The time has come for me to put together a new PC. The Core 2 Duo E6700 I bought within a few weeks of its release is now five years old, and though it's served me well, when it comes to playing the latest games and building software in Visual Studio, it's just not that quick. Both the processor and my video card, a Radeon HD 4830, need replacing with something newer and better. The October release of Battlefield 3 has finally pushed me over the edge: I need a new computer to play that game in all its glory.
This is going to be a major upgrade. I'm going to keep my hard disk, beloved Dell clicky keyboard, and twin Dell U2410 monitors, but everything else is going to go; motherboard, CPU, RAM, video card, TV tuner, optical drive, and case are all up for replacement. Optical drive, too; I have enough old games on CD and DVD that ditching optical media isn't really viable yet.
Many of the component choices are easy enough. But when it comes to processors and motherboards, I'm flummoxed. Intel's Sandy Bridge processors have the clear performance lead, so they're the natural choice for any system build. But then I fall foul of Intel's pricing scheme; which one should I buy?
It should in theory be simple—get the most expensive processor in a given price bracket/model range, and it'll be the best. That means getting a Core i7 2600 variant.
Intel has three Core i7 2600 variants. There's the standard mainstream/business-oriented 2600, the low-power 2600 S, and the enthusiast-oriented 2600 K. The 2600 is the cheapest, the 2600 K the most expensive, with the 2600 S somewhere in between. The 2600 S is actually something of an odd-ball; it's equivalent in terms of features and specification to the 2600, but its power usage is lower (65W versus 95W), and its base clock speed is lower (2.8GHz versus 3.4GHz), even though its turbo speed is the same. For my purposes, only the 2600 and 2600 K are worth consideration; I want the higher clock speed and don't care about the difference in power usage. So given the choice between the 2600 and the 2600 K, the more expensive 2600 K should be better, shouldn't it?
Nothing's ever easy
Unfortunately, it's not that simple. The Core i7 2600 K may cost more, but that doesn't mean you actually get more. The cheaper Core i7 2600 has a few specialized features not found in its more expensive brother. Specifically, a virtualization feature called VT-d, and a security feature called TXT. Neither of these are mainstream, widely used technologies—at least, not yet—but both have some interesting applications.
VT-d opens up the possibility of safely giving virtual machines direct access to hardware such as video cards, allowing the flexibility that virtualization affords to be combined with the performance of running an operating system on the bare metal.
Some virtualization software such as Xen can already make use of VT-d, when supported, to provide this kind of assignment of devices into virtual machines. Over the five-year life that I expect the new computer to last, I suspect that all virtualization software will support it. One day this technology will be everywhere, just as CPU virtualization extensions like Intel's VT-x and AMD's AMD-V already are, and so software using it will be widespread. Imagine, for example, a future version of Windows that can run legacy software in a virtual machine including 3D games or CAD software, with no performance penalty. It obviously enables progress.
When that software support arrives, I want my PC to be able to take advantage of it. But I can't do that if my processor doesn't support VT-d.
TXT is rather more esoteric, and rather more contentious, as it interacts with Trusted Platform Modules (TPM), and could be used to provide robust DRM protection. With TXT in the processor and TPM on the motherboard, the processor can carve out protected pieces of memory that are off-limits to the rest of the system; these could then manipulate DRM keys safely without fear of compromise. That's something that some circles will undoubtedly hate, but there are other things that can be done with the technology too.
The power to make the world a better place
Most significantly, TPM and TXT together can ensure trusted boot. That is, they can ensure that the system's boot files haven't been tampered with before letting the system boot. This has the potential to make life a whole lot harder for rootkit authors. Some of the widespread Windows rootkits today overwrite the Windows boot loader so that they can tamper with the operating system before it even loads, to disable certain security features and make themselves even harder to detect.
TPM can stop that kind of tampering dead in its tracks, and the world would certainly be better off if such rootkits didn't exist, but at the moment it's a safe gamble for rootkit authors to try to pull this kind of stunt, because most systems don't use a TPM in this way. TXT interacts with TPM to provide even stronger protection; TPM checks the system's integrity at boot, whereas TXT can do it at any time.
If these technologies were widespread, and if malware authors had to assume that tampering with the system would be detected, we'd be much better off. It would make things more difficult for the bad guys, almost providing a kind of herd immunity. But, alas, they're not.
I want both these features, but I can't get them with the Core i7 2600K. I have to go for the Core i7 2600 instead. Doing that, I save a few currency units and gain a few features, but there is, of course, a downside. I lose the ability to overclock—though personally, I don't care about that very much—and I get worse integrated graphics. That's more annoying. The integrated graphics in the Sandy Bridge processors come in three variants; the top end HD Graphics 3000, found in the Core i7 2600K, i5 2500K, and most of the mobile processors, the mid-range HD Graphics 2000, found in all the other Core-branded parts, and the low-end HD Graphics, found in the Pentium-branded parts.
This is peculiar, to say the least. The overclockable "K" series parts are the most likely to be used in conjunction with discrete GPUs. It's the mainstream parts—the ones that have TXT and VT-d—that are more likely to be used in business desktops, and hence more likely to require the use of integrated graphics. But for some reason, Intel has decided to put the good GPUs in the hands of people who will never use them, and force everyone else to make do with the slower parts.
I don't plan to use the integrated graphics for the most part, though I'm glad to have it there as a fallback, so perhaps the HD Graphics 2000 is OK. But I would like to use Intel's high-performance video transcoding, Quick Sync Video, to transcode videos for playback on a smartphone, and I'm told that that's faster on the HD Graphics 3000. Although Quick Sync mostly use dedicated hardware within the GPU to do its thing, some parts of the encode process are supposedly performed using the regular GPU hardware. HD Graphics 3000 has twice as many execution units as HD Graphics 2000, and it runs them at a higher clock speed; this in turn apparently gives it better transcode performance—though I am not sure how great this impact actually is.
I don't want either/or; I want both. I'm even willing to pay...
So essentially, neither processor really gives me what I want. 2600K isn't an upgrade from 2600, even though it costs more; it gains in some areas, but loses in others. Except for the ability to overclock, neither the gains nor the losses make a whole lot of sense—there's no reason to take away TXT or VT-d, nor much reason to include the improved graphics.
I understand the argument for offering different features at different price grades. Companies use minor features differences to distinguish between the high-end and the low-end, and alter pricing accordingly. This allows them to extract more money from buyers who place a higher value on their products, while still being able to offer similar products to buyers with less money to spend. I don't have a problem with this; I don't think making a profit is itself objectionable, and as long as everybody feels they're paying a fair price for the product, who's losing out?
But when the pricing tiers don't correspond with capabilities—pay more and get less, pay less and get more—and when the features aren't things that people even know about, it's hard to make sense of the price scheme. You can make people pay extra for more processor cores or more cache or more clockspeed, because they know about those, and can see the benefit. But that's not really true of esoteric processor features. Few people know about them, much less have any willingness to pay extra for them; unless you check out the minutiae of your processor on Ark you'd never know the difference.
The processor I want does exist, however
The real kick in the teeth is that Intel does have a processor that offers the features of the 2600 and the graphics of the 2600K. It's the Xeon E3-1275. For another $22, you get everything. Maybe more than everything; the graphics part is named "P3000"; Intel makes vague claims that these parts have superior performance in things like CAD software, though it's reluctant to provide any specific details. So there's certainly no technical impediment standing in the way of making a processor with the good graphics and the full range of processor features.
But the problem with the Xeon is that if you want to be sure they'll work, you need a motherboard with a Xeon chipset. I want to use the Z68 chipset because I want to take advantage of "Smart Response Technology"—a new feature that lets you use a (small) SSD as a cache for a (big) magnetic disk. Though there's a very close Xeon counterpart to Z68, named C206, its major difference is that it doesn't include Smart Response Technology. This is a rather odd decision. C206 is aimed squarely at workstation users, and they can surely benefit from the accelerated disk performance that SSD caching provides. It's just the feature is somehow disabled.
Technically, the Xeon and Core parts should be interchangeable. C206 supports the non-Xeon processors, and there's no reason for Z68 not to support the Xeons. The sockets are identical, the processors are as closely related as possible without being the same. There just aren't any guarantees. Using a Xeon on a Z68 motherboard would probably require firmware support, to ensure the processor is properly recognized, and I don't think any vendors offer such a thing. So the CPU I want is there, it's just tantalizingly out of reach.
We all lose out
Restricting the availability of processor features means simply that it's harder for software developers and system builders to take advantage of them. As long as processor features remain the same, a cheap low-end dual-core processor can run every single program that an expensive high-end quad-core hyperthreaded processor can run; it just does so more slowly. But when different processors have different features, developers now have to pick and choose which features they depend on, knowing that they eliminate some of their potential audience when they pick features that aren't universal.
Intel positions TXT, in particular, as a purely business-focused part—it's an integral component of the company's vPro platform—so no doubt feels that it's irrelevant to the enthusiast (that is, gamer)-oriented K series. This is true, in a sense—the only systems likely to make use of TXT and TPM are business systems—but that's a chicken and egg scenario. If TPM were in every Intel chipset and TXT in every Intel processor, of course system and software vendors would start using them.
This isn't new, of course. Intel has done this kind of peculiar segmentation before. Its VT-x virtualization feature was found on some, but not all, Pentium 4 and Core 2 Duo models. Generally, high-end ones had it, low-end ones didn't. This whole issue became rather messy when Microsoft wanted to make its Windows XP Mode virtualization feature depend on VT-x: Windows XP Mode would work on some systems but not others, with no easy way for end-users to tell if their processor would have the required features turned on or not. Microsoft eventually relented, dropping the VT-x requirement, but this wasn't progress. It was increased software complexity, decreased performance, all because Intel decided that some processors should have the feature disabled.
Treating processor capabilities as product differentiators just doesn't work very well—the VT-x debacle shows that much, the market just doesn't understand the distinctions—and having a model line-up that prevents even the possibility of an upsell—want a processor that's like a 2600 but better? Too bad—makes Intel plenty of money, but it hurts the broader PC ecosystem. We have software that doesn't take advantage of the latest and greatest capabilities, and we have systems that are less secure, because Intel thinks that people will pay more for processor features—and then denies them the ability to do so.
Listing image by Photo by David Sanders, illustration by Dominique Signoret
reader comments
188