As a number of people have eloquently
pointed out, a single swallow doth not a summer make (indeed, all of them
struggled to manage it in the UK this year). In this case, that means: while
Broadwell might well be delivered on BGA, it might also come on a socket. This
would make some sense, as the market for workstation-class hardware is still
reasonably strong and has been remarkably profitable for Intel.
![The 27” Apple iMac comes in only two processor specification, with three GPU possibilities - only its RAM is defined as a user upgradeable](http://programming4.us/image/012013/27/92741348.jpg)
The
27” Apple iMac comes in only two processor specification, with three GPU
possibilities - only its RAM is defined as a user upgradeable
Intel has done well from selling people a
cheap Celeron and then upgrading them to something better later, as it enables
them to spread the full cost of ownership over time. The lack of that upgrade
conveyer belt is difficult to quantify in Intel’s sales planning, but it must
have some figures tracking how many processors an average enthusiast might buy
for a single system. It also may have quantified what second hand CPU sales
cost it through the likes of Ebay. It’s complicated stuff, but you’d reckon
that someone is doing the math.
Whatever the numbers say, this isn’t
something that Intel will want to jump into with both feet. As such, it would
make more sense to see a transitional proposition where a new chip is both BGA
and socketed, to see if ending the latter tradition is a viable option. I would
think it’s safe to say that Intel is definitely considering this ambitious move
- though whether it’s written is stone (after crunching plenty of numbers on
how it might affect sales and upset the private system builders) is doubtful.
It may well be that, eventually, it concludes that it can leave the
enthusiast/upgrader business to AMD, without damaging its overall
profitability, as radical as that may be.
If anything resembling the plan we’ve
outlined comes to pass, it would mark the most radical change of direction that
Intel has ever made; serving to spotlight the seriousness with which the
decline of the desktop market is viewed internally.
“Intel has done well from selling people a
cheap Celeron and then upgrading them to something better later, as it enables
them to spread the full cost of ownership over time”
Is It Totally A Bad Idea?
When I first heard this my reaction was
probably the same as many of you reading this now: surely the strength of the
PC is its flexibility and upgradability, and ending socketed CPUs is the thin
end of a very big wedge? The truth, actually, is that we should have been
expecting this because as the number of pins that processors need goes up, it
becomes less practical to have a socket to support them.
![In a BGA chip the contacts can be placed much closer together (as you can see on this mobile Core-i7) because the robot that will solder them to the board can work within very fine tolerances](http://programming4.us/image/012013/27/92742348.jpg)
In
a BGA chip the contacts can be placed much closer together (as you can see on
this mobile Core-i7) because the robot that will solder them to the board can
work within very fine tolerances
I have an LGA 2011 test rig, and the CPU on
that machine is almost the size of a drinks coaster. The silicon part of that
package is much smaller than the whole CPU, the size of which is dictated by
the 2011 little dimples it needs to connect to the PC. Each of those little
pins in the socket is a potential point of failure, and at some point the
likelihood of getting a system to work reliably is going to be impaired by the
statistical certainty that one of the pins won’t be touching the chip fully.
There’s also the fact that a socketed chip must have all the lines that connect
it on one side of the motherboard, while BGA chips can be suspended so that
lines from both sides of the board can be accessed. It all amounts to a huge
issue surrounding board complexity, layers and the number of tracks that can be
practically supported.
Simply put: doubling LGA to 4022 pins could
easily render motherboards too expensive to be commercially viable. Also, a BGA
mounted chip is always going to be more reliable, and if it sits in a hole in
the motherboard it can be cooled from below and above.
However, it’s worth considering that doing
that only puts off the complexity problem with for a few years, because even
using BGA there’s a limit to how many connections a chip can reasonably make to
a motherboard.
To this writer’s mind I can foresee only
two possibilities for CPUs a decade from now. One is that very high speed
serial linkage is developed so, like USB and SATA, the data flows in and out of
the processor using a very small number of wires that are manipulated at high
speed. Another is that an optical interconnect is developed, where laser light
is channelled into and out of the chip in a single piece of fibre, where many
thousands of channels in both directions can be defined, or even dynamically
allocated.
“It’s a form of commercial cannibalism,
where Intel eats one of its many markets in order to ensure the profitable
development of another”
Ironically, if a simple optical connection
was established the chip would only require that and power, making socketed
chips again both practical and possibly desirable.
Rather than focusing on our own reaction to
this, it’s worth assuming Intel’s position, where it is trying to sell
processors in an increasingly competitive market. BGA packaging is cheaper, and
allows it to tie motherboard makers into the chip supply channel. It also
enables it to control major customers like Apple more succinctly, and
effectively turns the whole PC market into a bigger version of the mobile/tablet
sector.
Ultimately, it means it doesn’t need to
support an end-user chip channel, or cope with processor warranties, or people
re-badging its products and selling them as something better than what left the
factory.
The Downside
What I immediately thought when I first
heard about this roadmap was that all computers would be likely to degenerate
into those systems where everything’s melded to a screen, iMac style. Almost
without exception these aren’t good computers, and they usually best represent
a victory of style over functionality.
If you use a computer just to do word
processing or surf then you probably don’t care, but for those people who use
theirs for so much more they’re a retrograde step that’s a modern analogy of
the PET computer or the Sharp MZ series. Is that what we want?
Nor is it an environmentally-sensitive
template. Currently, if your motherboard develops a fault you can take your
CPU, memory, etc. and place it on a new motherboard. In a socket-less future,
if either your CPU or the motherboard has a problem then they’re both toast.
![The Microsoft Surface, poster-boy for the one-size-fits-all future where processors aren’t swapable](http://programming4.us/image/012013/27/92743348.jpg)
The
Microsoft Surface, poster-boy for the one-size-fits-all future where processors
aren’t swapable
Enthusiasts might not be so keen, and
motherboard makers might be wondering if it is a change that’s in their best
interests too, as it provides and additional challenge for those making these
systems: once a CPU is attached to a board, it can’t be pulled off should the
market make a sharp turn and demand for that product dry up. That in turn makes
new product launches tricky, because a manufacturer must decide well in advance
how many of each it is likely to sell. Getting it wrong will result in lost
sales and unsold stock, and a company being lumbered with CPUs (bought from
Intel) that can’t be returned as ‘unsold’.
This represents a massive movement of risk
because, while Intel currently carries the can if a CPU isn’t a rip-roaring
success, after the shift motherboard makers will get to share that pain too. I
can see that Intel might find that attractive on one level, but it’s also a
motherboard maker so it’s not all good news.
It seems doubtful that motherboard makers
of the future will want to carry the full range of Intel options; they make
enough different products already. Multiplying current lines by all the CPU
choices would surely be a logistical nightmare. A typical online vendor I
picked on for the purposes of this article currently sells about 23 Intel
processors on LGA 1155 (from the Celeron G465 to the Core-i7 2700K), while
Gigabyte (again, for example) lists 196 motherboards for that socket, over 11
different chipsets! You do the math.
The forcing of motherboard and CPU
combinations would, we suspect, drastically reduce the number of choices. If it
didn’t, the likes of Gigabyte and Asus would end up listing thousands of
products, made with short (and therefore expensive) production runs. A good
example of this is Apple’s iMac, where you can have two screen sizes and two
processor specs on each, and that’s it. What’s more, that decision assumes customers
want to spend a minimum of £1,099 on a system - not a price that many people
would accept for a Core i5 with a GeForce GT 640M GPU.
If this isn’t enough to convince the
shrinking number of motherboard makers to find some other business, then what
would it take, we wonder?