Once in a while it's the job of every technology journalist to gobble up some humble pie, roll back the cynicism and give credit where it's due. With the arrival of Intel's latest multi-core monster, the Core i7 processor, that time has come for me.
I don't mind admitting I doubted the first Core i7 chips would amount to much more than a mediocre upgrade over the Core 2 family. For sure, I was well aware it represented a major break from Intel's previous architectures. I knew that Intel was finally going to pension off the ghastly old front side bus interconnect in favour of the bang up-to-date Quick Path Interconnect and that the memory controller was belatedly moving on-die.
But I also knew Intel had done a stand up job of papering over the cracks of the Core 2's architecture with features such as massive on-die cache memory pools. Likewise, the Core i7's actual execution cores are only a relatively modest revision of those found in the Core 2 family. Similarly, one can't help but notice that AMD's Athlon and Phenom processors sport a very similar set of integrated features. A fat lot of good it is currently doing for AMD.
Am I still dubious about HyperThreading?
I was particularly dubious about the resurrection of HyperThreading, too. Surely this was an indication Intel was relapsing back to the bad old days where the marketing men called the shots. Just in case you've already forgotten, it was the silly marketing-driven obsession with MHz and GHz that got Intel into a sticky situation with the Pentium 4 Netburst architecture.
In single-core Pentium 4 processors HyperThreading typically added no more than a few percentage points of performance. For the dual-core Pentium D, it frequently sapped performance thanks to the way Windows manages threads. So, I wasn't exactly expecting it to perform miracles in Core i7. And anyway, surely Core i7's quad-core architecture alone has the multi-threading angle covered?
HyperThreading aside, therefore, my take on Core i7 was that it was more about laying solid foundations for the future. As Intel adds more and more cores to its processors, a modular architecture with plenty of bandwidth will be essential to keep performance scaling efficiently. That's exactly what Core i7 delivers. So, it's the eight-core version, pencilled in for a late 2009 launch, that I thought would be the defining model in the Core i7 family.
However, more than anything it was good old fashioned cynicism that put paid to my expectations. For starters, the Core 2 was so good, what were the odds that Intel would once again be able to raise the bar so dramatically? What's more, Intel already has the desktop well and truly locked down in performance terms with the Core 2 range. Its sole rival AMD can't get close to the Core 2 much less anything quicker.
Kudos must therefore go to Intel for sticking with its aggressive Tick-Tock product release plan. And shame on me for ever doubting that Core i7 would be anything less than a rocket ship. But the short version goes something like this. For multi-threaded software, it's the absolute bomb - as much as 60 per cent quicker than the best Core 2 processor in fact. That's at least as big an improvement as Core 2 managed compared with the borderline broken Pentium 4 NetBurst architecture. Honestly, I didn't think it was possible.
As for HyperThreading, well, strictly speaking it's probably not quite as earth shatteringly effective as Intel would have you believe. But the 15 per cent typical performance boost it delivered in my multi-threaded testing is certainly not to be sniffed at.
Single-threaded stall
That's not to say Core i7 doesn't have any weaknesses. It doesn't move the game on terribly far for single-threaded performance. Early indications also suggest that it doesn't quite deliver the enormous overclocking headroom of the last of the Core 2 processors. Oh, and for the record, HyperThreading does appear to occasionally put the kybosh on performance, but only a little and only in fairly isolated circumstances.
So, it's onwards and upwards for Intel's Tick-Tock strategy and no more doubts from me. Well, very nearly. I certainly expect Intel will be on target with the next two installments of Tick-Tock. First up is a process shrink to 32nm for the Nehalem architecture that underpins Core i7. Then comes Sandy Bridge, the next major architectural overhaul.
Those products are essentially done and dusted. The R&D has been paid for during the dying years of the global economic boom, the wheels are in motion and only some unforeseen catastrophe will prevent them from appearing on schedule. But look a little further out and my doubting Thomas ways begin to take hold again.
With memories inside Intel of the troubled Pentium 4 fading and AMD apparently unable to truly compete, surely complacency will eventually set in. Then there's the looming economic recession. Intel's revenues will inevitably suffer and the temptation to lift off the throttle will be even greater. I therefore suggest you make the most of the next year or two. If the world economy does eventually plummet into a long depression, we could be in the dying days of a golden era for personal computing.
You can read Jeremy's Technocrat column every month in PC Plus magazine. And check out our Ultimate Guide to Core i7.
