry if this was already posted. I did a search and didn't find anything) Here is the overview for someone not interested in reading it all:
The Xbox Next
So that's the Xbox Next, as I see it. Microsoft's best bet is to go with a CPU that mimics their current one, only with four times as many cores and some significant core upgrades. Or alternately (and especially if the company decides that backwards compatibility isn't important) to use something like Intel's Nehalem, sacrificing ownership of the design for a very robust and well-documented instruction set architecture that is flexible enough to bolster the system's graphics capabilities and gives you a year's jump on the rest of the industry in moving to future process technologies.
The CPU connects on a fast 20GB/sec or faster interconnect to the GPU, which is connected to RAM in a similar unified memory architecture as was in the Xbox and Xbox 360. The GPU uses the same separation of main GPU for texture address/filtering and shading and a separate embedded DRAM chip with logic for rasterization and anti-aliasing. Only this time, of course, more shaders, more eDRAM, more render back-ends, and enough flexibility in the GPU to give it at least DX11-like capabilities, where Compute Shaders can be used to further game physics, AI, and video processing. The RAM is roughly 3GB of GDDR5, giving developers six times the space as before and six times the bandwidth. A hard drive is out in favor of a SSD, around 128GB with future "elite" versions giving more storage, and possible entry-level systems providing 32GB or 64GB. This gives much greater performance for caching game data than a HDD, and over the life of the system becomes less expensive as well.
