Bits & Bytes
Written By Van Smith
Date: August 16, 2003
By now, everyone knows that Prescott is hot. "Prescott" is the code name for Intel's soon-to-be-released update to the Pentium 4 core. In fact, Prescott produces a lot more heat than the chipmaker had planned. Some in the industry believe that Intel will lick this problem, but the hurdle Intel is facing is more than meets the eye.
In previous articles we have contrasted Intel's "speed demon" versus AMD's "Brainiac" approach. Although Prescott improves upon the existing Pentium 4's performance at any given clock speed, the CPU still depends upon GHz to remain performance competitive especially with AMD's Athlon 64 looming soon.
Intel relies on fast transistors for the Pentium 4 and fast transistors are "leaky" transistors. What this means is that, typically, the transistors' gate size must be reduced as much as possible so that each transistor can switch more quickly. But, by shrinking the gate size, more and more electrons can seep through and around gates, causing power consumption even when the processor is doing nothing.
Even at 0.13-microns, the leakage current for today's Pentium 4 "Northwood" core is very high. In some cases, Northwood leakage power is greater than 20W. Despite the fact that worst case Northwood power peaks above 100W, the processor's real issue in the mobile space is its non-stop leakage power. An idle Pentium 4 will suck a laptop's battery dry very quickly.
That is why Intel introduced Banias, a CPU designed for the mobile space to run very efficiently at much lower frequencies than the Pentium 4. Banias doesn't demand the fastest transistors as it doesn't have to reach very high frequencies. For instance, a 1.4GHz Banias can often run applications as fast as a Pentium 4 clocked 1GHz higher. Consequently, Banias doesn't need higher voltages to ramp clock speed.
And most importantly for the mobile space, Banias's leakage power is a fraction of a Watt.
Almost certainly, the entire amount of excess power that Prescott will be consuming is due to leakage. At 30W or even greater, leakage power will make Prescott a horrific notebook part if battery life is any concern at all.
In fairness, SpeedStep will help mitigate the problem a little, but Prescott looks to be the worst mobile part ever introduced.
Believe it or not, but notebook manufacturers are working to accommodate Prescott and its warts by spec'ing 110W chassis, bigger than ever power bricks and thermal design power solutions of 90W!
In the desktop arena, Prescott looks to consume as much as 20% more power
than Northwood yet have a significantly smaller die size. In desktops,
leakage power of 30W can be dealt with relatively easily, but peaking well over
100W with a tiny-little die means crazy-high thermal densities. It will be
interesting to see how Intel attempts to cope with such a prodigious little
furnace.
But the bugaboos of 90nm technology are not just limited to Intel.
Everyone is going to have to deal leakage problems that are simply huge.
And unlike years past, there is no quick fix for Intel or anyone else in sight.
There is however a solution for Prescott's power problems. Intel would be much better served by dumping the P4-P5 line, optimizing Banias and its descendents for speed, and adding multiple cores per die. Such a processor would be cheaper and easier to produce, run a heck of a lot cooler and quieter than the P4, and would eat Prescott's lunch and dinner.
Alas, it is not likely to happen, because if Banias and/or its progeny hit the desktop, the Pentium 4-5's market share would implode, leaving many of Intel's vaunted and expensive fabs idle.
However, even Banias's successor, Dothan, is having 90nm power problems. Unlike Prescott, the issue isn't really a combination of overall power and leakage as much as it is just flat out leakage. Sadly, this actually makes Dothan's situation more delicate for the mobile space it targets since leakage is so critical in this area.
But the puny 10 Watts or so of Dothan leakage power would be no sweat at all in a desktop chassis.
One last irony, Banias is a very simple design, being the logical evolution of the Pentium III core (albeit an evolutionary path that has performance compromises in order to save power). A small, talented CPU design team of eighty or so, if given the Pentium III as a starting point, could whip out a Banias-like chip in about two years time. The Pentium 4, on the other hand, required a design effort that recalls the construction of the Great Wall of China -- and it remains a work in progress with Tejas, a core design that we exclusively reported, finally patching many of its holes.
And speaking of our exclusive reports, we revealed the existence of AMD64-like support in the first Prescott samples on June 26th of last year. This occurred at a time when Intel's Paul Otellini was vehemently denying the existence of 64-bit instructions in any Intel processor besides Itanium.
Irony or not, Otellini's denial came in the same timeframe we were told that Intel, together with Microsoft, would announce support for x86-64 in a new version of Windows. As we revealed, Microsoft had been the driving force behind Intel's adoption of AMD's 64-bit instruction set in the first place. And as we further reported, Microsoft's support of AMD's initiative would not be limited to the server space. Of course, it is now public knowledge that 64-bit server and desktop versions of Windows are scheduled to be released in the coming months.
And on the Yamhill side, Hans de Vries of Chip Architect has done some fine detective work that has all but confirmed the existence of the 64-bit processing potential of Prescott.
Now whether or not the 64-bit functionality of Prescott actually works is another issue.
===================================
Leaping to Linux
The dag-nabbed Blaster worm is causing a lot of fuss, and rightfully so. With Windows having such a glaring DCOM security hole, many more people seem to be looking towards Linux for a secure desktop OS alternative.
I have been tinkering with the latest SuSE, Mandrake, Red Hat and Lindows distributions. Although my personal favorite is Red Hat 9.0 for its unmatched polish, level of documentation, and its beautiful Gnome interface, Lindows 4.0 is simply head and shoulders above all other Linux distributions for Windows users wanting to migrate quickly to Linux.
But don't think that even Lindows 4.0 is going to be a smooth move from Microsoft's warm embrace (ahem) because Linux is still not as mature as Windows in terms of driver support, GUI system administration tools, software availability and power saving support.
Despite this, for many people Lindows 4.0 will be more than good enough right now. And what makes recommending Lindows 4.0 easier is the phenomenal pace of development of that OS. I have participated in Lindows OS testing from just about the beginning and the folks at Lindows have fixed bugs and smoothed out the rough edges almost as fast as you can spot them.
A couple of things I dearly hope they work on is a much more polished Samba GUI tool, so that different security levels can easily be set on shared directories -- without hosing-up the whole dang Windows network. And they have got to do something about fooze and mounting drives -- it can be a nightmare.
Despite being a notorious cheapskate, I like Lindows enough to have purchased a two years Lindows Insiders subscription for my family.
===================================
PocoMail
I don't like forking over money for programs, but I recently bought a license to PocoMail. Of course, with all of the free and powerful email programs available, that made this decision especially difficult for a tightwad like me.
However, PocoMail is that good.
Slavin Radic, the lead PocoMail developer, is obviously a consumed, obsessed, maniac bent on making the most feature rich email client on the planet. Oh yes, he has already succeeded at this goal a long time ago, but that hasn't prevented him from continuing to work on PocoMail unabated.
PocoMail has a head-spinning list of features. In fact, there are so many features and options that a new user can quickly get lost.
However, if Slavin can successfully coral all of the ponies he's trying to throw under PocoMail's hood, he'll have produced the best email client around. And after watching PocoMail's development over the last year, this is likely an inevitability.
===================================
Pssst! We've updated our Shopping Page.
===================================