by Ilya Gavrichenkov
10/31/2004 | 12:00 PM
During the past few months the situation in the CPU market has changed dramatically. Within a relatively short period of time there appeared new platforms, as well as new processor families: both, AMD and Intel have already introduced a few new solutions. However, against the background of high AMD’s activity, Intel’s actions look more like weak attempts to retain the existing market share, which are evidently less efficient than the rival’s. However, it is still too early to conclude that Intel suffered an unpleasant defeat in the new round of competition. In fact, the next year will dot all i’s, and in the meanwhile the company decided to sit on the fence and reconsider some of its future plans, moving the focus from the clock frequency growth to introduction of new features and extending new platforms’ functionality.
<%BANNER[article]%>Nevertheless, sometimes Intel continues announcing new processor models, which have already been put on the roadmaps. A week and a half ago we introduced to you two new AMD CPUs targeted for hardware enthusiasts and hardcore gamers with quite a bit of cash to spend. Today Intel is proud to introduce to us their response to that launch: one more Pentium 4 Extreme Edition processor with a slightly higher core clock frequency and faster system bus. However, even keeping this fact in mind we still have every right to say that AMD CPU line is somewhat fuller than that of Intel’s.
Unfortunately, there are no CPUs among the Intel’s offering, which would allow you to build a system priced between $1300-$1500. The top Pentium 4 on Prescott core, Pentium 4 560, has become too inexpensive for that, while Pentium 4 Extreme Edition processors cost more than twice as much as the top Pentium 4 CPUs. This gap in the product range was supposed to be filled with the Pentium 4 3.8GHz and 4.0GHz, scheduled for the end of the year. However, now Pentium 4 580 working at the actual 4.0GHz has been cancelled and there is only Pentium 4 570 working at 3.8GHz due in mid November, which will have to compete with the new Athlon 64 3800+ and Athlon 64 4000+ processors from AMD.
As far as the new Pentium 4 Extreme Edition is concerned, its clock frequency has grown up to 3.46GHz due to the increase in the Quad Pumped Bus frequency to 1066GHz. This CPU should become Intel’s fastest solution for the period until Q1 2005, when Pentium 4 Extreme Edition with 3.73GHz clock frequency comes to replace it. Note that this is a pretty hard task, actually, because the previous Pentium 4 Extreme Edition 3.4GHz worked at a slightly lower clock frequency, but was designed in Socket478 form-factor, i.e. could be used in faster platforms based on i875P chipset. The new Pentium 4 Extreme Edition 3.46GHz will be available only in LGA775 form-factor and hence will have to run only in systems with PCI Express x16 graphics and DDR2 SDRAM, which still yield a little bit in performance to their predecessors.
However, we will be comparing the performance of the new CPU with the solutions available in the today’s market later in this article, and in the meanwhile let’s try to find out what the new CPU is like and where it belongs according to Intel’s product positioning.
The new Intel CPU doesn’t have anything surprising about itself. It is based on the Gallatin core, which you should be very well familiar with and which is used in Pentium 4 XE processor family. It differs from the predecessors by slightly higher working frequency and higher bus frequency. This is also indicated clearly by the diagnostic utilities:

The new Pentium 4 XE 3.46GHz is based on the good old 0.13micron core, which is developed from the already retired Northwood core. In fact, this is the last Intel desktop CPU produced with the old 130nm technological process. As a result, this processor doesn’t boast any new technologies, which have been the talk of the town lately, and which have been successfully implemented in the competitor’s solutions. Here we are first of all talking about 64bit extensions (EM64T), Intel Enhanced SpeedStep technology intended to reduce heat dissipation, and Execute Disable Bit technology, which ensures better operation system protection against harmful software attacks. All these innovations are to appear in Pentium 4 XE only when they move to the new Prescott-II core, which should happen in the beginning of next year.
This way, the only innovation the new Pentium 4 XE 3.46GHz can boast is faster 1066MHz bus, which should somewhat speed up the processor, according to Intel engineers. Note that according to Intel’s current plans, the 1066MHz processor bus will be the prerogative of Pentium 4 XE processors only for the end of this and the entire year 2005. Other Intel’s processor families, such as Pentium 4 5XX, Pentium 4 6XX with 2MB L2 cache and even Pentium 4 7XX with dual-core architecture will support only 800MHz bus all next year. This way, if 2MB L3 cache is to be the distinguishing feature of the 2004 Pentium 4 XE processors, then next year this role will be taken by the fast 1066MHz bus.
Summing up everything I have just said, let me list the formal technical specifications of the new CPU:
Note that the heat dissipation of the new Pentium 4 XE 3.46GHz doesn’t exceed that of the faster Pentium 4 processors based on Prescott core, even despite the relatively outdated 130nm production process they use. It also means that the new Pentium 4 XE 3.46GHz will be able to work in the same LGA775 mainboards as Prescott CPUs. The new CPU requires only one thing from the boards: hardware support of the 1066MHz bus and Gallatin core support in the BIOS.
Since Pentium 4 XE 3.46GHz processor is the first CPU to support 1066MHz system bus, we had to use a mainboard based on a new chipset, which allows clocking the bus at this frequency. Intel decided to take the easiest way and didn’t develop any special sets of core logic for its new Pentium 4 XE 3.46GHz processor. The next generation chipsets, namely Lakeport and Glenwood, supporting 1066MHz bus and the whole bunch of other features are to come out only in the middle of next year. In the meanwhile, Intel adapted its already existing i925X Express chipset for the needs of the new Pentium 4 XE 3.46GHz processor. The new chipset version is called i925XE Express and its only difference from the predecessor by the new faster system bus support. Intel didn’t even include DDR2-667 SDRAM support in the specs of the i925XE Express, although unofficially this chipset should be able to work with this memory when the bus frequency is set to 1066MHz.

Note that speeding up the bus between the CPU and the chipset to 1066MHz, resulted into higher bandwidth of the bus between the processor and the system memory. Namely it grew from 6.4GB/s to 8.5GB/s. Moreover, when the system bus is clocked at 1066MHz and the DDR2 memory is working at 533MHz, the CPU bus and the memory bus work synchronously, which as a rule allows minimizing the chipset North Bridge latencies.
i925XE chipset is fully compatible with i925X and doesn’t require any new PCB layouts or wiring. Therefore, the mainboards based on the new chipset, which was released only because of one single CPU, will hardly take long to appear in the market. However, despite the minor differences from the i925X, the overclocking fans will be able to benefit from the new core logic a bit more. Due to the official support of 1066MHz bus, mainboards on i925XE Express should definitely be much more stable when working with the bus frequency exceeding 800MHz, than any of the older LGA775 solutions. So, i925XE Express has every right to be considered a great solution for hardware enthusiasts who are willing to spend quite a bit of money on the new Pentium 4 XE 3.46GHz. At the same time, it is also an excellent overclocker’s choice, which has already been proven by the first benchmark results for the mainboards based on the new core logic set.
To prove that the i925X and i925XE are identical solutions, let me give you the following example. Intel’s own mainboards, such as Desktop Board D925XCV2 and Desktop Board D925XECV2 based on them are built on the same PCB and even use the same BIOS code.

However, if you try to install the new Pentium 4 XE 3.46GHz into the older Intel D925XCV2 Desktop Board based on i925X chipset, you will hardly succeed in making the system work: i925X based mainboard will not even start with the new processor, even if you reflash the BIOS supporting it from the newer D925XECV2. So, I think I can conclude that there are still some hardware differences between the i925XE and the older i925X.
We tested the new Intel Pentium 4 XE 3.46GHz with the new i925XE Express chipset in two steps. Before we compared the performance of this bundle with the solutions offered by the competitor, we decided to estimate the performance gain provided by the faster 1066MHz system bus. Therefore, we had to include a bit more benchmarks and use more platforms to run the tests of the kind thoroughly enough.
Our test platforms were built with the following hardware:
We ran all tests in MS Windows XP SP2 operation system with the installed DirectX 9.0c pack. The testbeds were configured to ensure maximum performance. Note that we increased the Cycle Time (Tras) timing of Athlon 64 to 10, because according to our experience, the memory controller of Athlon 64 processors works more efficiently in this case than in case this timing parameter is set to the minimal possible value of 5.
The first part of our test session will be devoted to discovering what positive effect results from moving the Pentium 4 Extreme Edition processors to 1066MHz bus. Since the faster bus can appear the only difference between future Pentium 4 XE processor and the Pentium 4 6XX family next year, knowing these numbers could be quite helpful for proper evaluation of future prospects of the Pentium 4 Extreme Edition processors. At the same time we are also interested in another aspect. Since the transition of Intel processors to the LGA775 form-factor and i925/i915 chipsets resulted into a slight performance drop compared with what we had on older platforms based on i875/i865 chipsets, it will also be en interesting question if the 1066MHz bus can actually help to solve this issue. Since in case of 1066MHz bus, the DDR2-533 SDRAM works synchronously with the processor bus in systems built on i925XE. Theoretically, it should speed up the system, so that in the end it turns out faster than an i875P based platforms with DDR400 memory.
For our testing needs we assembled three systems: one based on i925X chipset with 800MHz bus, one based on i925XE chipset with 1066MHz bus, and one on i875P chipset with 800MHz bus. In all three cases we used Pentium 4 XE processors working at 3.2GHz core clock (the samples we had at our disposal allows adjusting the clock frequency multiplier). In other words, in systems supporting 800MHz bus the processor frequency was set as 16 x 200MHz, while in the system supporting 1066MHz bus this setting looked as 12 x 266MHz.
First of all we took a look at the synthetic benchmark results, which illustrated the memory subsystem performance in all three cases:
i925X | i875P | i925XE | |
SiSoft Sandra 2004, RAM Int Buffered Bandwidth, MB/s | 4699 | 4956 | 5396 |
SiSoft Sandra 2004, RAM Float Buffered Bandwidth, MB/s | 4699 | 4956 | 5393 |
ScienceMark 2.0, Memory Bandwidth, MB/s | 4008.1 | 4351.34 | 4536.27 |
ScienceMark 2.0, Memory Latency, cycles | 269 | 240 | 261 |
ScienceMark 2.0, Memory Latency, ns | 84.06 | 75 | 81.56 |
Both synthetic benchmarks we used, SiSoft Sandra 2004 and ScienceMark 2.0 unanimously show that the transition to faster 1066MHz bus really does result into a noticeable practical increase in the data bandwidth of the bus between the CPU and the system memory. Moreover, in the system with 1066MHz bus and dual-channel DDR2-533 SDRAM the practical bandwidth is higher than the memory bandwidth in an i875P based system. In other words, when we speak about the new systems with DDR2-533 memory we can state that they do ensure higher practical bandwidth of the memory subsystem. However, when we measure the latencies of the memory subsystem, we find out that DDR2-533 cannot compete with DDR400 SDRAM at all here. As we see, even faster system bus and synchronous working mode do not help the system with DDR2 memory to outperform the one based on i875P chipset.
Here I have to make one thing clear first. Namely, you should understand that the practical latency of the memory subsystem, just like its bandwidth, depends not only on the type of the system memory, but also on its timings. DDR400 SDRAM has already reached the top of its potential here: the widely spread memory solutions with the 2-2-2-5 timings is the top, there will be no memory of this type with more aggressive timings out there. And in case of DDR2-533 SDRAM, the evolution has just started, so we can expect the memory of this type to go far beyond the currently common 4-4-4-11 timings settings in the future. Moreover, the engineering samples of the DDR2-533 memory with 3-3-3-8 timings do already exist. So far we will not run all the tests for the systems featuring this type of memory, because it hasn’t yet gone to the mass market. However, we will still offer you a few most essential results obtained on an i925XE based platform with 1066MHz bus and DDR2-533 SDRAM supporting such aggressive timings:


Well, the conclusions are evident. The use of DDR2-533 SDRAM memory with more aggressive timings can significantly increase the memory subsystem performance by raising its bandwidth on the one hand and reducing the latency on the other. It means that DDR2-533 SDRAM with 3-3-3-8 timings has every chance to speed the LGA775 systems up to the level of i875P based ones. However, this is still a future prospect, and in the meanwhile let’s check out the results we can obtain in real applications if we use the DDR2-533 SDRAM with 4-4-4-11 timings available in the stores today.

As we have expected, even though the use of faster 1066MHz bus on the LGA775 platform does ensure a certain performance gain, this gain can still be not enough to obtain performance comparable with what the i875P based system would show.



However, it would still be incorrect to announce the defeat of the i925X/i925XE platform to the i875P based predecessor. In a few tasks, which are critical to the memory bus bandwidth, 1066MHz bus does guarantee sufficient performance gain, which cannot be compensated even by the low memory subsystem latency of the i875P based platform.
Here are more detailed results for a few benchmarks from the PCMark04 testing set, which will show you more clearly what type of tasks really do benefit from faster system bus:
i925X | i875P | i925XE | |
File Compression | 5.43 | 5.51 | 5.43 |
File Encryption | 52.05 | 49.46 | 52.11 |
File Decompression | 38.11 | 37.64 | 38.25 |
Image Processing | 14.63 | 14.37 | 14.67 |
Virus Scanning | 2754.72 | 2713.31 | 2843.78 |
Grammar Check | 2.35 | 2.35 | 2.40 |
File Decryption | 87.60 | 75.75 | 91.19 |
Audio Conversion | 2833.21 | 2834.79 | 2836.41 |
Web Page Rendering | 6.56 | 6.61 | 6.63 |
WMV Video Compression | 57.67 | 59.36 | 58.06 |
DivX Video Compression | 66.58 | 67.59 | 66.73 |
Physics Calculation and 3D | 202.02 | 205.03 | 202.87 |
Graphics Memory - 64 lines | 2706.06 | 3046.53 | 2705.82 |
File Compression | 5.39 | 5.78 | 5.76 |
File Encryption | 48.55 | 51.82 | 50.53 |
File Decompression | 38.03 | 38.55 | 38.53 |
Image Processing | 14.66 | 14.73 | 14.65 |
Grammar Check | 4.86 | 4.92 | 4.91 |
File Decryption | 79.18 | 92.70 | 92.83 |
Audio Conversion | 2830.35 | 2836.76 | 2844.90 |
WMV Video Compression | 58.04 | 56.92 | 58.59 |
DivX Video Compression | 66.44 | 67.71 | 66.77 |
As we can see there is a relative parity between the i875P based platform supporting 800MHz bus and DDR400 SDRAM and i925XE Express based platform supporting 1066MHz bus and DDR2-533 SDRAM. Both systems manage to win almost in the same number of applications, depending on the fact what parameter of the memory subsystem (latency or bandwidth) is very critical for the algorithm of each test.




In games the situation is even more evident: i925XE platform is completely defeated by the i875P based platform, even though the new platform supports faster system bus and the memory subsystem with higher bandwidth.
Summing up this mini test session I would like to state one thing clear: LGA775 systems based on i925/i915 chipsets remain slower than the i875P based ones. This situation wouldn’t change even if we speed up the system bus to 1066MHz and use the new i925XE chipset. DDR2-533 SDRAM still features pretty high latency, which causes all platforms using it to lose in tests. However, things are not that dramatic yet for LGA775. The arrival of DDR2-533 SDRAM with 3-3-3-8 timings can change the situation completely. Although, no one knows yet when these memory modules will appear out here in mass quantities.
The 1066MHz bus turned out to be no panacea here. Just this bus itself ensures very low performance improvement, which doesn’t exceed 1%, even though only in case of 1066MHz bus all contemporary LGA775 chipsets work in the synchronous mode. Of course, it should be the fact that i925/i915 chipsets have been initially optimized for work with 800MHz bus. So, in i925XE they simply had to speed up the system bus, in order to ensure at least some slight performance improvement of the new Pentium 4 Extreme Edition processors, because their clock frequency potential has already been completely exhausted and the 130nm Northwood core didn’t allow any further clock frequency increase. That is why in the next generation chipsets aka Lakeport and Glenwood the 1066MHz bus may turn out much more efficient.
The second part of our today’s test session is devoted to comparing the performance of the new Pentium 4 Extreme Edition 3.46GHz processor with that of the competitor solutions from AMD and with that of the predecessors: Pentium 4 Extreme Edition 3.4GHz (in LGA775 form-factor) and Pentium 4 560 based on Prescott core working at 3.6GHz clock frequency. I would like to stress right away that you shouldn’t expect the new Pentium 4 Extreme Edition 3.46GHz to work wonders. As we have already seen above, the performance gain resulting from the use of faster 1066MHz system bus can make no more than 1%. Besides, its clock frequency is only 1.7% higher than that of the predecessor. So, since the results of Pentium 4 Extreme Edition 3.46GHz appeared quite close to what the Pentium 4 XE 3.4GHz showed, there is nothing to be surprised with. Anyway, let’s get down to actual numbers now.
The average performance in “general” applications of different kinds was tested with the help of SYSMark 2004 testing set.






Here the new Pentium 4 XE 3.46GHz runs faster than its predecessor. However, it still appears not enough to indicate the advantage of the new CPU over the regular Pentium 4 on Prescott core. Higher clock frequency of Pentium 4 560 allows this CPU to run faster in most cases. Also keep in mind that the new Pentium 4 XE 3.46GHz loses in SYSMark 2004 to its No.1 competitor: AMD Athlon 64 FX-55.
Everything I have just said above is nicely proven by the average performance diagram for SYSMark 2004:



We see a similar picture in PCMark04 test, too. Again Pentium 4 XE 3.46GHz is only 1%-1.5% faster than Pentium 4 XE 3.4GHz, but at the same time it falls behind Prescott based CPU. So, we can say that in “general” applications measured by SYSMark 2004 and PCMark04 benchmark sets, Pentium 4 XE 3.46GHz turns out hardly worth the money they ask for it. However, PCMark04 also showed that Pentium 4 XE 3.46GHz still manages to outpace the competitor from AMD, Athlon 64 FX-55, which doesn’t support any Hyper-Threading technology actively involved in this benchmark.

The memory subsystem test from PCMark04 set shows that increase in the system bus speed of the new Pentium 4 XE 3.46GHz to 1066MHz has positively affected the bandwidth of the bus between the CPU and the system memory. So here Pentium 4 XE 3.46GHz finally manages to outperform Pentium 4 560, falling real short behind Athlon 64 FX-55 with the integrated memory controller onboard.

As we see, the situation turned out somewhat different in gaming apps, where 3DMark 2001 actually belongs, too. Large L3 cache of the Pentium 4 Extreme Edition processors allows them to show better results than the top Prescott based solutions do. However, even faster system bus and the clock frequency of Pentium 4 XE 3.46GHz do not allow this CPU to outrun the top Athlon 64 CPU models.


The similar situation can be observed in 3DMark03 test set.


In the new 3DMark05 test package Intel CPUs look somewhat better: the total score of all LGA775 platforms tested today appears higher than that of the AMD platforms due to faster PCI Express x16 graphics interface. However, the processor test from this package still awards AMD Athlon 64 FX-55 with the winner’s laurels, while the new Pentium 4 XE 3.46GHz again appears slower than Pentium 4 560 working at 3.6Ghz and supporting 800MHz bus.






In all 3D games we use for testing purposes Intel processors yield to AMD solutions. However, the advantages of the contemporary AMD platform in gaming applications are indisputable. As for the results of Pentium 4 XE 3.46GHz, this CPU appears about 1-3% faster than the predecessor, Pentium 4 XE 3.4GHz, and about 5-10% faster than Pentium 4 560 (the only exception is Quake3 game where Pentium 4 XE 3.46GHz manages to perform 19% better than Pentium 4 560).

Although a slight clock speed increase managed to make the new Pentium 4 XE 3.46GHz run faster compared with the predecessor, AMD and Intel processor retain certain parity during MP3 audio encoding.




Nothing outstanding happens during video encoding, too. The new CPU appeared slightly faster than the previous model, although Pentium 4 560 based on Prescott core is still faster in this type of tasks. It must be higher working frequency as well as architectural improvements and SSE3 commands support that tell here. So, if you do a lot of video encoding, then it hardly makes much sense to you to go for an expensive Pentium 4 Extreme Edition processor.

Despite faster system bus, Pentium 4 XE 3.46GHz processor is just a little bit faster than the predecessor in WinRAR. Well, there is nothing to be surprised with actually: data archiving is very critical to the latency of the bus between the CPU and system memory, rather than to its bandwidth. By the way, Prescott based CPU from Intel performs much faster in this test, which can be certainly explained by more efficient hardware data prefetch algorithms implemented in this core. However, even Pentium 4 560 is considerably behind Athlon 64 processors from AMD, which feature built-in memory controller working with DDR400 SDRAM.

During the antivirus check tests by one of the most widely spread applications, the winner’s laurels go to AMD Athlon 64 processor family.

Despite the 2MB L3 cache, one of the most popular graphics editors doesn’t favor Pentium 4 Extreme Edition processors, so that they lose to the regular Pentium 4 560 as well as to the competitor’s Athlon 64 processors.

We can see a similar picture in a popular Mathematica 5 package. However, this is not surprising at all, I should say. Computational tasks have always been a trump of AMD CPUs, even since the times of K7 solutions.
Besides Mathematica 5 test set, we also used for our testing purposes a popular Matlab 7.0 application with the benchmark built into it by default. This benchmark consists of 6 subtests. Here they are:
As a result of this benchmark we obtain a total performance score, which you can see on the graph below for all our testing participants.

Besides the overall score we can also show you the time it took our testing participants to complete each of the 6 subtests (of course, the lower numbers stand for better performance):
Pentium 4 560 | Pentium 4 XE 3.4 | Pentium 4 XE 3.46 | Athlon 64 3800+ | Athlon 64 4000+ | Athlon 64 FX-55 | |
LU | 0.359 | 0.328 | 0.313 | 0.34 | 0.331 | 0.31 |
FFT | 0.328 | 0.422 | 0.406 | 0.281 | 0.28 | 0.27 |
ODE | 0.313 | 0.281 | 0.281 | 0.2 | 0.2 | 0.18 |
Sparse | 0.484 | 0.422 | 0.406 | 0.451 | 0.441 | 0.421 |
2-D | 0.438 | 0.422 | 0.407 | 0.28 | 0.271 | 0.261 |
3-D | 0.219 | 0.219 | 0.219 | 0.2 | 0.19 | 0.17 |
As we see, AMD CPUs are beyond any competition in Matlab 7.0.


Visual C++ is another application where AMD processors show their strengths. Although Intel Pentium 4 XE works here better than the regular Pentium 4 processors, they are still unable to catch up with Athlon 64.




During 3D rendering Pentium 4 XE processors appear a little faster than their counterparts on Prescott core. However, the overall situation looks more in AMD’s favor, even though the new Pentium 4 XE 3.46Ghz processor improved the performance records.
Well, if you were expecting to get some significant performance improvements with the launch of the new Pentium 4 Extreme Edition 3.46GHz and i925XE Express chipset, then you are probably somewhat disappointed. The transition of the Pentium 4 XE CPUs to 1066MHz bus ensures a tiny performance gain, and the clock frequency of the top solution in this family grew up really insignificantly. As a result, everything we have already said about Pentium 4 Extreme Edition CPUs in our previous articles remains valid even after the launch of the new model in the family.
Pentium 4 Extreme Edition 3.46GHz is a crazily expensive CPU (the price at launch was set to $999) with pretty arguable performance advantages. Even the top model in the Pentium 4 Prescott family working at 3.6GHz - Pentium 4 560 – appears faster in quite a few applications. In fact, Pentium 4 XE boasts noticeable performance benefits in games and in computational tasks of various sorts. However, in all the applications where Pentium 4 XE 3.46GHz does outperform Pentium 4 560, AMD’s Athlon 64 FX-55 solution still turns out faster. This way we will not be able to claim that Intel’s newcomer would be a successful choice for this type of tasks. All in all, this doesn’t make our final verdict sound very promising: Pentium 4 XE 3.46GHz CPU can be of interest to real Intel hardware fans, who are ready to invest $1000 in the right to own a relatively rare and exclusive solution. So far, we do not see any real practical value of the new Pentium 4 XE 3.46Ghz CPU.
As far as the new i925XE Express chipset is concerned, this solution also doesn’t offer us anything really new. It is just a slightly enhanced version of the i925X core logic, which we are already very well familiar with. Therefore the new chipset boasts the same features and provides the same performance level (which is actually even lower than what we are still able to get from the good old i875P based platform). However, i925XE Express has quite a few promising prospects ahead. Mainboards based on it will become popular among overclocking fans (we will check it out first hand when the mass mainboards on the new i925XE Express appear in our lab). Moreover, the performance of i925XE Express based platforms can also improve when DDR2-533 SDRAM featuring more aggressive timings (for example, 3-3-3-8) comes out.