Why Silicon Features are so Important Now - Blue Crystals Hidden in the Northwood
(Page 2 of 3 )
The Northwood core had one more hidden talent: it was the first “blue crystal” that Intel brought out which easily showed its benefits to normal consumers via the use of a previously dormant function: Hyper-Threading (HT). Hyper-Threading was first enabled in the 3.06GHz model, and eventually turned on in the lower speed specifications as well. Unlike MMX and SSE(2), this was something that you could see within a normal operating environment, even without running a program since, as far as the operating system was concerned, your single CPU was actually two! I’m not going to get into the implementation or performance differences of what this accomplished, nor the legal issues
behind why it laid dormant for so long. Sufficient to say, however, that this was something that was not present in the competition’s line of products. HT was something Intel was quite adept at driving home in its ads, since it now had a technological advantage over AMD. While most software didn’t support this at first, many software manufacturers made an attempt to get on board and take advantage of it soon after its introduction.
Hyper-Threading was a bit of a pre-emptive strike by Intel. Many processor manufactures were already planning to eventually move towards putting multiple full CPU cores on one die. SUN, IBM, and AMD were all going in this direction. At that point in time, however, Intel’s only announced architectures going multi-core were their Itanium server level chips. Of course, it would have been only be a matter of time before that technology drifted down to the desktop sector. But by putting HT out there at that time, the software designers would work towards having all performance oriented programs set up for multiple threads, ensuring there was a decent base of optimized software for the future multi-core processors.
Few, however, saw what was really coming. Prescott, the .09 micron (90 nanometer) revision of the Netburst architecture, and Northwood’s replacement, was supposed to allow the P4 to scale well through 5GHz and beyond. Based on how well Northwood had moved on the .15 micron process, this seemed like a piece of cake. Unfortunately, Intel’s R&D had a big problem: speed increases on the new process were no longer “free”, like they had been in the past.
Typically moving the transistors closer together makes them smaller, using less voltage while switching faster. The chips require less power to run, and are therefore cooler than the previous generation’s top end processors. This is a bit like up-shifting in your car. You can go the same speed, while using less RPM, and as a result having higher headroom for speed.
Unfortunately for Intel, something broke. Instead of running cooler, Prescott was actually hotter than the equivalent Northwood! Not a good situation for scaling the CPU. Heat causes transistors to switch incorrectly, generating errors. There are ways around this of course. Computer enthusiasts and overclockers have for years been avoiding this issue by installing larger heat sinks, fan, and even more extravagant solutions such as water or phase change cooling. Companies like Dell and Compaq, who expect their systems to be quiet as a coffin in addition to cheap, weren’t exactly jumping with joy over a situation like this.
But Intel wasn’t the only one hitting this wall. IBM’s Power5 and PowerPC 970 processors were supposed to move down to a 90 nanometer process, and climb up to near 3GHz at the high end. They also found the finished CPUs coming off this process were much hotter than expected, and required more voltage to hit various speed levels stably, exactly the same issue Intel had. From all indications so far, while AMD’s 90nm CPU’s at this time aren’t experiencing the same heat issues that both IBM and Intel have, they aren’t having much success scaling the processor speed regardless. All of this begs the question, “what went wrong?”
Next: What Went Wrong? >>
More Computer Processors Articles
More By DMOS
