As expected, Sandy Bridge launched this month, among a flurry of revelatory announcements, news, and rumors within various areas of the tech industry. Sandy Bridge's launch is preceded by much expectation, amidst a string of products made in recent years by a company known for delivering on such expectations. In many ways Sandy Bridge continues on that legacy, but in many ways, it's largely forgettable, certainly the most forgettable since the Core 2 generation first launched.
As a new generation, you expect new features and a new architecture to be introduced, much like the Core 2 was a complete change in philosophy from the Pentium 4, and the first-gen Core i7 was a complete change in cache memory hierarchy, and the first to bring the northbridge on die for a consumer Intel chip. So what does Sandy Bridge bring? Improved IGP and an integrated video encoder. Granted, there are a lot more low-level stuff, but none of them are really revolutionary like the last two generations, and as far as the end-user is concerned, there isn't a whole lot of new stuff to be had. Of course, at some point you have to expect the advances to slow down. There were a lot of big-ticket items Intel needed to get out of the way in the move away from P4, and in the competition against AMD. AMD rarely had the performance crown, but there were plenty of enviable innovations under their heatspreaders, and for the most part, Intel's been spending the last four years copying those ideas while also making them perform at a benchmark level.
But now even speed is starting to get less attention. Intel's made huge strides in IPC, more than most people expected they could, but even with the benefit of a die shrink, it looks like the aces are beginning to dwindle from underneath Intel's sleeves, as the burden grows heavier on the need for parallelism to bring about the next strides in performance. To that end, Intel's got a couple of nifty additions. The first is a very fast 256-bit AVX vector engine, and the second is the aforementioned video encoder.
The video encoder is an interesting addition, and one not entirely expected. It's obvious Intel's beginning to feel the squeeze from GPGPU progression, or rather, they felt the squeeze some years ago when they were still formulating Sandy Bridge. It's not hard to foresee the growing desire to move such a parallel load to a chip that is best suited for such program types, and a CPU's core count can't hope to catch up to all the little stream processors comprising a GPU. So a little stop-gap measure is needed, and what you see is Intel installing a fixed-function vector unit specifically to address this concern. The good news is this stop-gap measure is a damn fast one, and it can produce some good results. The bad news is, just like CUDA and OpenCL and any other co-processor initiative, it requires new programs to be written to take advantage of it. The other problem is that the video encoder, because it is fixed-function, while efficient from a transistor standpoint, won't be able to support codecs beyond what was already considered in its design. Intel is hoping to leverage their software developer ties to get Multi-Format Codec (MFX) support, but while performance is great, the inflexibility will be less appealing compared to OpenCL solutions, especially when NVIDIA's future chips require less intervention from the CPU to impede on the performance picture.
The other appealing feature for consumers, at least the enthusiasts among them, is the overclocking capabilities. New for this generation is K-branded CPU models that sport unlocked clockspeed multipliers for a little extra money, a feature AMD users have been familiar with for some time now. While in many other areas Intel has done a lot to try to constrict overclocking ability in their products, this is the first time Intel has really embraced the idea, and it's already looking to be a very popular move on their part. No longer limited by FSB constraints, tinkerers are free to explore the furthest boundaries of their CPU's capabilities, and the results have been some very impressive breakthroughs on the existing barriers of air-cooled clockspeeds. I remember when 4GHz was a strain on even Vapochill-equipped systems. These days, it's just a starting point.
Even still, clock-for-clock, little has changed compared to Nehalem, and it's evident even in the name they chose for the new line of CPUs, sticking with the tried-and-true "i7" (and lower numerations) branding. In many ways this doesn't even feel like a new generation, as the first 32nm processors came about with little more than a whimper, mostly focused on the mobile side of things. Notice most people are still using the "Sandy Bridge" distinction, even though codenames usually die out after the product's release. Add to that, Intel still insists on their bombardment of new sockets and chipsets to go with every CPU iteration, so if you want to adopt the new family, you'll have to also befriend its uglier companion platform.
AMD has always been better about this, allowing lots of overlap of their new socket introductions, even across new memory standards. They won't get any advantages though when Bulldozer arrives, as you'll still have to upgrade your motherboard then, and if you're still on DDR2 which a lot of folks are, you'll have to upgrade that as well, whether there are tangible benefits to it or not. The notion of choosing an upgrade path becomes more ambiguous then, if you're not in a big hurry to improve your system performance. If you need the performance sooner rather than later, you'll doubtlessly be happy with a Sandy Bridge setup. If you're otherwise pretty well happy with what you have, and only want the speedups for more extreme demands, either wait until the next-next generation, or see what Bulldozer brings to the table. The architectures are looking strikingly similar between the two, with physical register files, macro-op fusion, and multi-threaded cores, and with AMD bringing back the FX moniker, it may mean a return to form for AMD's performance prospects, or at least, reason enough to wait it out and see, if you can.
Intel also more than doubled the IGP performance, but this is a moot point on the onset of the AMD's Llano release. If there's anything worthwhile to note about the improved graphics capabilities, it's the implication that Intel is finally taking that aspect more seriously, in hopes of bringing up the minimum graphics performance of their solutions to a level approaching acceptable. This is probably spurred by the same reasoning behind their video encode engine, and that is the threat of an AMD equipped with a mighty GPU powerhouse behind them. If AMD can eventually achieve the widespread proliferation of Fusion chips in the market, that will be a huge boom for their platforms in the OEM space, moreso if OpenCL applications ever take off. Intel isn't likely to ever catch up to AMD's IGPs, but the last thing they want is to look completely stupid sitting next to them, and an improved HD graphics component will at least help to save face.
Really, it looks more like Sandy Bridge is just the dip of a toe in the next-gen waters, waiting to see what the ripples stir up in the other camp. When Bulldozer hits, Intel can move in with their hexa-core and eventually octo-core parts, sufficiently effecting a counter-blow that will restore them the performance crown, if that's even in contention. It may even be as simple as a clockspeed bump, but either way signs point to a more apprehensive Intel than in the past couple go-rounds, and a much more competitive AMD. Here's hoping for the best.
