Review Once you've invested a few billion dollars in a fabrication plant and have a team of skilled engineers beavering away, you'll find it relatively easy to design and build an x86 processor. As time goes by you have to develop it to reduce production costs, raise efficiency and to increase its speed. Over the last 20 years we've seen chip manufacturers go through a ritualised dance to keep moving things along. Every few months they raise the clock speed multiplier. If things get tough, the engineers can increase the size of the L2 cache, writes Leo Waldock.

The real bonanza comes when the fabrication plant is ready to move to a new production process. The move from 250nm (Pentium II and III 'Katamai') to 180nm (Pentium III 'Coppermine' and Pentium 4 'Willamette') and on to 130nm (Pentium 4 'Northwood') has reaped huge returns. As the process reduces in size, Intel has been able to fit more transistors into the same area, or make the core smaller and cheaper. It's been able to continue upping clock speeds.

In short a new fabrication process is a joyous experience which offers the potential of massive profits - if it works.

As the processes have shrunk, chips have been experiencing some unusual physical effects. The main problem is quantum tunnelling. At school we're taught that atoms are like planets with electrons orbiting around them like moons. It's a neat image but quite inaccurate. According to quantum theory, atoms and electrons don't inhabit a fixed space but instead exist in a number of locations at once until they are fixed by being observed. It sounds like a plot line from Star Trek but it's an unfortunate fact of life that electrons leak through materials that appear to be solid. This errant behaviour has never been much of a problem before, but the big crunch came when Intel moved to the 90nm 'Prescott' Pentium 4.

Instead of seeing the usual gains of reduced power and raised speeds, Prescott actually worked slower than the previous Pentium 4. Intel had increased the length of the pipeline inside Prescott to make it easier to raise clock speeds in the future. That should have been invisible to outside observers, but we all noticed the performance hit, and none of us could ignore the TDP (Thermal Design Power) which broke past the 100W mark.

Intel responded swiftly by canning development of the 65nm 'Tejas' Pentium 4 and, soon after, by dropping the 4GHz Prescott. It also chose to accelerate development of dual-core processors. As far back as 1989, Intel's engineers knew that the day would come when conventional processors would grow too complicated and expensive, and the best way to increase processor performance was to add extra processing cores.

Taken at the most basic level, a dual-core processor works much like a dual-processor computer, except that it only has one processor socket. The big difference is that in a dual-Xeon workstation or server the two processors have to communicate with each other, as well as with the system memory, so there's inevitably a performance hit as the processors negotiate which data they will each work on.

In a Pentium 4 PC you already have two 'virtual' processors thanks to Hyper-Threading (HT), which mimics a second core by giving work to under-utilised execution units. That's fine if there are under-utilised execution units to work with - if not HT doesn't bring any benefit. HT won't ever deliver the same performance boost as adding a second, fully functional processing core will.

Intel will make its first stab at dual-core later this quarter, perhaps earlier than it would have liked. To pave the way for the launch, it recently began seeding pre-release dual-core test systems, and we got our hands on one.

Naturally, a new processor requires a new chipset, so the 955X steps in to supersede the 925X, while the 945P takes over from the 915P, but other than dual-core support it's steady as she goes with support for DDR 2 SDRAM and PCI Express. The Southbridge is ICH7R. Intel also decided that the new processor needed a new name.

Out goes 'Pentium 4'. In its place we get two flavours of dual-core processor for the desktop: 'Pentium Extreme Edition', which has twin cores each with Hyper-Threading, so one physical processor appears as four virtual processors. This will inevitably be the $999 part that looks good on paper but doesn't get bought. Hence the dual-core part called 'Pentium D' that doesn't have Hyper-Threading. It runs on the same 800MHz FSB and has the same cache but it appears as two processors. No doubt Pentium Extreme Edition (PEE) will move to a 1066MHz FSB, which will allow Intel to enable HT in Pentium D (PD) but for now it all looks like a bit of a marketing nightmare.

It's quite clear that Intel appreciates its predicament as it sent us a complete white box, rather than a regular press kit of processor and motherboard, along with a stack of documentation six inches thick. It insisted we should recognise that this is pre-production kit intended to give us a flavour for the PEE 840, rather than being a rock-solid retail-ready set-up.

From the outside, the PC looks quite conventional, and if you peer through the window in the side of the case you can't see anything out of the ordinary either. The motherboard is an Intel 955XBK with a 955X Northbridge and an ICH7R Southbridge, integrated Pro/1000 PM Ethernet, HD Audio and Sil3114A RAID controller. There's 1GB of dual-channel DDR 2 clocked at 667MHz (PC5300) memory running 5-5-5-15 timings, a 160GB Seagate 7200.7 hard drive, a Plextor PX-716A DVD writer, Sapphire Radeon X850 graphics card and an Enermax EPS 12V 550W power supply.

Intel advised us that the BIOS was incomplete and didn't support Serial ATA II but other than that it was good to go and as the processor was unlocked we might fancy overclocking it. Sadly, this was not the case. The BIOS was unusual as it didn't display POST information and there were no options to adjust either the FSB or clock multiplier. We also found that it wouldn't recognise our P4 EE 760 when we tried to do some back-to-back comparisons to see what effect the 955X chipset had on performance.

However, we could enable and disable Hyper-Threading, so we could turn the processor into a Pentium D at will, but that was about it. We got mildly excited when we found an option in the BIOS labelled 'Trusted Platform Module'. We asked Intel if it was introducing a TCP (Trusted Computing Platform), but no. TPM has been appearing on workstation motherboards for the last two years and is a hardened chip that can't be read by x-ray, so it can be used to store an encryption key to keep the hard drive secure. We've not seen this feature on a desktop motherboard before.

So how did this dual-core technology perform? We did some fairly quick and dirty tests as we only had a very short time to evaluate the system, so we felt it most beneficial to compare PEE with a dual-Xeon workstation which runs the same 3.2GHz clock speed but on a 533MHz FSB, and also against a 3.4GHz P4 EE.

Intel included documentation about tests we could run using sixteen different applications, and in each case the emphasis was on multimedia coding, which is understandable as this puts the emphasis on processing power. We ran SysMark 2002 on each platform (SYSmark 2004 would have taken too long) and then ran three of the suggested applications, and perhaps the biggest surprise is that Pentium D (ie. with Hyper-Threading disabled) only loses out to PEE 840 in movie encoding. No doubt other applications will show a similar effect but we would suggest that the vast majority of users will see no benefit from the higher-end chip.

Probably the least surprising result from our tests is that the PEE 840 behaves very much like our dual Xeons, despite the fact the latter uses the older 875P chipset and a slower FSB. Both configurations have four virtual processors, of course, but the similarities are startling.

That said, the average PC user should see a big performance gain when they run a dual-core processor. This ties in well with Intel's concept of the digital home, where a single PC will be performing multiple tasks for multiple users simultaneously. Imagine one person watching an HD movie while someone else plays Half-Life 2, without any degradation in performance.

To get an idea of how good the PEE is at multi-tasking we started by playing Doom 3 on High Quality settings. As we expected, the ATI Radeon X850 handled it flawlessly. We quit the game and started a full system scan running in Norton AV which bumped CPU usage up to 75 per cent or so and then went back into Doom 3. Although it took a while longer to start up, gameplay was completely smooth and Norton's presence was completely undetectable. Once again, we quit Doom 3 and with Norton still running we set iTunes to transcode two albums' worth of MP3 files to AAC format. We opened Doom 3 up again, leaving Norton and iTunes running in the background, and the gameplay continued to be superb. It provided us with a computing experience that we have never had before. Very, very impressive, so just remember, the benchmarks don't give the full story.

Verdict

Would we buy a brand new PEE-based machine? Probably not, but that's the answer we give with most new technology. We said it about Serial ATA, PCI Express, DDR memory and DDR2 memory, and now we're saying it about dual-core processors - ultimately, the price premium for the Extreme Edition will be too high.

That said, dual- and multi-core processors are definitely the way forward. Unlike Hyper-Threading, using a dual-core processor really is like adding a second CPU to your machine, rather than faking it up, as HT does. You really do have double the processor resources, all the time, not just when there happen to be some execution units going spare.

As is always the case, early adopters are going to pay a high price for small gains, but as Intel rolls out dual-core chips across all its platforms, software developers will be forced to code applications with multi-threading support. In the meantime though, anyone who's doing a significant amount of multi-tasking should see a significant performance boost right now.

How We Tested

After we had run SYSmark 2002 we used Windows Media Encoder 9 to encode a 416MB AVI movie file in WMV 9 format which finishes up 4MB in size. Next we used iTunes 4.7.1 to transcode a 22MB MP3 (128Kbps) audio file to AAC format at 128Kbps. Finally, we opened a 47.2MB TIFF in Adobe Photoshop CS 8 and resized it and increased the resolution so it grew to 658MB in size.

Pentium Extreme Edition 840 3.2GHz with HT enabled, 800MHz FSB on Intel 955XBK with 1GB 667MHz DDR 2

SYSmark 2002 Overall 361, Internet 515, Office 253
3DMark05 1.2 5729
Encoding Kitesurfing AVI to WMV 9 CPU load 70% 1m 25s
iTunes MP3 to AAC conversion 41s
Photoshop conversion 32s plus 47s to write the TIFF

Pentium D 3.2GHz (Pentium Extreme Edition 840 with HT disabled), 800MHz FSB on Intel 955XBK with 1GB 667MHz DDR 2

SYSmark 2002 Overall 369, Internet 530, Office 257
3DMark05 1.2 5721
Encoding Kitesurfing AVI to WMV 9 CPU load 80% 1m 44s
iTunes MP3 to AAC conversion 41s
Photoshop conversion 32s plus 47s to write the TIFF

Dual Xeon 3.2GHz/533MHz FSB/2MB L3 cache on Asus PCH-DL with 875P chipset and 1GB PC3200 memory

SYSmark 2002 Overall 319, Internet 436, Office 232
Encoding Kitesurfing AVI to WMV 9 CPU load 70% 1m 24s
iTunes MP3 to AAC conversion 37s
Photoshop conversion 30s plus 43s to write the TIFF

Pentium 4EE 3.4GHz 760 on D925XCV with 1GB 533MHz DDR2

SYSmark 2002 Overall 387, Internet 499, Office 300
Encoding Kitesurfing AVI to WMV 9 CPU load 90% 2m 11s
iTunes MP3 to AAC conversion 37s
Photoshop conversion 28s plus 39s to write the TIFF

Review by

Recent Reviews

Dell Latitude D410 sub-notebook
HP iPaq hx2750 PocketPC
Sapphire Radeon X800 PCI Express
Buffalo TeraStation Network-Attached Storage
Creative MuVo V200
Creative MuVo Micro N200 MP3 player
Fossil Wrist PDA FX2008