
The successor to the PlayStation 3, apparently codenamed "Orbis,"
will use an AMD x86 processor with an AMD "Southern Islands" GPU,
according to rumors emerging last week. Xbox 360's replacement, purported to be named "Durango", is also rumored to use an AMD GPU—either a Southern Islands variant or an equivalent to a Radeon HD 6670—this time paired with a PowerPC CPU.
Though these rumors are thoroughly unconfirmed at the moment, they're
all well within the realm of plausibility. But if they prove true, the
Orbis and Durango will be decidedly mid-range at launch when compared to
top-of-the-line PC hardware. The Xbox 360, launched November 2005, and
the PlayStation 3, launched November 2006, were both cutting-edge
systems at their release. Their capabilities were unmatched by PCs of
the time. If these rumors are to be believed, the eighth console
generation won't be a repeat of the seventh.
The stupendous seventh generation
The Xbox 360's Xenon processor, a three-core six-thread PowerPC unit
running at 3.2 GHz, had a theoretical peak number crunching throughput
of 115 gigaflops. A
contemporary Pentium 4 at 3 GHz had a theoretical peak of around 12
gigaflops when the system launched. The PlayStation 3 was in a similar
situation; its Cell CPU,
jointly developed by IBM, Toshiba, and Sony, had a theoretical
throughput of 230 gigaflops. Contemporary Core 2 Duos were topping out
at 24 gigaflops at the time—and cost many hundreds of dollars to boot.
The GPUs found in these systems were not quite so impressive compared
to those available in desktop systems at launch, but they were still
high-end. Xbox 360's Xenos was built by ATI, falling somewhere between
the capabilities of its R520 (sold as the Radeon X1800 series, released
in October 2005), and its R600 (retailed as the Radeon 2900 series,
released in May 2007). The PlayStation 3's Reality Synthesizer was
designed by NVIDIA, as a slightly cut-down G71 (marketed as the GeForce
7900 series, released in mid-2006).
In short, the (theoretical) CPU performance of the current generation
consoles was out of this world when they launched. Their GPUs went
toe-to-toe with discrete cards costing as much as the consoles
themselves.
The (potentially) unexceptional eighth generation
The Southern Islands GPU, shipping in the HD 7970, has been on sale
for three months already. With neither next-generation console likely to
hit the market until 2013
(and probably late 2013 at that), Southern Islands will be the best
part of two years old when those systems finally hit. Southern Islands
is a fast and powerful GPU, but it's already lost the top performance
spot, displaced by NVIDIA's brand new GTX 680. It'll be falling further
behind with the launch next year of AMD's Sea Islands GPU architecture.
If the next-generation Xbox really does use a Radeon HD 6670 part, it'll
be even less impressive.
Estimates of CPU performance are harder to make, given the dearth of
information about these consoles. Being realistic, we can't expect any
great leaps for the CPU either. If AMD could produce processors that
were competitive with or superior to current shipping x86 processors, it
would be doing so. Unfortunately for AMD, its newest Bulldozer architecture
hasn't reached the performance levels the company originally announced.
The next-generation PlayStation CPU could be a Bulldozer derivative, or
it might be based on the company's low-power Bobcat design. In either
case, it's unlikely to boast the kind of remarkable theoretical
performance that the Cell claimed relative to its contemporaries.
Seventh-generation consoles leapfrogged the top-level PC performance
of the time. The systems were enormously powerful, and enormously
expensive to build. Both Microsoft and Sony sold them at a considerable
loss for their first few years on the market. Thanks to these subsidies,
they offered phenomenal value for the gamers' dollar, affording gaming
experiences that would be prohibitively expensive for PC gamers to mimic
at launch. If the current architecture rumors prove to be true,
eighth-generation consoles aren't going to pull off the same feat.
They'll be a substantial step up from current console hardware, sure.
But they likely won't be able to offer the same wow-factor the seventh
generation did.
If Sony and Microsoft have indeed slowed down their console hardware
arms race, building for more modest specifications instead, then this
could be good news for everybody—except perhaps console gamers.
The cutting edge has lost its point
Cutting-edge hardware is expensive to produce. While Microsoft could
probably stomach another round of massively subsidized gaming hardware,
Sony probably can't. Subsidized hardware is a risky proposition. More
modest systems, selling perhaps at break-even at launch, are much more
palatable to shareholders and beancounters alike. Nintendo and Apple
have both demonstrated that selling hardware profitably can be done
successfully. This is certainly the more sustainable model for the
long-term health of the industry.
Cutting-edge hardware is also, arguably, pointless for a new console.
While PC gamers can always slap on a huge 2560×1600 or 2560×1440
monitor—something that taxes even dual high-end video cards these
days—consoles are for the most part limited to the 1920×1080 at 60 Hz
that HDTV sets allow for. 3D sets, which ideally need 120 frame per
second inputs, do raise the bar somewhat, but speccing the GPU for this
niche audience would be a foolhardy endeavor. It would make the GPU more
expensive for 100 percent of customers, with benefits seen only by a
handful.
Contemporary CPUs are already overkill for many games. Developers
have struggled to exploit the large numbers of hardware threads that
processor designs now support. Even a good-looking and moderately
physics-rich game such as
Battlefield 3 rarely demands more
than three cores of a current Intel Sandy Bridge processor. There are
games that can take more advantage of multiple cores, but they're the
exception, not the rule. As long as the CPU is at least
adequate, the GPU is probably the best place to invest money.
With a 1080p60 graphical upper limit and recognition of the
complexities of multithreaded programming, there isn't a compelling case
for building hardware that's streets ahead of what we have today.
Media machines
Keeping the hardware inexpensive is also important for another
reason. Modern consoles aren't just used for games. Xbox LIVE Gold
subscribers spend more hours per month watching streaming TV
than they do playing games. This is a burgeoning market that greatly
expands the appeal of games consoles—console gaming is still a niche
activity; watching TV isn't.
Streaming media has mainstream appeal in a way that games won't
achieve for another decade or two. It's an audience worth going after,
but it changes the economics of console hardware development
substantially. The game consoles can be subsidized and sold at a loss
because each game also includes a cut for Sony or Microsoft. As long as
gamers buy a handful of games, the money can be recouped. Boxes used
predominantly for streaming media don't provide access to that same
revenue stream.
Microsoft does still make money from some streaming media users,
since many services are locked behind its Xbox LIVE Gold paywall. But
with competition from other set-top boxes with comparable streaming
capabilities and no monthly cost, it's not clear if this is sustainable.
To accommodate, selling the hardware can't incur losses—which means it
can't include expensive, high-end components.
Good news for developers
The more conventional system architecture would be good news for
developers. The current Cell architecture in the PlayStation 3 has
proven difficult for developers to make the most of. Its design—a single
PowerPC core with eight simple but fast vector cores (of which six are
usable by third-party developers)—is quirky. The Xbox 360, with its
three identical cores and six hardware threads, and the PC are both
easier to use and understand.
This is not to say that the next generation PlayStation will
necessarily be identical in design to a PC (though that has been tried
before, with the original Xbox). Sony and AMD might have a few custom
tricks up their respective sleeves. AMD's plan
is to produce highly-integrated systems-on-chips, and the company has
said that it's keen to include additional processing units in these
designs. It's easy to envisage a custom-produced design that combines
perhaps 2 or 4 CPU Bulldozer or Bobcat threads and a Southern Islands
GPU—both "standard" AMD parts—with, for example, a high-speed memory
unit, or a dedicated vector processing unit similar to those found in
the Cell processor.
A conventional design means developers can take full advantage of the
hardware much earlier in its lifecycle. As a rule of thumb, games
released later in a console's life look better than those released
earlier. Early in the console's life, developers don't yet know the best
way to wring out every last bit of performance from the system. The
more unusual and complex the architecture, the longer it takes to
understand how best to use it.
While the hardware companies might not like it, developers like
systems that aren't strange outliers. Most major games from major
publishers are not exclusive to any one platform. Huge franchises like
Call of Duty
are cross-platform titles, released for Xbox 360, PlayStation 3, and
PC. As a result, these games tend to be developed based on the lowest
common denominator. A PlayStation 3 might be particularly good at a
particular task (a fancy graphical effect, say), but if the Xbox 360 and
PC aren't equally adept at that same task, cross-platform developers
will have no option but to ignore the PlayStation 3's aptitude, or spend
a lot of development time tuning a version specifically for the system.
This might mean platform exclusives don't have any special
capabilities to take advantage of, but with platform exclusives normally
negotiated according to studio ownership or cash payments—rather than
the nature of the hardware in question—the impact of this is likely
minimal.
An AMD victory
If AMD has scored the GPU design win for the next Xbox, and both the
CPU and GPU designs for the next PlayStation, this is enormously good
news for the company. It will provide a steady stream of income for many
years to come.
It might also help the company undermine NVIDIA's attempts to court
game developers. NVIDIA's "The Way It's Meant To Be Played" promotional
program sees NVIDIA work with developers to some extent to help market
or develop their games. In theory, TWIMTBP games are developed on and
developed (or at least, optimized) for NVIDIA hardware. In practice, the
extent varies; some games are developed on NVIDIA hardware with NVIDIA
offering advice with performance tuning. For others, the branding is
applied only after development has been completed, purely so that
publishers can take advantage of NVIDIA's marketing and promotional
dollars.
At a minimum the games should run reliably on the company's hardware;
it may or may not contain additional tuning to ensure optimal
performance on it.
With both next-generation consoles using Southern Islands, it's
inevitable that games for these consoles will be developed on, and
developed for, AMD GPUs as their first priority. NVIDIA will still have a
role to play, as its GPUs will continue to be found in PCs. But with
consoles taking the lion's share of the market for most games,
optimization for NVIDIA is unlikely to ever rival that for AMD.
What about the gamers?
While bad news for NVIDIA, it's probably worse news still for another
demographic: current PlayStation 3 owners. The radical shift in
architecture, from Cell with NVIDIA graphics to x86 with AMD graphics,
means that the next generation PlayStation is unlikely to offer
backwards compatibility with existing titles (rumors are already
pointing towards Sony removing this feature, in fact). Emulating Cell on
the CPU will be impossible, as the CPU simply won't be fast enough.
Sony could potentially integrate a Cell processor into the new
system. The company did a similar thing with the PlayStation 3; initial
models included the PlayStation 2's Emotion Engine for backward
compatibility. Then Sony dropped the chip as a cost-saving measure in
2007. Adding hardware purely for backwards compatibility is hard to
justify on a cost basis: the older games have limited appeal to new
buyers, and even existing PS3 owners could continue to use their old
hardware. There's an outside chance the GPU could be roped in to allow
Cell emulation, or that a vector co-processor could be integrated into
the CPU. But in all likelihood, the next PlayStation will break from
Sony's backwards compatibility trend.
Console gamers of all kinds may also be disappointed the new machines
won't be as tremendous a leap over current systems as past systems have
been. Consoles have already been eclipsed by PCs—with a result that
games like
Battlefield 3 offer PC players larger maps with more
players than the consoles can cope with—and it looks like that will
still be the case come the eighth generation.
If current rumors are to be believed, the next generation of Sony and
Microsoft consoles will gain performance parity with PCs, but not much
more. Consoles will still have their advantages—the range of
peripherals, the plug-and-play simplicity, the reduced maintenance, the
low up-front cost—but they won't be able to offer best-in-class gaming,
even at their debut. For that, only a PC will do.