Ever since the Xbox One and
PS4 were unveiled as being based on 28nm technology, it’s been clear
that both Microsoft and Sony would transition to smaller nodes as soon
as those technologies were ready for prime time. It now looks as though
Microsoft may make this jump before its rival. A sharp-eyed member of
the Beyond3D forum
picked up on the LinkedIn profile of one Daniel McConnell, an SoC
designer at AMD, which discusses how he, “Successfully planned and
executed the first APU for Microsoft’s Xbox One game console in 28nm and
a cost-reduced derivative in 20nm technology.”
So what could it
mean for Microsoft’s console to make a quick jump down to 20nm? In
theory, such a shift could give Microsoft several advantages. Power
consumption and die size both come down modestly at 20nm, which improves
yields and price structure. The system chassis could theoretically be
redesigned with new features or MS could target a cheaper model with
certain capabilities removed. In the past, both Microsoft and Sony have
used these periodic revisions to target modest improvements to storage capacity, performance, peripheral hookups, or other similar options.
The current Xbox One SoC – a 28nm APU made by AMD
But I think there’s an interesting question here, that strikes at the concept of what a console actually is
and what Microsoft can or can’t get away with doing. For decades,
consoles have been presented as fixed units. Sure, the external shell
and some of the I/O hookups might change over time, but an Xbox 360
purchased in 2005 should be just as fast as an Xbox 360 bought in 2014.
Consoles have been inviolate, even as PC hardware shifts fairly rapidly.
As
Microsoft looks at its plans for a 20nm die shrink, it has to be asking
if there’s a way to change the Xbox One’s design to better match the
PS4. And it probably can — as we’ve previously discussed, there are two
unused GPU partitions on the system that would give it a significant GPU
performance boost. It might also be able to increase the size of the
shared ESRAM cache.
The problem, though, is optics. Early adopters of the Xbox One can’t be particularly pleased that they bought a now-useless doorstop
in the form of Kinect 2.0. No one is going to be happy if Microsoft
rolls out a new version of the Xbox One (Xbox 1.5?). And the company
would risk bifurcating its dev teams between haves and have-nots. For a
company whose message has already been badly diluted and confused, the
presence of two different types of console would be problematic at best.
Furthermore, McConnell’s profile makes it clear that Microsoft chose to
drive engagements that would minimize costs over increasing performance in future iterations of the SoC.
We’ve toyed with the idea of upgradeable consoles in other posts, but
after the problems the Xbox One has had with brand and message, I think
shifting the system’s design now would only confuse and frustrate users
further. 20nm die shrinks are also expected to be relatively modest,
with the following 14/16nm technology delivering more raw performance and the bulk of improved thermals.
What about DDR4?
One
intriguing option Microsoft might take would be to outfit the console
with DDR4. There’s been a great deal of discussion over whether or not
the Xbox One’s quad-channel DDR3 memory bus is a problem for most games,
with some general consensus that it likely is. Microsoft could close
this gap, at least in theory, by upgrading to a faster form of DDR4. 8GB
of DDR4-2700 or even DDR4-3200 would improve memory bandwidth by
25-50%. The question, however, is whether games could be programmed to
run equally smoothly on both sets of hardware.
Again, that’s
problematic. If memory bandwidth is really the issue, Microsoft might
wind up having one console that can run 1080p safely, while the other is
stuck in 900p territory. Is it possible to build a resolution toggle
into games? PCs certainly have done it for years, but it’s not clear if
this is possible on the console side.
Microsoft could
theoretically swap out the DDR3 on the Xbox One for DDR4 without
changing any of the underlying specs, but I suspect this will be
difficult. DDR3 and DDR4 have very different latencies and matching the
two well enough to make them identical at the design level might be more
trouble than it’s worth.
Source: http://www.extremetech.com
No comments:
Post a Comment