Does megahertz matter?
With its latest series, the Athlon XP challenging Intel's Pentium 4 range, chipmaker AMD would like buyers to forget clock speeds _ and concentrate instead on overall performance. Anand Parthasarathy tries to separate fact from hype in this game of processor oneupmanship.
SIZE DOES matter' they said _ and that may well be true about rampaging reptiles like Godzilla. But when it comes to computer processor chips, 'Speed matters _ but not all that much' might well emerge as the ''mantra`` we need to mug up.
For some weeks now _ ever since November 5 when Advanced Micro Devices (AMD) unveiled its top-of-the-line ''AthlonXP 1900 plus``, to challenge Intel's Pentium 4 line _ we are being stalked by these two heavyweight chipmakers even as they continue to play the game called 'galloping gigahertz'. AMD crossed the 1 gigahertz lap first, but Intel caught up fast. And having clocked 2 gigahertz ( that's 2 billion clock cycles per second), it has been cruising along gently, with no competitor in sight.
The latest AthlonXP ( cannily named to suggest perhaps, that this is a great chip to work with the new Windows XP operating system!) has been confusingly named 1900 plus. In fact, the small print shows that its real clock speed is 1.7 GHz, but dealers have been seemingly cued to say that the 1900 means that this chip is better than a comparable Pentium 4 at 1.9 GHz.
In fact with its new naming system, AMD would like customers to move away with a fixation on clock speed alone, to compare processors. It's performance that matters, it says _ and a chip with a slower clock speed could still end up doing most of the PC's jobs, faster.
Avis , long the number 2 car hire company in the US, after Hertz, once came up with an inspired advertising slogan: ''When you're second best, you try harder``. Is what we are seeing now, in the PC chip arena, a similar case of the number 2 guy ( AMD is the main competitor to market leader Intel), trying very hard, with some slick technical ''jadoo``, to convince us that second best is indeed better? Or do megahertz really not matter?
One respected watcher of the chip scene, Mr Manoj Nadkarni, founder- President of Chipinvestor.com, a US-based company that advises investors in the processor industry, was quoted recently as saying quite plainly: ''Megahertz alone is not a complete measure of performance``. Indeed senior Intel designers whom I have had the good fortunate to meet at their annual developer forums, have also been candid enough to admit that chip clock speed is just one, not the only measure of a processor's performance.
So what are the others?
One needs to understand first, how in the contemporary computer chip, the central processor unit or CPU works. It must communicate with the computer's memory to retrieve the instructions from the programme. It does this using the data bus. The bus today is 32 bits wide ( the early chips like the 8088 was only 8-bits wide).
This means it handles a byte or word, 32 bits at a time. The bus itself runs at around 133 MHz today on most PC mother boards. But ( and this is the clever part) by jacking up the clock speed, that is the speed at which the chip acts as an on-off switch, to many times more than the bus speed, the designers make the CPU work effectively much faster.
One may ask, if the bus which carries data to and from is still plodding along at a hundred megahertz or so, how can it be deceived into thinking in terms of 1000s of megahertz or 1 GHz?
The answer is to put a small cache of memory on the chip, so that the to-and-fro-ing from chip to main memory is cut down for some common tasks.
Thus, while the main memory typically consists of 64 or 128 or 256 megabytes of Random Access Memory (RAM), today's newer processors, like the Pentium 4 and the Athlon have an 'on board' cache of a few MBs to jack up the overall processing speed.
If one compared two chips from the same manufacturer, using identical architecture, one's task is easy; the 800 MHz Pentium 3 will work faster than a 650 MHz Pentium 3 and so on. But when the design changes _ as for instance between a Pentium 3 and a Pentium 4 _ the comparison is not straight forward.
In fact experts have said that for a P 3 and a P 4 working at the same clock speed, the P4 actually is 20 percent less efficient. The problem is compounded when one compares chips from different manufacturers. Unlike the AMD and Intel chips of a couple of years ago, today's Athlon and Pentium 4 have radically different architectures, in ways they organise their on board cache, what tasks are allotted to the cache... and so on.
So comparing an Athlon and a Pentium 4 purely on the basis of their clock speeds may be like comparing an apple to an orange on the basis on the number of pips or seeds they have. Last week AMD launched an advertisement drive claiming to tell the truth about processor performance. It provided a number of analogies: A car with the higher RPM may not go faster, but a car with more cylinders does; a child takes more steps than an adult to cover the same ground, but it cannot walk faster etc. All these merely tended to cloud the issue.
And that issue as far as it concerns the customer is really quite simple: How can I do all things I want to do on my computer in the most efficient manner and at the least cost?
Many PC users have discovered to their chagrin, that if some of their most-used applications, Excel spread sheets, or photo editing or whatever, tend to work sluggishly, the solution did not lie in upgrading to a costly new chip with a faster clock speed.
Rather,a modest increase in RAM memory, from 64 to 128 MB which today costs only around Rs 600, may result in dramatic improvement. Why? Because every computer reads and writes data much faster to RAM than to the hard disk; so the more RAM you provide, the less the application has to move to and from the hard drive.
Then there are other elements that matter when we use the PC's multimedia features like watching movies on CD or downloading animation .
Here the video card comes into play _ and a variety of third party cards mostly from Taiwan or Korea are available and each model has different amounts of cache on board and is optimised for different applications.
A good assembler will advise you about the best video card for your particular application. It is also worth remembering that while the G4 Power PC processors that are used in the Apple range of computers have always been slower than the Pentiums ( they top out at around 733 MHz), Apple Macs have always been superior platforms than any Wintel ( Windows software and Intel chip) PC for applications like iMovie or iDVD.
But for the rest of us, who do only fairly basic things with our PCs _ surf the Internet, write some letters, keep our accounts, see some movies hear some CDs, and make the occasional presentation _ the truth is the PC of contemporary specification is already a massive overkill _ we are paying for a lot more processing power than we need. For us, the ever increasing chip speed is a non issue. We could change to a chip at twice the speed and would be hard put to tell any difference in performance.
That's the cunning of today's marketing techniques: we are nudged into worrying about gigahertz and gigabytes when we should be demanding: ``
Give us the configuration that will do our job at an affordable price, while offering us a small margin for growth. And keep those galloping gigahertz for the gullible ones for whom the bytes are always bigger on the other side of the cyber fence.``
Send this article to Friends by