At the end of 2006, the average home/office PC is still the same crooked X86 architecture, still considered by many as the worst instruction set ever created, and still running a GUI follow-on to the old DOS, the Windows family of malware "operating systems", if one can consider Windows a true proper OS. The casing is basically a vertical version of the old IBM PC - even the I/O slots, CPU and mainboard are all in similar position all these years.
While IBM itself finally got rid of its PC division, selling it to brand-hungry Beijing, the two other key parties in the original PC venture, Microsoft and Intel, made obnoxious hundreds of billions of $$ in the meantime from it, and cemented their monopoly positions fore PC hardware and software respectively. All other attempts, no matter how technologically superior, failed to make a serious dent - remember the superb 64-bit Alpha CPU family and its infamous murder by Carly&Curly? Or the equally superb OS/2 and its equally lame pillowsmothering by IBM? Not to mention Netware, WordPerfect, SmartSuite, Netscape - in the end, we're still relying on Wintel, just like 25 years ago.
The landscape did change a bit as AMD is now stronger than ever, even though it missed the chance of a lifetime to accelerate its product development and deliver the knockout punch in late 2005/early 2006: if there was a K8L then, Intel would be in real trouble - even Core 2 wouldn't save it. The lesson is: don't rest on your laurels just thinking that Intel would have continued the suicidal P4 push. Alas, it's too late now - I seriously doubt Intel will ever repeat such a mistake again, and AMD will have to satisfy itself by being perennial No 2 by far, even with the DAAMIT buy.
Also, on the OS front, the Mac-Intel deal might just give Apple a chance to push MacOS beyond those meagre few percentage points of its current market share: it seems to run wonderfully on the Core 2, and gives us all a good alternative to the ever more complicated and insecure Windows. As long as Apple doesn't fall into its Ipod-Itunes and start polluting its OS with plenty of DRM like Windows Media Center, not to mention Vista, do - too much DRM can kill (the OS market acceptance, that is).
Also, almost all Linux distros are now available in all-inclusive easy-install packages, both 64-bit and 32-bit that provide all the required functionality - stable GUI OS, office, multimedia, 3-D, games, and more - for free. For the first time, over half of PCs in my house - 5 of them - run Linux either in dual-boot or alone. Doing a credit-card transaction? No Windows, please.
So, what to expect this coming year then?
CPUs - more of the same
Except for the up and coming POWER6 at the very high end, and continuation of ultrathreaded SPARC follow-ons, all the rest of the action in the general-purpose computing (server, workstation, desktop) is still the old irritating X86. I believe that Intel will have its first 45 nm CPUs out before the end of 2007 - to steal the march from AMD's new 2007 cores, at least. Too bad that FSB1333 looks like to stay the highest official bus speed next year. If I were at Intel, I'd have tried to make the best of that process advantage by pushing two Conroes and shared large L3 cache on a single die with single FSB load for a native 4-core chip without the FSB bottleneck that the Kentsfield suffers. Intel can still squeeze at very least a regular 3.33 GHz/FSB 1333 Conroe (if not Kentsfield) part on the 65 nm process - as a guide what the future tweaking can do, look at what overclockers, including myself, achieve with the current parts on air cooling.
Also, I expect to see a very potent answer to AMD's 4x4 - simply, put two Clovertowns together (8 cores total), add a third party (can't mention which one) dual FSB chipset with either quad-channel DDR2 or dual-channel DDR3 in sync with those FSBs, plus 2 x 16 PCI-E SLI with some extra I/O lanes on a side, What you have is a '8x4' setup now - if it wasn't for the chipset, Intel could have shipped it now. But well, in any case, I guess 2008 will bring the real news on the Intel X86 front - Bloomfield and CSI, I am waiting!
AMD's new cores will be the thing in 2007, however, despite the matching instructions-per-cycle problem and FP throughput, I don't believe it will have a per-core per-clock advantage over Intel's offerings - it will probably be a neck-to-neck race. Its dual-channel DDR2-1066 on-chip memory controller and HT 3.0 will be its biggest pluses - it's hard to say no to 17 GB/s of peak memory bandwidth per socket and easy four or eight socket availability, plus an expected 3.2 GHz frequency for the 2007 year-end quad-core AMD offering.
Oh, I forgot Itanium... well let's leave it at that: I won't bother much about the Montvale half-step. If Tukwilla doesn't appear by the end of 2008, in my mind it's bye-bye for the platform - but again, that is beyond the scope for 2007.
Memory - DDR3 time comes
Starting with Intel's new chipset line in Q3, DDR3 will finally make its overdue appearance on the scene, and signal the end of all those senseless 2.4+ volt overclocks on the DDR2 front - over 25% above the stock 1.8v! With DDR3-1333 at 1.5v and DDR3-1600 at 1.6v, plus good latencies (between CL5 to CL7) to boot, the new standard's performance and low power should make it the choice for late 2007, as long as there isn't much price premium over DDR2.
I do believe that we may see Rambus back in the PC market: its XDR memory could be an interesting alternative, not just for GPUs, but for on-chip high-bandwidth memory controllers of Nehalem-generation CPUs. Remember how Alpha 21364 (EV7) so nicely matched the L2 cache bandwidth with 8-channel RDRAM bandwidth to basically use caches as a low-latency buffer for full-speed code&data streaming from the main memory.
System design - back to low-power desktop
With lower-power CPUs from both Intel and AMD, and the switch to DDR3 memory, there will be an opportunity to shave another 30W or so of consumption for a typical dual-core late-2007 desktop PC. Assuming you stick with a mid-range graphics device, say a DX10 equivalent of GeForce 7900GT or Radeon X1900GT, a good 2 GB Vista PC in such config should be more than happy with a 400W PS - in fact, I believe that a 350W PS should be more than fine, unless you have water cooling and RAID in there. So, we'll be back in power supplies where we were three years ago, the pre-Prescott age - good for the environmental causes, certainly.
Graphics - 4K displays (re)appear on PC
As 2560x1600 displays become common and inexpensive (Apple, Dell, HP...) the ultra high-end will come back to where IBM T221 ruled few years back: the ultimate desktop resolution of 3840x2400 QWUXGA. At this level, you can map any lower PC or TV / HDTV resolution directly, without interpolation, onto full or near-full screen, using integer multoplies of pixels. A 30-inch QWUXGA display would provide a nice, sharp but still discernible, 150 dpi resolution, and provide 4K 70 mm cinema-quality experience on your desk.
On the other hand, I expect 22- to 24-inch 1920x1200 screens to break the $500 price barrier next year - not only every desk can then have a true HDTV++ display, but it would also be the end of the 'oddest ever' monitor resolution of 1680x1050. Who the %#&@ even figured out that number? It doesn't match any other (previous or future) computer, TV or HD resolutions, and I want it a quick end.
Input - 3D mice and Spaceballs are back
With Vista's AeroGlass, and 3D enabled coming GUIs on Linux and MacOS, the 3D (full X-Y-Z movement) mice and spaceballs, once the expensive domain of niche workstations, will now have a good reason to go to the mainstream - another good excuse will be PC versions of Nintendo Wii games, whose 3D & motion-aware gamepad needs a PC equivalent.
Software - multithreading galore
No excuses anymore - nearly every new laptop, desktop, workstation and server in 2007 will be multi-core. To win on the performance front, code writers have to not just support, but fully optimise the apps for multithreading - unless the code is so hopelessly serial that it just makes no sense to even try it. The rumours of AMD's Reverse Threading, enabling a single thread to run across two cores, didn't materialise in 2006. For 2007? Let's see...
New superviruses: Virtualisation + AMT = DANGER
Virtualisation in hardware lets you safely & reliably run several different OS environments in parallel, without resource conflicts. AMT-like remote boot / power-on / management capabilities let you easily turn on, run and diagnose your or company PC, no matter what happens to its operating environment (read: infected / corrupted Windows). Now, if you combine virtualisation and remote management, on many 'always connected' PCs, and you are a truly nasty hacker / 'Big Brother' government / David Icke's reptilian or all these in one (the 'historical baggage' 3-in-1 package), you would be tempted to enjoy turning on thousands of such unsuspecting populace's PCs remotely when they're away, and launching parasitic remote OS sessions under virtualisation that could do anything to their systems, including scanning / copying / malware-planting, yet without a trace left in their 'regular' OS sessions.
While this sounded just like a bad nightmare few years ago, now all the hardware needed is in place for this to happen - quick recommendation: unless you ABSOLUTELY NEED IT, do disable virtualisation and remote boot. also make sure your PC isn't connected to the Net when you're not using it.
Final: Vista - the endgame or end for Windows?
Vista was long expected by all the Windows fans and yes, there are such people, yet it seems that, besides the 'eye candy' GUI ,which other OS environments aren't short of either and few productivity enhancements, there isn't much else to bother about. Oh no, there is: even worse resource hogging. Now you need 1GB just to launch M$ Office 2007 - ten years ago you needed just 64 MB for basically the same app functionality, new unknown security concerns and, to top it all, complicated and truly irritating DRM infection, pervading the whole OS. No, I am not favouring pirates and believe that, if every song is sold at 50 pence without major restrictions, everyone should legally pay for what they get - but, inserting rules that completely screw up legal users' experience in an age where everyone will want a seamless HD multimedia experience, will just support piracy.
And more, it may just go too far for many to swallow, and open their eyes on how easy Linux has finally become to use, besides being truly free - couple it with China's EVD disc, offering high definition play without mad royalties and regions, and voila - the ideal HD platform comes up. Fast, stable, cheap and made in China - and no Windows in sight.
Many users are finally seeing through the wool pulled over their eyes - ever-increasing licence fees, ever-higher hardware requirements which stink of collusion with certain hardware vendors, endless security threats and holes coupled with ever-persistent rumours of NSA backdoor access for every Windows copy - why do you think Red China doesn't exactly endorse it, and, after all, there is only so much that 95% of PC users really need to do on their PC. Hard-core gamers, engineers and multimedia creators together are still a very small portion of the total market to justify the endless software bloating - the rest of the users are fine with what non-Windows platforms may provide for free.
The problem is that, with open-source model if Linux becomes strong, it is fairly easy to move the complete code library to another new platform. Let's say Alpha EV9 is revived tomorrow, or POWER6 becomes dirt cheap, or Cell CPU finally has better yields - with all the code in open source, you could have all the software you need ported and tuned on the platform in a matter of days. Not to mention getting rid of all the ugly quirks of the X86 ISA. So, if Microsoft cheeses everyone off, and Windows starts to lose the market share - it could be start of the end of Wintel dominance in the long term.
Let's see what will happen in 2007. I like Vista AeroGlass and its 64-bit capability, even though it doesn't play games, just like 64-bit XP didn't, but everything has its limits - I hope Micro$oft corrects these 3 key irriations in the Vista SP1. Otherwise, my next PCs will follow the current 'newest' one: that one now runs 64-bit OpenSuse 10.2 - only! µ
Getting botter all the time
It's the best of the rest from Google's week
Just like we promised ourselves we wouldn't do again