Intel chips away in the Coppermine

The recent battles between processor giant Intel and its nearest rival, AMD, have been well documented. In case you have somehow managed to avoid this conflict, in essence, it revolves around AMD's release of the Athlon processor.

AMD managed to impress everybody by releasing a processor that was quicker than its Intel equivalent, in this case the Pentium III processor. This had never happened before - say what you like about Intel's processors, the simple fact is, they were always the quickest and best on the market.

But with the Athlon, AMD managed to overtake Intel and cause some trouble along the way.

Intel bites back
But Intel is no shrinking violet, and has a great deal of experience in processor development. And so it returns to the field with its Coppermine technology to try to wrestle the crown back.

The first point about Coppermine is that it is not an entirely new technology, but more of an update and refinement of the existing Pentium III processor. You don't start to enjoy the benefits of Coppermine until you start buying processors of speeds greater than 600MHz.

What does Coppermine actually offer?
The first new feature is a result of the way the chip is manufactured. Intel has managed to scale down the manufacturing process to use 0.18-micron technology. Current processors, including the Athlon, are manufactured using 0.25-micron technology.

What does this mean in the long run? The answer is in two parts. First, the smaller manufacturing process means that the processor core uses less silicon. In fact, production costs for a Coppermine processor are lower than those for the original Pentium III.

Second, Coppermine requires less power to run. Less power means a cooler processor, allowing notebooks and other mobile platforms to take advantage of the superior power consumption.

These factors have come together to enable Intel to immediately release a 733MHz processor, compared to AMD's 700MHz Athlon.

Following neatly on from this is a new method of regulating power for the mobile user. Having a notebook is only worth it while you still have power, and for most applications that you will run while on the move, performance is less of an issue than a consistently working machine.

With this in mind, Intel has provided SpeedStep, a function that automatically decreases the amount of voltage and clock speed to your processor when the notebook is running out of battery power. This reduces performance, but on the plus side, it increases battery life. It's a trade-off, but you can disable one if you absolutely must have your notebook constantly running at full speed.

New components
It's not just the manufacturing process that has come in for a major overhaul. Several of the Pentium III's key components have been updated to suit the faster processor speeds. Perhaps the most important of these updates comes from the Level 2 (L2) cache. To explain better, an understanding is needed of exactly what L2 cache is, and what job it performs. So we'll start with a basic overview of Level 1 (L1) cache.

L1 cache sits on the processor and runs at the same speed. In the case of the Pentium III, there is 32Kb of cache, split into two 16Kb sections - one for data, and one for instructions. Should the processor need the same information again, it can retrieve it from the cache, which works out quicker than actively performing the same task repeatedly.

In the same way, L2 cache sits between the processor and main memory. No matter how fast memory gets, the central processor is still quicker and ends up waiting for it to respond. The solution is to fit a secondary cache to help alleviate this problem.

L2 cache is often called external cache because it sits off the processor. However, Coppermine moves the L2 cache to being on-die - i.e. the whole processor, plus cache, is just a single chip - and slices the amount down to 256Kb from 512Kb from the earlier Pentium III models.

Cut down the amount of cache by half? Doesn't this make the Coppermine worse than the Pentium III? The simple answer is no. But keep reading and we'll explain why.

The older external cache versions of the Pentium III only had the L2 cache running at half core speed, so immediately you have twice the bandwidth available to you. This extra width actually means that the cache is now faster than before.

This is part of the reason why Celeron processors achieved such popularity: their onboard L2 cache meant that once you started to over-clock them, they could outperform their Pentium II rivals.

The L2 cache isn't just on-die, running twice as fast as external cache. Some design improvements have also been made. Intel calls this new design 'advanced transfer cache' (ATC). It sounds impressive, and even has the appropriate computer industry acronym; but what does it mean?

ATC refers to the increased bandwidth and decreased latency of the new Coppermine cache. First, Intel increased the data path width to 256-bit. As a comparison, the standard Pentium III has only a 64-bit wide data path.

In addition, the other task was for Intel to reduce the latency of the L2 cache. Reduced latency helps when the processor fails to retrieve something from the L1 cache and needs to move on to the L2 instead. So Intel managed to reduce latency by a quarter. The new eight-way associative design of the L2 cache, coupled with the fact that it is now running at core speed, means that cache transfer rates can rise as high as 11.7Gb/sec.

Hitting the buffers
Another term you may come across with reference to Coppermine processors is advanced system buffering (ASB). It refers to the increase in buffers for the processor that enables it to take advantage of the 133MHz bus speed. Put simply, the more buffers there are in place, the more outstanding operations there are that can exist, which means that no component is waiting for another one to catch up. They can all keep running as normal.

More changes in the pipeline
With regard to the processor itself, this is what has been changed. Still in the core are support for MMX, and Streaming SIMD instructions, which offer greater performance in applications such as 3D modelling. Because Intel provides developers with a large amount of support, and compilers which take advantage of their architecture, expect to see Coppermine well supported by upcoming applications.

AMD hasn't managed to achieve anywhere near this amount of market penetration, so unfortunately 3DNow! isn't as popular as it could be.

However, that isn't where Intel stopped. I'm sure that some readers are thinking, 'what about the motherboard?' After all, AMD created a new chipset to work with the Athlon.

And that's the next point. To take full advantage of Coppermine, you'll want to take a look at the new 820 chipset as well.

First on the 820 is support for a 133MHz front side bus (FSB). Currently, the maximum on an Intel board is 100MHz, giving the new chipset a clear 33 per cent advantage over its predecessor.

For increasingly data dependant operations, such as improved graphics throughput, the 133MHz bus is going to start showing its worth. It's worth mentioning that not all of the new Coppermine processors will run at 133MHz FSB, so take a look before you buy to make sure that you get one that does.

If you've followed memory enhancements through the years, you'll know that PC-100 SDRam is the current performance standard, running at 100MHz. As you'll see, it's no use having memory that runs at 100MHz if your bus runs at 133MHz. So the 820 chipset will have support for PC-133 SDRam instead.

But before you rush out to do this, you might want to consider the alternative.

Keeping up with the chip
Intel has quite sensibly pointed out that as processor speeds leap forward at a tremendous rate, memory performance has never really been able to keep up. Having 133MHz memory is only really a short-term solution to a growing problem. Intel has therefore come up with Rambus (RDRam) memory.

Rambus is currently available in three varieties, ranging from 300MHz to 400MHz. At the high end, it will offer twice the performance of PC-100 SDRam. There are a couple of downsides to consider, however.

Because Rambus is a new technology, expect to pay in the region of five times more than PC-100 SDRam. And while it may be quicker, the processor can only take advantage of an additional 33 per cent speed increase, because its bus is running at 133MHz. RDRam is actually slower as regards latency than PC-100, so for random reads and writes, it's actually going to be outperformed by the older technology.

RDRam is also going to affect the performance of SDRam, thanks to the way the 820 chipset deals with the memory. The entire chipset is geared to take advantage of RDRam: any SDRam on the motherboard has to pass through a Memory Translator, so that SDRam signals look like RDRam. No matter how small the performance hit. Performing a translation will have some effect on performance overall.

Perhaps the most important new feature the 820 has to offer is support for AGP4x interfacing. As a direct comparison, AGP2x, which all Athlon boards use, runs at 532Mb/sec. Thanks to the increased bus speed to 133MHz, AGP4x will allow 1064Mb/sec.

It doesn't take a genius to work out that the increased throughput is going to make a big difference on high-end graphics applications, which really are bandwidth hungry. Newer graphics cards, such as those based on Nvidia's GeForce 256, support this new standard, and it's really here that Intel is quite a way ahead of the competition.

A level playing field
In all, Intel has addressed certain issues with its processors and managed to increase speed thanks to changes throughout the system. These mostly centre around available bandwidth for system peripherals such as memory. The solution here was to increase cacheing speed and available bandwidth.

The best combination is to use Coppermine with the 820 chipset, because this is where you'll begin to see the differences. And if you can be sure of one thing, it's that Intel has the facilities and money to make sure that supply of the new processors is constant - something AMD has had some trouble achieving.

The real question at this point has to be, should you bin your Athlon and move back to Intel? At the moment, no!

The Coppermine and the 820 chipset take back some of what AMD managed to steal away from Intel, but the Athlon is still a great processor. The Athlon wins out when it comes down to the floating-point instructions, but for other applications Coppermine is the processor of the moment.

The playing field is now quite even as both sides get ready for the next stage. AMD is planning a 0.18-micron version of the Athlon and a 64-bit processor to compete with Intel's IA-64 architecture.

When it comes down to it, fierce competition offers more choice for consumers, which is what we really want, after all. And there's nothing like a tough marketplace to force better technological development. Keep watching to see what AMD does in retaliation for Coppermine - it's going to be an interesting ride.