Hurdles on the upgrade path

Originally published 2004 in Atomic: Maximum Power Computing

Last modified 03-Dec-2011.

You're supposed to be able to upgrade PCs. That's one of the big advantages of an open architecture; standardised, interchangeable components.

Now, we all know that's not true all the time. The brand-name proprietary PCs that so many regular people buy are often hard enough to repair, much less upgrade; it can be disturbingly difficult to even add more RAM and a bigger hard drive, let alone swap out the motherboard.

And, at the other end of the nerd spectrum, really cutting-edge machines can't often be turned into another really cutting-edge machine a year later, without a complete everything-but-the-case brain transplant.

But, by and large, worthwhile upgrades are often possible. New CPU, more RAM, better video card, another hard drive; ten minutes' work and you've pretty much got a brand new PC.

Every now and then, though, there's a big nasty discontinuity. Stuff that doesn't work with any existing stuff.

This whole BTX/Socket T thing, for instance. Six months ago, it was to upgraders as gold is to Cybermen. New-fashioned machines invariably sell quite well in the minimally-informed-consumer market, though, and the result was months and months - centuries, in performance PC terms - of pointless products for us enthusiasts who like to spend our money on, you know, something.

BTX is supposed to make PCs run cooler and quieter, thanks primarily to a new giant bolt-mount heat sink standard. BTX doesn't have anything else to commend it over ATX, microATX and FlexATX. But Intel love BTX. They want it to be the industry standard. Next week, by preference.

The basic BTX case form factor is not just ATX with the mobo on the other side and the PSU at the bottom instead of the top. BTX also tells motherboard makers how to lay out their boards - not only the slots, but the CPU and the RAM too. This is essential to accommodate that vast heat sink (which is no bad thing, in itself) and integrate cooling ducts into the case (another good idea; increasing CPU wattage caused ATX PCs to outgrow their original simple single throughflow fan design many years ago), but it's a big problem for 64 bit AMD systems.

The AMD64 processors have the memory controller on the CPU, so they need the circuit traces between the CPU and the memory modules to be short, and similar in length. Ain't gonna happen, with BTX. If nobody ever solves this problem, and if Intel stick to their guns, then case makers who don't want to give up on half of the market are going to have to split their product lines and factory tooling. They are not happy about this prospect.

It's possible that case makers will follow Lian Li's lead and start making "upside down" ATX cases that can share a lot of parts with a BTX sibling, but it still won't be as clean a transition as the AT-to-ATX one.

Which took four years.

And then there's Socket T, a.k.a. LGA775, which is what you now have to use if you want to keep on the cutting edge of Intel's desktop product line.

Compatibility-breaking socket changes aren't anything new. The P4 started out on the "Up Yours, Early Adopters!" Socket 423, then moved to Socket 478 about eight minutes later. AMD should be commended for sticking with Socket A for so long, but now people trying to make sense of the AMD64 lineup have to cope with 940, 939 and 754-pin sockets, all at once. AMD want nothing but Semprons to be using the 754 pin socket, but they haven't managed that yet.

But Intel wanted to stop making CPUs with pins on them, partly because customers bend the pins, and then send the chip back for a warranty replacement. So they switched to the radically new LGA775, with the pins on the socket and little "cups" on the chip. People are still going to bend the pins, of course, and thereby bork the whole motherboard, but at least that probably won't be Intel's problem! Hurrah!

Socket 775 systems were initially reported to be unlikely to survive more than 20 CPU insertions, but that wouldn't have been a big deal even if it were true. Early P4 systems had some scary issues too, like the fact that the CPU coolers pushed down so hard that they bent the whole board like a banana - but they worked.

The problem was the big upgrade discontinuity. LGA775 was made to be paired with PCI Express (PCIe), which is a Good Thing generally (one bus for all kinds of cards including video, with enough bandwidth to handle things like gigabit Ethernet without requiring Special Server Slots), but which explicitly does not support AGP. And as if that weren't enough, the top-end LGA775 systems required DDR2 memory, which at the time was actually slower than regular DDR, despite costing more.

PCIe doesn't offer any significant extra speed for current graphics hardware. The "x16" PCIe graphics card slots are twice as fast as AGP 8X, but AGP bus speed bumps down the years have been underwhelming, and so is the x16 step. PCIe for video, in itself, is no real upgrade at all. There were, fortunately, at least some regular PCI slots on the new boards in addition to the tiny "x1" PCIe slots for cards that someone might, one day, buy, so you didn't have to trade in every card. Just the most expensive one.

And the new LGA775 processors cost more than the existing Socket 478 ones, but weren't any faster.

So these early LGA775 systems were, understandably, received with a rousing round of indifference.

There were some fits and starts over the following months to try to get sensible people to buy LGA775, and dropping prices for DDR2 RAM and Socket T processors helped. Only now, though, are there actual realistic LGA775 options - and they only exist because of an old trouper of a chipset.

Now, at last, you can buy a motherboard that takes an LGA775 CPU and accepts your existing PCI and AGP cards, and regular DDR RAM. Quite a few of them exist; this is a good example. All of them have one thing in common - the i865PE chipset, which dates back to June, 2003.

Amazingly enough, though, these old-chipped boards aren't actually significantly slower than the latest shiny stuff, all other things (CPU and graphics card) being equal. And all other things are equal, by and large.

It's now, finally, possible to get DDR2 modules that beat DDR1, but RAM speed has little impact on PC performance for most tasks (thank goodness).

You can also get slightly faster LGA775 chips than you can Socket 478, but nobody sensible's shelling out for the 3.8GHz LGA775 P4 "570" or the half-the-price-but-still-far-too-expensive 3.6GHz LGA775 "560"; the 3.0-to-3.4GHz chips are far better value, and they're matched by (now more expensive) Socket 478 alternatives. If you're in love with the good old Northwood core then you're out of luck, because all LGA775 P4s use the slower-at-a-given-clock-speed-yet-hotter Prescott. But on the plus side, the Prescott Celerons are, unexpectedly, great performers for the money; much better at a given clock speed than the old Northwood ones.

PCIe still doesn't have a graphics advantage, either. The Radeon X850 is only available in PCIe, but even the top-clocked X850 variants just keep on mixing it with the similarly expensive GeForce 6800 Ultras; there's no very pressing reason for even a rabid gamer to buy one. And everything else is still available, possibly with minor model number differences, in AGP and PCIe flavours.

All PCIe graphics has over AGP, so far, is the ability to run a pair of Nvidia cards in Scan Line Interlea... I'm sorry, Scalable Link Interface mode. That really does give a significant speed boost over what any single card can manage at the moment. But for everybody who doesn't have a monitor that RAND Corporation stole from the future, this just means running your games five times faster than you need to, instead of only 2.8 times.

Even when Socket T was new, people bought it, just so they could get a big fat too-hot Prescott that ran a whole 6% faster than the fastest Socket 478 chip. Heck, some even bought the very first server-chipset BTX systems with two PCIe video slots, so they could be the first on their block with new-fashioned SLI, in a computer that cost as much as a good used car and would be worth as much as a very bad one three years later.

Now, though, the magnificent bleeding edge gesture of LGA775 has sunk back down into the bubbling muck of Stuff Sane People Buy. Against all reason, the slowing pace of CPU development means that people who built a Tiny God a couple of years ago still have little reason to upgrade anything but their video card (well, maybe another drive or two), but i865PE-based LGA775 boards have knocked down the wall across the upgrade path. And there is much rejoicing.

Early BTX adoption, however, still offers users a whole bunch of nothin'. In that department, there has still never been a better time to wait for a standard to bounce rudely off the real world and turn into something people can actually use.

Hurdles on the upgrade path

Other columns