Foreword

There is not an issue in which opinions further apart and tempers more heat than in the comparison between Intel and AMD processors. Often this but too happy to taken simply unaware of the facts and programmable features of a candidate, but most insist on a sometimes emotional position that they represented by all means.

Exactly this to change and to bring some light into the darkness we looked closer so the top models of Intel and AMD in this review and compete against each other can be. This will concern not only the pure speed of processors, but also we want to share our experience in terms of stability and compatibility you. A consideration of the features as well among the mandatory points such as the question of whether AMD’s “performance rating”, although of AMD can explicitly not designed to pass on the performance of Pentium 4 be.

As we we deal in this review mainly to the history and the new features of the Pentium 4 and does not respond to the Athlon XP, I first put all readers of this review our AMD processor Roundup on the heart.

History

With the history of Intel have processors we us already in the last article to sufficient busy. However, we thereby deliberately neglected the Pentium 4, to take his history and what is new at exactly this point under the microscope.

Before we address but its features in detail, we deal first with the external development of the processor.

Since the Pentium-III by further increases in clock you showed only marginal performance gains, Intel with the Pentium 4 has developed a completely new processor that has to do hardly anything to its predecessors. Is he based NetBurst on a new architecture, called. But this later more.

As the Pentium 4 format produced by the way only in the FC-PGA and not like its predecessors in a slot version on the market came, he placed first of all new demands on power supplies and housing. Because he needs a very large heatsink to quickly enough from the CPU to transport the heat, you can attach him not only to the CPU socket but also on the mainboard. Would you stall him only at the base, always a danger that rips the socket in the transporting of the PC by the Board and thus destroyed the Board would be because of the high weight. To get this issue out of the way, Intel provided for four additional holes around the CPU socket on the socket 423, on the one called, screws the cooling body holder (Rentention module) directly to the housing. This screw options not offered the most Boardträger of the housing but at that time yet, so that the Board and housing manufacturers were forced to switch to Intel’s new ATX 2.03P1-Spezifikation. It specifies the exact location of the processor on the Board and allows such a uniform design of the housing. The required Rentention module provided at that time with Intel and the important method of fixing of the cooler no longer left the cooling body manufacturers.

But still not enough. The Pentium 4 placed new demands on the power supplies to be used, afterwards, however we go further in the course of the review, because they have lost even when the Socket 478 is not of importance. Certainly, the new requirements for the enclosure and the introduction of the Pentium 4 power supply facilitated not just because some manufacturers was somewhat delayed on the new train and some initially even tried to get around the new specifications. Meanwhile, all manufacturers but adhere to the new specifications. Nevertheless you should look for when buying a new housing or power supply unit is still on the term “Pentium 4 compatible”. With the i850, Intel was also fully on Rambus memory which was abschreckte by its higher price compared to SD and DDRRAM many customers also before a purchase of the Pentium 4.

But some just described properties belong to the past again. Everyone will be aware that Intel for Pentium 4 already has been a change of the base, of the Sockel423 on the Sockel478.

So you need a Pentium 4 suitable power supply now at the Socket 478 no longer mandatory. This depends however on the respective motherboard and whether the manufacturer requirements by working on Intel’s. Intel’s specifications which still dictate the 12 volt plug. The Rentention module has been changed and now represents a kind of border for the heatsink that is already associated with the motherboard and not more in addition must be screwed with the housing.

At the same time with the switch to the Socket 478 moved one ahead of the competition to the new 0, 13µm technology and ensures to operate beyond the 3 GHz limit.

But this is of course not everything that is different from the new Pentium 4 with Northwood core and Sockel478 the Pentium 4 Willamette core and Sockel423. We dedicate this is why in the next section the differences of these two processors and a comparison to the Athlon XP.

Overview

< div >

The table shows up very already the differences of individual processors, but it indicates a not at all for now which processor is the fastest. Because the best specifications use something is limited if not also the interplay of all these components to work correctly. Here, Intel’s 20 stage pipeline is enormously important, but it we enter later more accurately.

The table illustrates beautifully the differences between the Willamette and Northwood. In the course of the shrunken manufacturing process Intel turned to Additionally the traces of aluminum to copper because copper is the better Director. Also could Intel on the now smaller DIE area 512 KB rather than as so far at the Willamette 256 KB place L2 cache. Also the increase of transistors to explain 13 million, all of which can be attributed to the additional cache is by this 256 KB cache of Northwood. Otherwise, not much has changed besides a lower core voltage justified also by the new production process and thus also lower heat.

At least on paper Intel’s Pentium 4 has ahead technologically and technically a lot the Athlon XP. Of course, we highlight point of whether to this change in performance, in the course of the review.

Another related change directly to the processor, Intel has also made during the conversion of the production process. Intel used the now larger wafers, i.e. the wafer slices are now 300 instead of 200 millimeters in diameter. This has the advantage that almost twice as many cores on a wafer fit on the one hand and on the other hand the blending, logically results from a round wafer, is percentage significantly reduced at a larger wafer. Because Intel has switched the production process of 0 18µm on 0 13µm, the magnification of the wafer is double paying off. On an old wafer found yet 144 Willamette cores place, on a 300 mm wafer for this already just over 200. The new production process increases this number 484 Northwood chips once again at a maximum. The significantly better yield of a wafer reduces also the production costs, what sooner or later should be reflected in the prices. At the moment, Intel produced on 200 mm large wafers, but the switch is in full swing.

Sets a Northwood in addition to an Athlon XP falls immediately to, that of the Pentium 4 thanks to his lead in the manufacturing a lot is smaller and has a so-called Heatspreader that covers the processor core. But this more in our next section, the technology of the Pentium 4.

Technology

NetBurst architecture

First we turn to now the generic term for Intel’s new technology of the Pentium 4: the NetBurst architecture.

To cope with the demands of the fast-growing processor market in terms of speed, it is usual all 3-5 to develop a completely new processor architecture years at Intel. The architecture introduced recently by Intel was the P6 micro architecture, which in 1995 for the first time was with the Pentium Pro used. But just above the 1 GHz border this architecture pushed to their limits and Intel was forced to develop a new, what was done with the NetBurst architecture. The following block diagram and the DIE view give a brief overview of the individual components of the new architecture, on which we shall enter into further more.

But first a small insight into the identifiers on the Pentium 4, made by Intel. You provide information on the processor clock, the cache size, the front-side bus, the core voltage, and the country of origin. But everything else shows the following graphic:

Hyper pipelined technology

The new pipeline of the Pentium 4 is certainly the most fundamental and far-reaching at the same time in many ways change compared to his predecessors. The pipeline is twice as long as that of the Pentium III with 20 levels. The extended pipeline has the enormous advantage that she can do many tasks at once, and only through them the high clock speed of the Pentium 4 are possible. The Pentium III was technique with his 10 stage pipeline already at one GHz in the 0 18µm in the end, so the Pentium 4 started only at 1.4 GHz, although he was made at that time still in 0 18µm.

But first we should address only once more on the 20 stage pipeline.

As most can probably do much with the term “Pipeline” and the meaning, we look at first the basics of a wired in series and a two-cycle pipeline. We go to the simplification of five steps out, to be carried out, to fully complete an operation. Here, the following five points of importance are simplified:

• Prefetch: The next command is loaded from the cache.
• Decode 1: The x 86 commands are decoded in MikroOPs
• Decode 2: If more data is required to calculate it to load now
• Execute: The instruction is executed.
• Host back: The result is stored in the cache or system memory.

A CPU without pipeline works from these steps one at a time and starts again only when a command is fully completed. Therefore always four of the five departments in the inactive mode are located in our example, and only a scope works. This prevents a CPU with pipeline. All five sections work here, so that an unnecessary inactivity of areas is excluded. As soon as the first instruction has left an area, the next is loaded immediately. It always five instructions are at the same time in the editing.

A 20 stage pipeline has the Pentium 4, thus 20 instructions are available at the same time in the editing and all areas are constantly busy. Therefore there must be in any area of a delay, because otherwise all other areas will have to wait and the work is significantly impeded. But to do so, we come back again later.

The higher clock rates of the Pentium 4 are above all to reach the steps of each command were divided into smaller parts that can be edited individually faster than the complete command.

This course also has its disadvantages at a 20 stage pipeline. A command has to go through more parts than earlier, finally been processed to be, and this increased to total time in which an operation is complete. Precisely herein lies the reason why a Pentium 4 with 1 GHz would be slower than a Pentium III with 1 GHz in a software that is not specifically designed for the Pentium 4! The Pentium III can process each operation faster thanks to its only 10 stage pipeline as the Pentium 4 at same clock speed. For the Pentium III can reach but by far not the clock speeds of a Pentium 4 just just because its only 10 stage pipeline. To make the Pentium 4 to his debut not equal slower than its Pentium III performs counterpart, Intel has obtained probably first start him with 1.6 GHz.

This point is extremely important for the further understanding of the results, because he also explains why a Pentium 4 at the same time is not faster than an Athlon XP, used a mix of 10 and 15 of stage pipeline. Therefore we find always something questionable, it if is compared to a Pentium 4 with an Athlon XP of same bar. Here, you should already know that in this case the Athlon XP in most applications just front should have had because of its architecture. But also each must be clear that an Athlon XP can reach the clockrate of a Pentium 4 with its current architecture.

Intel and AMD go here two different ways. Intel must settle significantly higher clock rate on its Pentium 4 due to its architecture, in order to be able to handle as many operations at the same time, like an Athlon XP with a lower clock speed. AMD sets here more performance per clock and Intel on more power with more tact.

Advanced dynamic execution

To wrong predictions in the processing of the next instruction as possible not occur to let some advanced features Intel of 20 has added stage pipeline to the team. All these improvements were in the “advanced dynamic execution”-combined engine. She should prevent that the pipeline by missing instructions or by the processor device mis-predicted branches of programs stalled, what would enormously slow down the Pentium 4. While Intel has improved compared with earlier in particular, the algorithm of branch prediction and the “branch target buffer” increases. In the “branch target buffer” the ready jump commands are stored and later once again can be controlled by the processor. The buffer has been increased for the Pentium 4 on 4 KB. The Pentium III had only 512 bytes to compare “branch target buffer”. Due to the longer pipeline, an increase of the buffer has been but inevitable. The prediction is to be improved so by a third compared to the Pentium III. Fairs not be these values for us of course, so that we here are dependent on the information of Intel and you should perhaps enjoy it with some caution. All in all would have to meet the Pentium 4 overall but with a 95% probability a correct prediction.

Trace cache

Some of you will have wondered at our overview of the processors already on the data in the L1 cache. Now, we want to again bring light into the darkness.

The Pentium 4 is not like his predecessors and the CPUs of the competition on a shared L1 cache, which stores the data as well as the instructions, but in the L1 cache only the data, but no longer stores instructions. There is now only the trace cache available for the commands.

To understand the trace cache, you must first have knowledge on the functioning of a processor. So far, it was common that the commands in the designated area of the L1 cache is stored. But today’s processors no longer directly facing the old x 86 commands, they assume yet to ensure software compliance. Later, these commands must be decoded only, but only the x 86 commands will be kept. And here comes into the trace cache. He no longer stores the x 86 commands, but (so called) equal to the decoded micro operations (micro OPs). This course has an enormous advantage when again and again same commands are used. Because the decoded MikroOps has not been saved, but only the x 86 commands, these had to be decoded in MikroOPs again and again, they should be used again. The Pentium 4 can directly access the already decoded x 86 commands, so the MikroOPs, and thus saved the time-consuming way of further decoding.

Unfortunately Intel specifies no longer the size of the trace cache in KB, but only in maximum storable MikroOPs, which is actually the better but more difficult to compare method. 12,000 MikroOPs in the trace cache can be stored in the Pentium 4, were previously decoded.

Rapid execution engine

Also the beer (arithmetic logic units) for the calculation easier integer Intel is a new way. Overall the Pentium has three ALU units, but is only in one of a conventional. The other two work with the double CPU clock. On a Pentium 4 2.2 GHz, two of the three ALU units with 4.4 GHz work accordingly. This double computer clock has two significant improvements: on the one hand is the time of a calculation in itself and on the other hand, the ALU-latency time is significantly reduced. This is the acceptance time of new commands on a Pentium 4 1.4 GHz under the 0, 4ns border. A Pentium III with 1 GHz needs for this still 1ns.

The Pentium 4 should issue thus in the integer calculation from its competitors, if the changes to be read as nice on paper because all so work as they should. Have an impact on FPU performance (MMX, SSE) of the Pentium 4 the two with dual CPU clock of running beer not course, which we would immediately arrived at the next topic: SSE2.

SSE2

Intel has gone not the way to develop further its FPU (floating point unit), but has supplemented his previous SSE instruction set to 144 new instructions, so that has the Pentium 4 on the 70 old SSE and the other 144 new instructions. SSE2 is fully backward compatible with its predecessor. New is also that the 144 added commands for 128-bit integer and floating point math are designed with double-precision. SSE2 is by the way for “streaming single instruction multiple date extension 2″. Thus, the SIMD in the name stands for commands that are able to process more data at a time. Already Intel’s SSE technology has shown that this can bring a significant performance gain with it especially for games and other applications through proper support.

The new SSE2 technology course brings no speed advantage to old applications that are optimized for SSE.

Because thought but even AMD into their processors SSE2, Intel can have been so very wrong with the further development and a significant performance gain can be expected particularly for applications that are specifically optimized for SSE2.

Quadspeed

The Pentium 4 works only with a 100 MHz bus clock, however, a completely new “Quad data rate (QDR)” comes the use of technology that allows to send four packets, so that the bandwidth of a clock rate of 400 MHz corresponds to a single bar. You can compare this technique has long known DDR technology which facilitates physically only 133 MHz, but anyway, thanks to its two packets per stroke, a 266 MHz performance achieved rapid bus. Although the front side bus with 133 MHz is physically higher on the Athlon, thanks to Intel’s QDR technology, the Pentium 4 but achieved significantly higher bandwidth in the transmission to the chipset. It is the Pentium 4 at 3.2 GB/sec, at the Athlon is only 2.1 GB/sec. Of course, this higher bandwidth reduces the time in which the processor must wait for new data.

To not to give away this bandwidth in memory, Intel placed in the introduction of the Pentium 4 on PC800 Rambus, because only he was in the “dual channel” method can reach 3.2 GB/sec data transfer.

Now there are also chipsets for anti-communism and SDRAM, Rambus was still very much more expensive especially with the introduction of the Pentium 4 and abschreckte many customers at scale. At the moment, the price difference between DDR-RAM and Rambus is no longer given.

Heat spreader

For many users and hobbyists the “heat spreader find 4 used in the Pentium” certainly is a very pleasing fact.

It has two advantages: on the one hand it protects the core from outer damage which may result in the setting up of the CPU cooler, if this tilting, on the other hand he also has the task, to ensure as a better heat mouse Exchange and scatter to the cooler over the heat in the core.

First pictures of “hammer”, which also has a “heat spreader” show that AMD is not averse to this “heat spreader”. Apparently, AMD learned here from the past in the, but the one or the other crafters broke up his core when fitting the heatsink.

Temperature sensor

The temperature sensors of the Pentium 4 are another very pleasant for the user feature. The Pentium 4 offers two completely independent temperature sensors that can cause two different protection mechanisms.

The first protection mechanism of the Pentium 4 is the so-called “thermal monitor”. It is composed of a temperature sensor directly in the core, a signal (PROCHOT#), indicating if the processor has reached its maximum temperature, and a control mechanism (thermal control circuit TCC) which can affect the CPU temperature through the adjustment of the CPU cycles. Once the CPU therefore has reached a set fixed temperature, the PROCHOT# signal is activated, the TCC deploys and omits some clocks when the CPU. Around every second measure is usually omitted. This is of course to keep in mind that this has a direct impact on the performance. Is every second beat is omitted, to the performance be reduced. Much more important in this context also the CPU temperature decreases significantly, and the processor is protected from corruption will however. The fixed temperature value is again below, consists of the PROCHOT# signal and the processor will work again with full power.

The second temperature sensor is intended only for absolute emergency if the cooling fails and omitting several bars brings no success. As soon as the Silicon reaches a temperature of about 135° C, reacts to the Pentium 4 with the THERMTRIP# system bus signal and turns itself completely off. Because THERMTRIP# operates completely independently of the processor, it produces no bars. The signal remains active as long as until the engine taken from the power source and has been a complete reset.

And here we must commend big Intel. The protection mechanisms of the Pentium 4 work perfectly and the processor continues to run even passively cooled many hours, albeit with reduced performance. Horror videos as they showed up for the launch of Athlon XP and where you saw a smoking CPU, because the protection mechanisms are not attacked, belong to in the Pentium 4 in the past, when it comes to the failure of two temperature sensors, what but probably never should occur under normal circumstances.

Power consumption

Before we chase our test candidates by the benchmark course, we can first compete them in usage. That the new Northwood should have an advantage due to its smaller manufacturing technology and lower core voltage, should be clear from the front come in.

< div >

The table shows clearly that the consumption of the Pentium 4 with Northwood core below which an Athlons. In addition, you can see the savings in consumption by moving to the 0 13µm manufacturing technology also very beautiful. At same speed, the Pentium 4 with 2.0 GHz and Northwood-core needs whole 23 Watts less than its predecessor with Willamette core and 0 18µm technology.

At this point, we want to touch briefly on the temperatures of various processors identified by us. The following table shows the temperatures even during normal Windows operation and on the other after a game Quake 3 arena:

< div >

As long as the CPU is stable and no crashes occur due to overheating, it may be actually pretty indifferent to the user whether the processor is hot 28 ° C or 50 ° C. However, the technological progress of the Pentium 4 clear, which is manufactured in 0, 13µm and therefore significantly less heat produced will be here. At AMD, the conversion of the production process is currently in full swing and you may count on Athlon XPs manufactured soon with first in 0, 13µm.

Condition

What are the requirements of the Athlon XP to a motherboard provides, we have already clarified in our large AMD processor comparison. Also the Athlon XP 2000 + with an effective frequency of 1.66 GHz is no different. On the Pentium 4, we want to lose more words, however, at this point of a few. The most important of these is that Intel has introduced the new Pentium 4 with Northwood core with a long-term plan in the market. Already as the start of the first Pentium 4 1.4 GHz socket 423 and the Willamette core fell, it was clear for Intel itself and thus also the mainboard manufacturers that this socket 423 is only an intermediate step to the current socket 478. Intel is extremely generous with roadmaps, so that the manufacturers of motherboards at least six months before know they must reckon with which products within a specific time period. Precisely for this reason, each is also quasi Socket 478 motherboard for the old Willamette and the new Northwood Pentium 4 ready, at least as regards the required voltage of 1.5 volts. For proper recognition only a BIOS update is missing and has nothing in the way should be happiness. This is encouraged by the fact that the disparities in architecture between will mat and Northwood are extremely low.

Another point to consider 4 processors generally for Pentium and the already input shortly mentioned, would be the power supply to be used. The Pentium 4 placed new demands on the power supply and the regulator on the boards could only meet them so that provides Intel for 12-volt directly from the power supply for the supply of the regulator, rather than going as usual about the 5- and 3, 3-volt lines to provide. Intel also pretended the new ATX12V specification provides for a four-plug, which connects the regulator directly to the power supply and provides them always with at least 10 A and maximum 12 A. Therefore the power supply manufacturers were in demand, adapt their power supplies to these new specifications, want to sell their power supplies 4 customers even on Pentium once again. Additional should have 4 PSU Pentium still the power factor correction, to ensure as far as possible uniform voltage and power supply. The new Pentium 4 with Northwood core, this solution is actually superfluous, because the processors are developing not quite so much warmth with this core. Therefore, there is now also Pentium 4 motherboards that do without a 12 volt connector and operate the Pentium 4 only with the normal ATX (2 x 10). Intel of it but nachwievor in its specifications above. On mainboards where this plug is present, must it also, in this respect unless specified in the manual, connected to the power supply, because otherwise the Board can be damaged.

Finally, Intel makes special demands on the body itself. The processor giant this recommends an average body temperature below 40 ° C to ensure a peaceful and safe work. This recommendation applies only to the in-a-box Variant of Pentium4, bundled together with fan and heatsink. It is this but more a recommendation as to a respected commitment. The fact is that the processor even in warmer environments properly performed his duty. The additional holes for the Rentention module, as mentioned above, are again unnecessary when the Socket 478. The next two pictures show the layout of the Rentention module now used.

Overclocking

While the AMD processors are the friend of Overclocks, makes it considerably more difficult Intel just these fellows since ages. For something like the L1 bridges to the unlock of the multiplier for the Athlon, Athlon XP and Duron knows the Pentium 4 not. Rather, the multiplier in the production is firmly branded at Intel. Thus, you can elicit more performance Pentium 4 only by increasing the front side bus. This is however little point because bar is raised automatically also the PCI and the APG and thus runs outside the specifications. Each megahertz more strikes at the FSB is thus in an increasingly unstable system again. This is so tragic, because the 0, 13µm of produced Pentium 4 is a great overclocking potential. The combination of head-spreader and the Northwoord core, radiating heat significantly less when compared to its predecessor, the Willamette, prove to be dead right. 2.5 GHz would be in there without major problems due to the architecture, this however would have to raise the FSB clock on 114 MHz. The AGP would in this case 75.27 MHz (66 MHz standard) and the PCI 37,62 MHz (33 MHz standard) run. A more than shaky matter. All in all still is Intel also Pentium 4 faithful. The processors have much potential, which can be but not easy to use. True to the motto, stability is, remains only the purchase of a new processor, if the performance is no longer sufficient.

In addition to manipulating the front side bus, it is still possible to increase the core voltage of the processor, at a bar outside of the intended sizes the signal strength to increase and hence to ensure more stability. The outgoing of the processor voltage signal that the mainboard chipset delivers what voltage must be used for the processor, is encoded as follows.

< div >

As we see it is that a 4-bit signal through four pins on the back of the Pentium 4 output can be read off. This outgoing signal is actually binding for the motherboard, however circumvent the signal the most basic input/output system (BIOS) of the Board and make it possible to manipulate the processor voltage freely within the boundaries. But care to much tension can damage the processor. Also is the processor due to physical laws (Q = CF m² U) warmer with increasing voltage. This however only as information to the edge.

Known bugs

Compared to AMD Intel has written a certain transparency on the Hat itself has long been, as far as concerned functional errors in the own processors.

As a result, also the information that you may receive an error, are more detailed. Taking a look in the list of known errors in the Pentium 4, so you must document paper with a total of 61 to 49 pages errors to deal. This is a lot of wood, bearing in mind that the Athlon XP according to AMD has only 9 known bugs in the sum. A number of which you certainly can fight, on the one hand, no one knows how many architecture errors are still in the processor, and on the other hand, how the errors found were all publicly documented. How many errors may also always be there, none of the two producers lets say these issues as possible in the area. For this reason, there are from time to time new processor Steppings that you want to create a few problems from the world. The following table shows the current Steppings of the Pentium 4.

< div >

As can be seen quite clearly in this table, Intel brought a new Prozessorstepping in the market for a 200 MHz clock increase on average. While the Willamette in the B2 stepping wore still 57 erratas, so mistakes, in itself, there were only 38 pieces in the C1 stepping. A considerable improvement, which is however only a few noticeable during normal operation. For the latest Pentium 4 with B0 Northwood-core, to better distinguish to the Willamette as nB0 known, there are only 23, of which only five in future Pentium 4 versions to be removed by means of slight modification. The fact that Intel’s documentation only in the B and not A stepping starts clearly only all Pentium 4 from which A generation come from the Development Department, and were thus still not for sale released. There were incidentally almost 100 spots, which Intel had to take the old Pentium III. However, it was here not as successful as in the Pentium 4, because after a total of seven Steppings, there are still good 80 mistakes the Pentium-III with the final Tualatin core.

Test system

To compare the current processors of the two leaders came on the part of AMD a VIA KT266A system with 256 MB PC2100 DDR memory used. At Intel we decided to use a Rambus system with Intel i850 chipset. Thus, the currently fastest platform was used for both systems. The argument often inserted at this point, Rambus was much more expensive compared to DDR SDRAM, memory of Infineon is no longer, because for 256 MB you’ll pay not more than 130 euros in both variants. It otherwise looks, however, on the motherboard. The ASUS A7V266-E with RAID, used by us onboard the Intel D850MD, which came for the Pentium 4 used costs less than at a better facilities well 40 euros. The benefits of the Intel platform are here clearly in the effective throughput of memory, although the Intel Board used by us under no circumstances among its kind the fastest representatives. In how far exactly the Pentium 4 can inspire, we will see in the next few sections.

The most complete systems for the Pentium 4 are sold at the moment although with DDR-RAM and DDR-RAM will be most used also in the local restaurant, but we approach this review with the approach, that we want to first of all highlight what you can get with the faster Rambus from a Pentium 4. On the course, in the performance of existing differences to a DDR-RAM we will enter into system more closely in future reviews.

The usual us Windows XP Professional was used as the operating system. Functions such as system restore and automated system updates have been disabled. In order to create a normal working environment, we have worked as a quality settings for the system interface with smoothed fonts, and the default Windows XP user interface. All other Visual options were also enabled. As the video card driver for the Inno3D GeForce3 Ti500 we related 27.30 the detonator, which is further are introduced with the GeForce 4. To prevent issues now follows a complete comparison of the hardware used and the installed driver.

• Processor
  • Intel Pentium 4 2.2 GHz
  • Intel Pentium 4 2, 0a GHz
  • AMD Athlon XP 2000 +
  • AMD Athlon XP 1900 +
  • AMD Athlon XP 1800 +
• Motherboard
  • AMD platform: ASUS A7V266-E (KT266A)
  • Intel platform: Intel D850MD (i850)
• Memory
  • 256 MB PC266 CL2 DDR SDRAM
  • 256 MB of PC800 Rambus
• Graphics card
  • Inno3D tornado Geforce3 Ti500
• Peripherals
  • Pioneer 16 x DVD
  • IBM IC35LC040
  • 10 / 100 Mbps Ethernet card
• Driver versions
  • nVidia detonator 27.30
  • Via 4in1 4 37a
  • Intel inf driver + IAA 1.1.2
• Software
  • Windows XP Professional

Benchmarks

Assessing the performance of the test scenario used already in the AMD processor comparison used was that this time around the magic music maker and the 3D render software Lightwave 7.0 (b) has been extended from NewTek. Also SYSmark 2001 is now with the game. As a special extra, we have not spared even tests with SETI@home. Each of the processors was allowed to continue to calculate PI to the first 1 and 4 million decimal places. Used benchmarks should therefore cover all areas relevant to the user.

• Game performance
  • 3DMark 2001
  • 3DMark 2001 SE
  • 3DMark 2000
  • Aquamark
  • Final reality
  • GLMark
  • Quake 3 arena
• Office performance
  • SYSmark 2001
  • Winace 2.11
  • FlaskMPEG
  • Lame
  • Magic music maker
• 3D of rendering performance
  • Cinema 4 d
  • Lightwave 7.0 (b)
  • Spec ViewPerf 6.1.2
• Other
  • Sandra 2002
  • SETI@home version
  • SuperPI

Who would like to understand the benchmarks in their home even once, will find much of the above listed test programs with us in the download section.

Windows boottime

The program developed by Microsoft Bootvis was used to measure the speed of the boat. This program has actually been developed to optimize the boot process and analyzes startup for this purpose in every detail. Among other things in / output examines the boat activity, CPU utilization, disk the driver delay and much more. Generally at boot time has shown, that these as well as not is dependent on the frequency of the processor, because usually other components unnecessarily delay boot. While the boot time when the Athlon XP 2000 +, Athlon XP 1900 + and Athlon XP was 1800 + on average only 12,85 seconds, the Pentium was 4 2.2 GHz 4 2.0 GHz well 14.2 and the Pentium only after 15.69 seconds ready for use. However, these results are not directly comparable, because when the system, in particular the mainboard and the onboard components have a big impact. Because you can’t just use a Pentium 4 but bad in a socket A Board, we have to live well or badly so. How the installation of applications on startup may have, was after the installation of SYSmark 2001 more than clear. By entire two seconds, the boot process was delayed by this application package. This shows once again, but how important it is, to free the operating system regularly from unnecessary ballast.

Sandra 2002

Before we compete against the team of five processors in real-world applications, we want to look once more the theoretical values of the car at this point. To this end we have used Sandra, since both the 3DNow!, extension of processors correctly recognize SSE1 as well as the SSE2 and accordingly also exploited.

Sandra processor test
Sandra 2002 provides equal to two benchmarks, which are to determine only the performance of the processor. This is the Dhrystone benchmark on the one used, which was originally developed by Siemens, to measure the performance of the main processor. On the other hand, the performance of the co processor is determined via the whetstone benchmark. Both tests be carried out without considering the extended such instruction sets. This result confirmed also when using floating point numbers (floating point data).

< div >

Already in the first theoretical results the Pentium4 and Athlon XP deliver a head-to-head. While the Athlon XP when the Dhrystone, an arithmetic test, the Pentium 4 can, clearly beat even the Athlon XP 1800 + is here faster than a Pentium 2.2, a slightly different picture emerges when the whetstone. At least then, if the optimized version of the benchmark is staff on SSE2 by an Intel used. Here it is possible to lose weight the fastest Athlon XP well 100 points the Pentium4 2.0 GHz. However, the standard whetstone benchmark is used, the results of the Pentium 4 can convince in any way. More than 1000 points separate in this case of Pentium 4 and Athlon XP. The floating point unit of processors, tested by the whetstone is thus much more powerful for the Athlon. The Athlon XP thus decides this test despite his significantly lower the bar for themselves.

Sandra multimedia test
The multimedia test by Sandra 2002 an (chaos theory by Mandelbrot) algorithm, which is used also when generating realistic natural objects such as mountains and clouds. This benchmark consider the extended command sets of the Pentium 4, or that the Athlon XP. Since the implementation of SSE1 in this partial testing better than that of 3DNow! is that we have let the Athlon XP with his SSE1 work unit (i.e. 3DNow professional). The Pentium 4 was, however, both SSE1, including SSE2 to use.

< div >

Also at the multimedia test, the results of the test before confirm. In integer performance (whole numbers), the Athlon XP is again superior. In particular with the help of the SS1 extension makes clear this superiority. In this case, even the Athlon XP 1800 + can beat a Pentium 2.2. When using the SSE2 extension this picture is shifted only slightly. The Athlon XP 2000 + remains still unbeaten and even the Athlon XP 1900 + with a real frequency of 1.6 GHz can all line up before the fastest Pentium 4.

< div >

In such test with floating point data, there is a different result this time. Surprisingly, SSE2 not accelerated the Pentium 4 in this case. Apparently, here is the optimization still left to be desired, although this has been carried out by Intel. We focus results so this time on the SSE1. As we see the Athlon XP 2000 + makes a very good figure here too. However, due to the Pentium 4 2.2 GHz this time the first place. The Pentium 4 2.0 GHz does, however, clearly behind. He must give himself defeated the Athlon XP 1800 +.

Sandra memory test
The memory test by Sandra occupies at least 50 percent of available memory. This benchmark to determine the performance of the storage subsystem, as well as the cache. For this, both arithmetic and floating point Opera stations are carried out. As this test very much depends of the platforms, we want to take only a short look at.

< div >

As we see, the Pentium 4 based on the Rambus platform from the DDR memory can break away system, which was used for the Athlon XP. Specifically, we should bear this fact later in the test in mind.

3DMark 2000

The 3DMark 2000 has found a successor already, is however also still ideally suited to give a statement about the Direct3D performance of the system. Even the processor performance can be measured quite well with this test. Because it is a synthetic benchmark results can be transferred while not directly to the games. But nevertheless it is used everywhere and is easy to each test.

< div >

With these results, one is clearly clear: the 3DMark 2000 is not the right terrain for the Pentium4. The Athlon XP 2000 + is the measure of all things here again. This is due, above all, that introducing 3DMark 2000 only the SSE1 and 3DNow, but not the SSE2 was expanding in the market. Thus SSE2 of the Pentium 4 is broke. Thus, it is already clear that the Pentium 4 in old applications, due to missing optimizations, feels not so good.

3DMark 2000 CPU mark
Also the CPU mark is an important part of 3DMark2000. Although here also the graphics card plays an important role, as well as chipset and memory of the mainboard, are these factors here by the fixed harness off.

< div >

Also here the results from the 3DMark show run is 2000. The Pentium 4 has once again been unsuccessful. Even stronger in this test, as it shows in the overall results. The cause is the fact that the CPU test the memory interface plays only a subordinate to no role at all. The next partial testing shows the influence of the memory bus.

3DMark 2000 HPC
Also the high polygon count is a part of the 3DMark 2000. Since this, the performance of the storage subsystem is more important, it is suitable particularly well, to clarify the interaction between the processor, RAM and graphics card.

< div >

As was to be expected can be the Pentium here easily issue 4 2.2 GHz from Athlon XP 2000 +, whereas the Pentium’s 4 2.0 GHz now in the midst of the test box, even though he 2000 bringing up the rear was CPU test on the 3DMark. This is found in the Rambus quickly, which, in 2002 showed as already the storage tests by Sandra, the DDR system is clearly superior.

3DMark 2001 (SE)

If there is currently a benchmark, which must always represent the performance of the own computer as the flagship programme, this is certainly the 3DMark 2001. To identify possible differences in processor optimization, we have tested 2001 SE with 3DMark 2001, as well as with the successor of 3DMark.

< div >

As we can see, Pentium 4 and Athlon XP are located on more or less the same in the ancestor of the benchmark. The deviation from a point may be neglected here. The revised version of 3DMark 2001, this image turns easily. Now, the Pentium could slide 4 2.2 GHz to the top of the test box. Also, the Pentium 2.0 GHz could outdo the Athlon XP 1800 +. On the theory it seems yet so as would be the Athlon XP at same clock rating (i.e. Pentium 4 2.0 GHz with the Athlon XP 2000 +) the better choice. In how far this result later to confirm, will show up later.

Final reality

The final reality benchmark comes from the year 1997 and thus constitutes a rather old application. Originally intended as the graphics card test, it means no major obstacle for today’s graphics cards. In addition to a test for 3D graphics, operations be tested also 2D, where the graphics card has little influence. The benchmark supports the MMX technology and is therefore not by 3DNow! or SSE1/2.

< div >

Final reality certifies the Pentium 4 excellent performance. Although this test relies not on current processor optimizations, the Pentium 4 with Rambus can break away clearly from the Athlon XP. The reason for this is to look in the larger L2 cache of Pentium4. While at the Northwood 512 kb are used, the Athlon XP in turn only has 256 kb cache. While Intel can clearly rise with an increase in the clock real 200 MHz, the 133 causes MHz clock increase at the Athlon XP no really appreciable performance gain.

Aquamark

The Aquamark massive is a DirectX 8.0 benchmark from the House and is based on the blatantly engine as well as Aquanox. A detailed underwater world is represented. AquaNox uses the GeForce3 nFiniteFX engine. With more than 160 vertex shaders are special effects such as layered fog, caustics, photo-realistic materials, radiosity lighting and shadows generated in real time. Pixel shaders are used for bump mapping and customised lighting effects. The frame rate from the Aquamark are significantly below from Aquanox. Also here, the processor understandably affected the results.

< div >

Clearly, two thumbs up for the Pentium 4. Even if you start with the naked eye difficult to carry out the four frames, but a clear gap between the Athlon XP and Pentium 4 exists. In the large quantities of texture plays a role to be reckoned at this point also the memory. The processor from Intel can thus clearly decide for themselves the last two tests (Aquamark and final reality).

GLMark

The GLMark has been developed by the German software company Vulpine and is a purely synthetic benchmark for OpenGL as well as 3DMark2000 and 3DMark2001 for DirectX. The benchmark is highly complex and can take advantage of some of the new GeForce3 features. Also here, you can achieve a performance by a new processor.

< div >

The successes of the Pentium 4 is easily broken with GLMark. In lower resolutions, which are limited as is well known by the processor, not the video card, it is able to conjure up some more pictures on the screen of the Athlon XP 2000 + on average. However you cannot recognize the difference again with the naked eye. At this point, you would almost think the Athlon XP would have a Quantispeedzahl (performance rating) 2200 + deserved. But should you be not too hasty with this judgment. Finally sets off AMD of the performance rating for the Pentium 4, however, it seems to turn quite well on him, and AMD has taken certainly the step of performance ratings for marketing reasons, to make the difference in the speeds between Athlon XP and Pentium 4 less will appear.

Quake 3 arena

Quake3 uses only OpenGL arena. First person shooter is characterized by a large number of polygons and complex scenarios. Because many games on the Quake III engine based, the tests are very practical with Quake III. We came in addition to the in Quake 1.17 already contained demo 001 still the NV15 demo by nVidia used that affects the processor to the utmost. Was tested with the normal config as well as with a custom config (16-bit, low texture, high Geometic detail).

< div >

The first non-synthetic tests in our test suite is common as in our Quake 3 arena. The results speak clearly for the performance of Pentium 4 with Rambus. Well 40 frames separate the two fast processors in lower resolutions. Clearly, the winner of this test is called Pentium 4.

SYSmark 2001

The SYSmark 2001 has often caused excitement on the part of the AMD fan base in the past. This benchmark is divided up into two areas. The first area simulates the work environment of a “Webmaster” or “Web Designer”, bearing the name Internet content creation. The following real-world applications with a scripting language are controlled in this test section: abode Photoshop, Macromedia Dreamweaver, Adobe Premiere, Microsoft Windows Media Encoder, and Macromedia Flash. While the Athlon XP applications makes a good figure, he failed completely in Windows Media Encoder. You must blame the Windows Media Encoder 7.0 in the shoes at this point but. Weirdly, this application optimized for SSE1 from this command extension. Unfortunately, the Media Encoder only on Intel processors tests whether the processor signals a support of these commands. Because at the time of development not yet on the way was thought that AMD will be put next to 3DNow still on SSE1 could, this move seemed more logical. This does of course in this portion of SYSmark 2001 quite well again, because the application on the Pentium 4 course with SSE1 extensions. Nevertheless the SYSmark simulates the real performance with commercially available applications once. This makes him an important test for the evaluation of processor performance or the performance of an entire system. Thus, we have to accept this test with the knowledge of this small limitation.

In the second area of SYSmark 2001 the Office life (Office content creation) is measured with a whole latte applications, raised in parallel to each other via multitasking. These applications include Microsoft Office, Dragon Naturally speaking, Netscape Communicator, WinZip and McAfee VirusScan. The respective partial results stemming from the Internet content creation and Office content creations, enter each to 50 percent in the final result. But first of all consider the results in detail:

< div >

As we see, the Pentium 4 due to the previously mentioned limitation quite clearly from the Athlon XP can break away. 50 Points separate the two fastest cars. However one has to admit clearly at this point that the Athlon XP 2000 + even with correct SSE1 landed implementation here at most behind the Pentium4 2.2 GHz in second place.

< div >

In the segment of Office content creation, the result is somewhat more realistic. The Pentium 4 2.2 GHz ranks because of the high bar, the larger L2 cache, and the fast Rambus, as was to be expected in the first place, followed by the Athlon XP 2000 + and 1900 +. The Pentium 4 2, 0a joins behind it a. The average of the two sections following result results.

< div >

The great head-start on Internet content creation is reflected clearly in the final result. Due to this fact, Pentium can slide 4 2.2 and 2.0 GHz at the top of this test.

SETI@home version

We are especially proud in our comparison test course on the inclusion of Textclienten of SETI@home version (3.03). Due to the long term, a clear picture of the performance of individual processors should arise here. To keep the results comparable, always the same work unit came with one angle rate of 0,417 used.

< div >

As we see, the Pentium finishes as the Athlon XP 2000 + 4 2.2 GHz two minutes more with the given workunit. The 200 MHz of low-clocked Pentium 4 must be, however, the rest of the test field beaten type. Especially in SETI shows here how much an increase in the processor clock can impact even if the results are only 20 minutes apart between fastest and slowest processor.

SuperPI

The SuperPI benchmark is on for is a rather old application, is however still pretty well, to determine the performance of operations-processor. Unnecessarily long to move the calculation period, we have limited ourselves PI to the calculation of the first million decimal of places of the number. The result was a 1.5 MB or 4 large text file with the collection of posts.

< div >

The results of SuperPI differed only slightly from results already put the processor test by Sandra 2002 open. The integer unit of the Pentium 4 also always doesn’t seem to be grown of the Athlon XP at what time. Here, the architecture of the Athlon XP which uses more power per stroke, their speed advantage in operations that were not optimized for the Pentium 4, to be able to play off appears. Also the 20 stage pipeline of the Pentium 4 is here not conducive, to edit an operation, as described before, longer compared to the pipeline of Athlon XP takes da. Since this test but has no special relevance while working or playing, we leave this result times.

WinACE

To the sense and nonsense of a faster process