Original Link: https://www.anandtech.com/show/818



In the nine months since its introduction, the Pentium 4 has increased in clock speed over 40%.  Today Intel is making a tremendous step forward in the eyes of the public as the first microprocessor manufacturer to offer an x86 CPU clocked at 2.0GHz.  In the eyes of the AnandTech readers, we know that today’s launch surrounds no architectural improvements to the Pentium 4 core but is more marketing than anything else. 

Over a year ago AMD beat Intel to reaching the previous highly marketable 1GHz clock speed.  While both the AMD Athlon and Intel Pentium III hit 1GHz within a matter of days of each other, only the Athlon was readily available by the end of that month.  Fast forwarding to the present day, in the same amount of time that Intel has been able to achieve a 40% increase in clock speed AMD has been able to take the Athlon from 1.2GHz to 1.4GHz, an increase of 16%.  Is the AMD Athlon inferior to the Pentium 4?

Absolutely not.  The above comparison is mainly to show you the futility of clock speed based performance comparisons.  When dealing within a particular processor family such as the Pentium 4 or the Athlon, clock speed comparisons can give you an idea of relative performance; there is no doubt that a Pentium 4 running at 1.8GHz is faster than a Pentium 4 running at 1.6GHz.  When making cross-family processor comparisons, clock speed isn’t the only way to look at the performance picture.

How much work the processor can do in a single clock cycle, measured in Instructions Per Clock (IPC), matters just as much as clock speed.  At the same time, measuring a processor’s performance in IPC wouldn’t make much sense either since a CPU capable of an average IPC of 5 instructions per clock yet only capable of running at 50MHz wouldn’t be faster than a CPU capable of an average IPC of 1.5 instructions per clock yet capable of running at 1GHz.  The combination of IPC and clock frequency determines the true performance of the CPU.

From our previous articles on the Pentium 4 architecture, we explained that the Pentium 4 is not able to perform as many instructions per clock as an Athlon.  The Pentium 4 also relies on proper usage of more elegant features such as its Trace Cache and SSE2 in order to reach peak performance levels.  Mainly the difference in IPC renders clock speed comparisons between the Pentium 4 and Athlon useless.  This ends up working as a double edged sword for AMD and Intel.  On the one hand, it allows AMD to say that their processor is faster than the Pentium 4 (on a clock for clock basis).  On the flip side however, it also brings into question whether the higher clock speeds allowed for by the Pentium 4’s architecture will allow it to eventually outperform the Athlon.  Remember, you can’t just look at IPC or clock speed; the combination of the two makes up the performance of the processor. 

If the majority of the market understood this concept then clock speed wouldn’t carry as much weight as it currently does.  Unfortunately, communities like AnandTech only make up a small percentage of the total market.  For the majority of the individuals out there that don’t have the time or the resources to research, CPU clock speed is the primary determining factor in the performance of two systems.  With the Pentium 4 now running at 2.0GHz, AMD is clearly at a disadvantage when it comes to the clock speed race.  The task upon us today however, is to find out if AMD is at a disadvantage when it comes to the performance race.

How hard is it to hit 2.0?

The core architecture of the Pentium 4 processor has not changed since its introduction last November.  Now running at 2.0GHz, the basic specifications remain the same.  We’ll spare you the detailed description but if you are interested in the NetBurst Architecture behind the Pentium 4 then you've got a couple of options. Here's Intel's take on it but we strongly suggest that you have a look at this quick one page description of the micro-architecture from our Pentium 4 1.7GHz Review.

If you haven’t already found out, one of the major features of the Pentium 4’s architecture is its Hyper Pipelined Technology.  This is nothing more than marketing spin on the fact that the Pentium 4 has a very long pipeline.  A long pipeline itself doesn’t translate into performance as the name would imply, rather there are actually quite a few downsides to having a long pipeline.  Branch mis-predict penalties are very high (although reduced by the use of a good branch prediction unit and a trace cache) and un-optimized code can perform quite poorly as the average IPC when dealing with a longer pipeline is considerably lower.  The benefit of this is that because the processor is doing less per clock, it can run at higher clock speeds.

With our 1.8GHz Pentium 4 we were able to run at 2.0GHz without any additional cooling, indicating that Intel’s journey to the 2.0 mark wasn’t that difficult.  The CPU uses a 20.0x multiplier to run at its 2.0GHz frequency; remember that the Pentium 4’s FSB is quad-pumped meaning that although data is transferred four times per clock, the true operating frequency of the bus is still 100MHz while offering data transfers equivalent to that of a 400MHz bus. 

Although the Pentium 4 is fundamentally unchanged, Intel did introduce a new socket for the Pentium 4 2.0 launch.  The original Pentium 4 used a 423-pin Pin Grid Array (PGA) socket interface;  while the 1.9GHz and 2.0GHz Pentium 4s being announced today will be available in Socket-423 versions, they will also be introduced in a new form factor as well: Socket-478.

The main difference between Socket-478 and Socket-423 is that the new socket features a much more densely packed arrangement of pins known as a micro Pin Grid Array (µPGA) interface.  This allows the CPU to be much smaller and the space occupied by the interface socket on the motherboard to decrease as well.  The entire Pentium 4 line will be available in both Socket-423 and Socket-478 versions, although eventually all of the processors will be Socket-478. 

We have heard from motherboard manufacturers that it should not be too difficult to make a Socket-478 to Socket-423 converter for owners of older Pentium 4 motherboards.  Newer Socket-478 motherboards will still primarily be based on Intel’s 850 and 845 chipsets so there isn’t much of a reason to upgrade boards other than to gain CPU support for Socket-478 processors.  Now do you see why we didn’t recommend going down the Pentium 4 path early on?

From left to right: Socket-423 Pentium 4, Socket-603 Xeon, Socket-478 Pentium 4


Click to Enlarge


Click to Enlarge

This isn’t the first time Intel has introduced a microPGA CPU interface.  If you’ll remember to the Intel Xeon launch of this past May, the Intel Xeon processor features a 603-pin interface socket that is no longer than the Pentium 4’s old Socket-423.  The main reason being that the pins are much more densely packed in a microPGA package. 



Very Aggressive Pricing

With a tremendous amount of pricing pressure from AMD, Intel was finally forced to drop prices on the Pentium 4 line to much more competitive levels.  In July we looked at Intel’s 1.8GHz Pentium 4 and did a quick pricing comparison based on street/Intel pricing at the time:

Pentium 4 1.3GHz - $162
Pentium 4 1.4GHz - $169
Pentium 4 1.5GHz - $233
Pentium 4 1.6GHz - $294
Pentium 4 1.7GHz - $335
Pentium 4 1.8GHz - $562

Today their pricing is as follows:

Pentium 4 1.3GHz - $133
Pentium 4 1.4GHz - $133
Pentium 4 1.5GHz - $133
Pentium 4 1.6GHz - $163
Pentium 4 1.7GHz - $193
Pentium 4 1.8GHz - $256
Pentium 4 1.9GHz - $375
Pentium 4 2.0GHz - $562

What’s happening is that the lower end line of the Pentium 4 family is getting replaced by some of the faster CPUs, thus driving the price down to a pretty low point.  One thing you’ll notice is that the Pentium 4 1.3 – 1.5GHz CPUs are all priced as high as the most expensive Athlon CPU at the time of publication (Athlon 1.4GHz). 

Intel has stated that if these price cuts do not have their desired effect then they will pursue another price cut later this year.  The price of the 2.0GHz part is still pretty high; this is in line with Intel’s history of charging a premium for their fastest CPUs.  AMD used to do this in the past however recently their pricing structure has become very aggressive, with the majority of their CPUs selling for under $100. 

The combination of very aggressive pricing and cheaper 4-layer motherboards make the cost of Pentium 4 ownership much less than what it once was.  The continuously decreasing price of RDRAM helps as well, although it has yet to achieve a price parity with DDR SDRAM.



Next Stop: Northwood

The 2.0GHz Pentium 4 will be Intel’s last Willamette based processor launch.  Around November you can expect to see Intel’s first 0.13-micron Pentium 4 based on the Northwood core debut at 2.2GHz. 

The 2.2GHz Northwood, as we reported months ago, will feature a 512KB L2 cache.  The performance improvements from a larger L2 cache will be particularly evident in large dataset professional applications, newer 3D games, and heavy office/IT usage patterns.  Northwood will also be available in a Socket-478 form factor and will work on current Socket-478 motherboards with a BIOS update.

The 0.13-micron manufacturing process will decrease the manufacturing cost of the Pentium 4 (although the doubling of the L2 cache size will reduce some of the cost savings) and enable cooler operation.  The cost savings won’t directly be transferred to the end user as the 2.2GHz Northwood is expected to debut at close to $600 in 1,000 unit quantities. 



The Board & Cooling

We tested on ABIT’s TH7-II RAID, their Socket-478 i850 based Pentium 4 motherboard.  The board has a very flexible SoftMenu III Jumperless setup to configure the CPU, memory and FSB settings.  Unfortunately with all retail/OEM Pentium 4 CPUs shipping in a clock locked state; there isn’t much configuration necessary.  The board ran fine during our test period but we will continue to test it and report on any problems that do arise.


Click to Enlarge

We used a GlacialTech Igloo 4300 cooler for our tests. The fan was relatively loud compared to Intel's retail heatsink/fan combo but it did the job. The 4300's attachment mechanism was superior that which was on Intel's retail heatsink.



The Test

Windows 2000 Test System

Hardware

CPU(s)

Intel Pentium 4 2.0GHz
Intel Pentium 4 1.5GHz
AMD Athlon-C "Thunderbird" 1.4GHz
AMD Athlon MP "Palomino" 1.2GHz
Motherboard(s) ABIT TH7-II RAID (i850) ECS K7S5A
Memory

256MB DDR266 Crucial SDRAM (Micron CAS2)
256MB PC800 Samsung RDRAM

Hard Drive

IBM Deskstar 30GB 75GXP 7200 RPM Ultra ATA/100

CDROM

Phillips 48X

Video Card(s)

NVIDIA GeForce3 64MB DDR (default clock - 200/230 DDR)

Ethernet

Linksys LNE100TX 100Mbit PCI Ethernet Adapter

Software

Operating System

Windows 2000 Professional Service Pack 2

Video Drivers

NVIDIA Detonator3 v12.41 @ 1024 x 768 x 16 @ 85Hz
VIA 4-in-1 4.32V was used for all VIA based boards



Business & Content Creation Performance

As usual, we start with Winstone 2001 to illustrate performance in a work and play environment.  First comes the work wth Business Winstone 2001:

Highly governed by the performance of the disk subsystem, Business Winstone 2001 gives us a good indication of performance in real-world business application scenarios since they are usually very disk limited.  In this case, the Pentium 4 2.0GHz is just about as fast as an Athlon 1.4GHz. 

While we would normally point out that all of these processors are overkill for regular business applications, with the transition to Office XP and Windows XP you can count on needing a bit more processing power (about 30 – 50%) to get the same performance levels as you are used to.

The picture is no different under Content Creation Winstone 2001.  The 1% performance difference between the Pentium 4 2.0GHz and the Athlon 1.4GHz is negligible.



Business & Content Creation Performance (continued)

SYSMark 2001’s Office Productivity suite is actually much more stressful on the CPU and platform in general than Business Winstone 2001.  This gives the Athlon 1.4GHz a 8.6% lead over the Pentium 4 2.0GHz.  The Athlon is still the better choice for Business/Office applications however with 512KB L2 cache the Northwood core may be able to tilt things in the Pentium 4’s favor.  Remember that these types of applications rely on being able to run primarily out of cache.

The Internet Content Creation tests under SYSMark 2001 provide a much different picture.  The Pentium 4 is clearly dominating in this test mainly because of the Windows Media Encoder portion of the benchmark.

Summing the two parts up results in a 12% lead for the Pentium 4 2.0GHz over the Athlon 1.4. 



Enterprise/IT Computing Performance

As basic office machines, all of the test beds completed the Office Bench suite without any performance issues.  Even the slowest platform performed respectably.

Cranking up the loading level a notch creates a good performance spread.

Truly stressing the system with the second loading level allows the Pentium 4 2.0GHz to outperform the Athlon MP 1.2GHz but still keeps a noticeable distance away from the lead Athlon 1.4.



3D Rendering, Raytracing & Video Encoding Performance

We know for a fact that 3D Studio MAX is very x87 floating point intensive which gives the Athlon the performance advantage over the Pentium 4 2.0GHz. 

A new addition to our performance measurement suite, CINEMA 4D’s Raytracing test closes the gap between the Pentium 4 2.0GHz and the Athlon 1.4GHz.  The two perform approximately the same.

Another newcomer to our test suite is MPEG-4 encoding using Flask.  We used the MMX iDCT algorithm for all of the platforms (on the Pentium 4 the SSE2 iDCT algorithm did not perform any better).  Using the 1.1 patched beta of v0.060 and the DivX 4.0 codec we ran an encode at 352 x 288 of a MPEG-1 source with de-interlacing enabled. 

Here we can see the enhancements of the Palomino core truly flex their muscle as the Athlon MP 1.2 is able to outperform all of the competing solutions including the higher clocked Athlon 1.4GHz part.  It won’t be much longer until the desktop Athlon gains the Palomino enhancements.



3D Gaming Performance

Always a strongpoint of the Pentium 4, Quake III Arena primarily shows off the benefits of the Pentium 4’s hardware prefetch.

A similar performance advantage is seen under DroneZ indicating the nature of future gaming performance to a certain degree.

…in spite of the aforementioned conclusions, AquaMark does not seem to favor the Pentium 4 at all.  The Athlon MP even comes out on top of the Pentium 4 2.0.



The future: 133MHz FSB?

Lately Rambus has been talking about yields doing so well on PC800 RDRAM that they are actually capable of running the modules at 533MHz or at PC1066 spec.  There is actually a bit of significance about that number since Intel has been toying with the idea of increasing the FSB on the Pentium 4 from 100MHz to 133MHz, resulting in an effective 533MHz FSB. 

Just to satisfy our own curiosities we used our unlocked Pentium 4 to obtain benchmarks at 20 x 100MHz (2.0GHz) and 15 x 133MHz (2.0GHz) to see the benefit a 133MHz FSB would have on the Pentium 4’s performance.  In order to keep this a controlled experiment we decreased the RDRAM multiplier to 3x which kept the RDRAM frequency at 400MHz (100MHz x 4 vs. 133MHz x 3) or PC800. 

We ran the entire set of benchmarks featured in this review and normalized the results to the regular Pentium 4 2.0GHz running at 20 x 100MHz. 

The results are actually quite impressive.  On average, performance improved 6% across the board.  In some cases such as under Office Bench and SYSMark 2001 the improvement was significantly higher than the average (12 – 14%) while in other cases such as under 3D Studio MAX and CINEMA 4D there were no tangible performance gains.  The test platform ran flawlessly using the 133MHz FSB meaning that it shouldn’t be hard to implement.  Unfortunately for Intel, it’s unlikely that we’ll see the Northwood core debut with 133MHz FSB support.  It will probably be 2002 before we see the first 133MHz FSB Pentium 4 processors.  The performance difference between the 100MHz and 133MHz FSB will only increase as the Pentium 4’s clock speed goes up meaning that the performance benefits will improve with higher clock speed Pentium 4s. 



Final Words

At 2GHz the Pentium 4 is even more competitive with the Athlon than it has ever been before.  If it weren’t for such a large price discrepancy between the two processors then the recommendation would be a tough one to make but for now the price to performance ratio of the Athlon continues to be superior to the Pentium 4.  If price isn’t a concern, then it’s worth noting there are some situations in which the Pentium 4 2.0GHz is noticeably more than what AMD’s current flagship can offer.  What you really must take away from this review is that the potential for the Pentium 4’s future is becoming more and more visible as the processor matures. 

The Northwood core will bring even more attractive performance to the Pentium 4 and a continually increasing clock speed before the year’s end.  While we seriously doubt Intel will introduce the 133MHz FSB this year, it would only make sense for Intel to eventually transition the Pentium 4 to a 133MHz FSB.  A 0 – 14% increase in performance across the board isn’t bad at all, not to mention what kind of an increase you’d expect with a similar boost in memory bandwidth as well.  Intel’s 850 chipset already works fine at this frequency as does VIA’s P4X266 chipset; the only thing that’s left is official support from Intel.

The performance race will be pretty competitive through the end of this year between AMD and Intel.  With AMD’s upcoming desktop Palomino processors they will be able to remain competitive with Northwood and the Pentium 4 processors announced today.  What will truly heat up is the marketing competition between the two companies as Intel has to convey that they are still the best, while AMD must fight that notion of clock speed equaling performance.

Log in

Don't have an account? Sign up now