Intel's upgradable processor: good sense or utter catastrophe?


Intel is about to experiment with a new concept in mass-market processors with its forthcoming Pentium G6951 CPU: upgradability. The chips will be upgradable by end users via a purchased code that is punched in to a special program. Previews of the processor quietly hit the Web last month, and with Engadget's post of the retail packaging, people took notice with reactions ranging from surprise to outright disgust.

The Pentium G6951 is a low-end processor. Dual core, 2.8GHz, 3 MB cache, and expected to be around $90 each when bought in bulk—identical to the already-shipping Pentium G6950. The special part is the software unlock. Buy an unlock code for around $50, run the software downloaded from Intel's site, and your processor will get two new features: hyperthreading will be enabled, and another 1 MB of cache will be unlocked, giving the chip a specification just short of Intel's lowest Core i3-branded processor, the 2.93 GHz Core i3-530. Once unlocked, the G6951 becomes a G6952.

The processors will ship as part of a pilot program next quarter to a select number of resellers in just four markets: US, Canada, the Netherlands, and Spain. Systems with the G6951 will have to use one of two specific Intel motherboards (DH55TC and DH55PJ), as well as an up-to-date BIOS. The process also needs an Internet connection: before authorizing the upgrade, the unlock code and ID of the OEM need to be validated by Intel. One OEM that will participate in the scheme is Gateway; the upgrade package that Engadget posted was for that company's forthcoming SX2841-09e desktop that will be available from Best Buy.

Reactions to the scheme have varied, but a common criticism has been that it's unfair to charge customers just to unlock things that a processor can already do. As Cory Doctorow put it, these are crippled processors that you have to pay extra to unlock. If Intel is shipping G6950-class processors that actually have 4 MB cache and hyperthreading support, the company should just unlock the features for everyone. Some have even suggested that the upgrade mechanism will be hacked, to allow people to upgrade for free.

There's an upside

For end-users, the processor offers a potentially attractive upgrade option. The systems built around these parts will be low-end, in the ballpark of $500 or less. A $50 upgrade—that doesn't even require you to open the case—that can offer a performance boost of 25 percent or more (depending on how much benefit the user's workload can gain from hyperthreading) is an attractive proposition.

For system-builders too, there are attractive features: the reason that the OEM's ID is validated with Intel is because OEMs will receive kickbacks for any systems that get upgraded in this way through a revenue-sharing system. Intel also argues that the scheme will allow OEMs to simplify and consolidate their product lines—instead of having to offer a bunch of similar systems with different processors, manufacturers will be able to offer fewer different processor options, relying on end-user upgrades to reinstate the variation in specifications.

The cost of a G6951 processor plus its upgrade is more than the cost of a Core i3-530 ($113 in bulk), so in that sense is poor value for someone who knows they want that level of performance to start out with, but for its target market that is beside the point. A Core i3-based system will cost more up-front, and that's not what buyers of these systems are after. People aren't going to buy the upgrade as soon as they get their shiny new computer home from the store. Rather, it'll be a way of extending the life of an existing machine, and on those terms, the pricing is something of a bargain.

You might not like it, but it ain't new

Whether buyers will resent unlocking capabilities that their systems already have is harder to say. Intel's move is not without precedent. In fact, this kind of unlocking has been a feature of the computer market for decades. Many mainframe systems were sold with what was, in effect, a throttle to limit their performance. Upgrading the performance would be as simple as opening up the case and flicking a switch or cutting a wire.

More recently, Asus' P5P800 motherboard offered a similar kind of upgrade for its integrated audio. Install the right software, and new features become available. And Apple charged $ for software to unlock the 802.11n features of many of its systems.

Moving from hardware into the software world, this kind of thing is routine. There are a million pieces of shareware that disable features until a license key is entered. A lot of enterprise-grade software has multiple pricing tiers with the more expensive versions supporting, say, more concurrent users or larger datasets, even though the software is fundamentally the same across all versions. More controversially, there's the decision of companies like EA to charge secondhand buyers for multiplayer access, resulting in the situation where the same licensed copy of the game either does or doesn't support multiplayer depending on who's playing it.

Though these decisions have been met with grumbles from the buying public—especially in the case of game unlocks—they've all been broadly accepted by that same public. People might resent paying extra for capabilities that, in a sense, they already "have," but when push comes to shove, they'll hand over the cash anyway.

Market realities of the processor business

A case can also be made that far from being a repellent money-grab, this is in fact an interesting way of reacting to realities of the microprocessor market. Intel sells many different processors, with a huge range of cache sizes, speed grades, and core counts; to that mix, there are also a number of features such as TXT and hyperthreading that are only found on certain parts. But Intel doesn't design a specific processor for each particular mix of cache size, speed, number of cores, and extra features.

Instead, the company designs a few processors that can do everything ("real" variations include core count, presence of QPI connections, number of memory channels, and a few other things), and then selectively disables features. Sometimes the decision is made for Intel—a chip might have a manufacturing defect that limits the amount of cache it can use, and not all chips can run at the same frequency within a given power envelope—but a lot of the time, the company is disabling functional hardware. For example, every Pentium G6950 processor has the hardware to do hyperthreading. It's just that it's been permanently disabled at the factory, because Intel's bean-counters have decided that that particular grade of processor won't have hyperthreading.

By offering software unlocks, users can reclaim these disabled features at a later date, if they decide that they'd be worthwhile. Given the choice between permanent disablement and unlockable upgrades, the latter is clearly the better option.

Intel is not alone in this, of course. Those mainframe manufacturers did the same, for the same reason as Intel: it doesn't make sense to manufacture different systems for each price point. AMD does it—those Phenom X3 processors are built with four cores, and sometimes can even have the fourth core re-enabled—NVIDIA does it—the GTX 470 is a GTX 480 with 32 shader cores disabled. Intel certainly takes the differentiation further than its competitors, but its overall approach is unexceptional.

Profit-making companies want to make a profit

If all Intel processors can support all the different features, some have argued that all Intel's processors should have these things unlocked. While that would obviously have some appeal—who wouldn't want the specs of a $1,000 CPU for $100—it ignores market realities. Remember, Intel wants to make a profit. Sure, the company would certainly be able to enable all the features on all its processors, but if it did so, it would no longer be able to charge many hundreds of dollars for the high-end parts.

It might still be able to differentiate according to clock speed, but there are limits even there; once Intel's fabs have got into the swing of making a new processor, it's fair to say that a large proportion of them will run at the top specified speed grade, if Intel wanted them to. Enabling the full range of a processor's capabilities would leave the company with just a handful of different models.

This leaves the company—the company that has an obligation to maximize its shareholder returns—with two options. It could drastically cut the range of models and increase the price of each processor to some kind of average, or it can artificially introduce variety by disabling features or running some parts at below their maximum possible frequency. If it did the former, a lot of people that are today gladly buying systems built around parts costing $90 or less would find themselves priced out of the market. Sure, it'd be great for people picking up a Core i7-980X for a few hundred bucks instead of its current $999, but it'd be lousy for those at the other end of the market. The "third" option—sell full-spec processors for bargain basement prices—isn't realistic given the desire for profit.

Which is why it's not a bad thing that Intel doesn't do that, and it's why selective disabling of processor features is here to stay. Some may be offended at, in some sense or other, not getting all the power their processor has to offer, but it was always like this. The new processors don't change that basic fact.

In the light of this, should we not be welcoming these new upgradable processors? They allow CPU manufacturers to address market realities, while also providing cheap, easy-to-use, waste-free upgrades to those who want them. Surely this gives consumers more options and better value.

Or to use Cory Doctorow's terminology, are we not now being given the choice between "crippled processors" that we can't pay extra to unlock, and a new generation of "crippled processors" that we can pay to unlock? It's hard to see how this is anything other than a step forward.

0 comments:

Post a Comment