Old trick aids Intel in 1GHz notebook spin
Not throttling, dolt! It's thermal management
Intel's SpeedStep technology allowed it to intro a 1GHz Pentium III mobile this week, beating AMD to the coveted speed tag.
But the jury is out on just how much chips used in high end notebooks really can be truthfully described as running at 1000MHz.
Intel's SpeedStep (GeyservilLe) technology goes head-to-head against technology from AMD (PowerNow) and Transmeta (LongRun). All are techniques to control power consumption in mobile microprocessors and so let the battery last longer. Via also has a system for the few fanless mobile CPUs it flogs.
But, according to one source, whether an Intel notebook is plugged into the mains or running on its battery, it only hits 1GHz for a very short term. Comparing CPU speed in a desktop with a mobile is therefore misleading.
He said: "A 1GHz Pentium III mobile is too hot to run at that speed all the time, so manufacturers need to "clock-throttle" the chip, which means momentarily reducing the speed to avoid overheating.
"There is little difference between the various speed grades since chips running at higher MHz ratings require much more clock throttling then the lower speed grades. The scaling is very bad compared to that of desktops where clock throttling doesn't exist."
Benchmarking a notebook with a claimed speed of 1GHz and a desktop at 1GHz - and ignoring hard drives and VGAs because of the innate differences of these technologies in these platforms - would mean a huge difference in performance, despite the speed tag, he claimed.
Other readers of The Register, however, have been swift to point out the differences between the technologies from the competitors in this lucrative neck of the PC woods.
"Unlike all other modern mobile chips from Intel, this chip is basically a desktop processor with SpeedStep," one claimed. "It runs at the full desktop voltage to get to 1GHz, so it's using near to 38 Watts. You'd drain your average laptop battery pretty fast at that pace. Of course, it doesn't run at 1GHz when it's unplugged unless you disable SpeedStep, so it's really only a 700MHz chip when you're on the aeroplane. That's still fast, but the 1GHz is a little bogus."
Another said: "The APM/ACPI part of the BIOS will slow the clock if the CPU runs hot. That is the same part of the BIOS which does the SpeedStep thing, by the way. I'm not privy to any BIOS source code anymore so I can't tell you if they do this by reducing the FSB (front side bus) frequency or by fiddling the multiplier on the CPU.
"The best place to look is probably the chipset docs, because it is the Northbridge which implements the STOPCLOCK hardware protocol with the CPU which makes it possible to fiddle with multipliers and bus clock rates.
"Most operating systems these days will halt the CPU whenever they have nothing else to do that to wait for the next interrupt (keyboard, disk, network, whatever), that way you can run the CPU at say 1GHz as long as you only do some light work like typing in a word processor because it really only runs 10 per cent of the time so the thermal output corresponds to say 300MHz.
"As soon as you throw real work at the CPU, calculations, reformatting a 100 page doc with a new template, MPEGs and so on, the CPU will quickly hit the thermal limit and be slowed down by the APM/ACPI BIOS so that it can continue to run without melting the plastic around it.
But, he added: "The battery lifetime issue is a very good reason for slowing the clock."
Replacement batteries for top-of-the-range notebooks are often sold by vendors at a hefty premium to those flush enough to have found $2,500 in their Four Plus pockets*.
But clock throttling is not a new phenomenon in Intel chips. Far from it, according to one source close to the development of all former and latter day x86 processors
"Clock throttling was around since the original Pentium; it was certainly around before Intel invented SpeedStep, to wit Transmeta responded with LongRun. Clock throttling was Intel's original form of controlling power consumption in mobile chips.
"Here's how it works. A 4KHz square wave is applied to the STPCLK pin of the CPU. Depending on the duty cycle, a greater degree of clock throttling occurs. When the square wave is active, the CPU actually stops executing code and essentially shuts down (this saves power). As you can see from this method, the greater the duty cycle of the square wave, the longer the CPU is inactive; hence the greater the power savings.
"I believe clock throttling begins at 75% (25% duty cycle) and goes down to 12.5% (87.5% duty cycle).
"Here's where the fun begins. Guess how Intel acheived their "1-W CPU" to answer Transmeta? First, they took their 1GHz CPU from the top of their yield. Due to yield characteristics, these top of yield chips also have the ability to run at the lowest voltages. So, first they robbed their most profitable chip line to create this chip. Second, they reduced the frequency to 600MHz and lowered the voltage down to something like 1.15V. Third, they clock throttled the thing down to 12.5% utilisation -- thereby creating the performance equivalent of a 75 Mhz Pentium III (600 MHz * 0.125 (12.5%)). Finally, they sat idle in MS-Word and measured 1 Watt.
That, he said, should be compared with Transmeta, which can play a DVD at one Watt at full speed, using LongRun.
It's certain that tier one vendors, a rash of which endorsed Intel's 1GHz PIII mobile launch earlier this week, implement the technology in different ways. Late last year, we saw Toshiba technology which implemented water cooling in a high end notebook.
A technician at a tier one vendor, who didn't want him or his company to be named, said: "I was at an Intel meeting about a year ago where it was explained that Intel marketing didn't like the term throttling, and changed it to thermal management." ®
* RegRoyalFactoid Zips, zippers were brought into vogue by Edward VIII, who abdicated as Emperor of India and King for the love of an American woman. It is not known how many of these different facts are related. King Ed is also alleged to have made Plus Four trousers the top pants in the US fashion scene in 1924.