home › Education › Intel Pentium 2 4 GHz Dual Core Processor

Intel Pentium 2 4 GHz Dual Core Processor

Alexey Shobanov

Continuing the series of spring premieres, Intel introduced the next model in its line of processors for high-performance systems for home and office - the Intel Pentium 4 processor with a clock frequency of 2.4 GHz. The transition to a 0.13-micron technological process has significantly expanded the “frequency horizons” opening up for the flagship of the processor market from Intel, and now quarterly presentations of new, ever faster processors seem quite common to us. Like its predecessors - Pentium 4 2 GHz and 2.2 GHz, also built on the Northwood core using 0.13-micron technology, new processor has a second level cache of 512 KB in size, which is twice the size of the L2 cache in the younger models of this line, created on the basis of the Willamette core (0.18-micron technical process). The Pentium 4 2.4 GHz is made in the mPGA-478 form factor using the FC-PGA2 (Flip-Chip Pin Grid Array) package, which has the most advanced thermal dissipation scheme to date. Speaking about the thermal regime of the Pentium 4 processor on the new Northwood core, one cannot fail to note the fact that the transition to a new 0.13-micron technology made it possible not only to increase the number of transistors on the chip to 55 million, while reducing its size, but also to reduce The core supply voltage is up to 1.5 V, thereby reducing heat dissipation. So, for the first processors on this core, which operate at a clock frequency of 2 GHz and 2.2 GHz, it is 52 W and 55 W, respectively, and for the new Intel Pentium 4 2.4 GHz it does not exceed 58 W. For temperature control, the processor uses the so-called “Thermal Monitor” technology, the essence of which boils down to the use of a thermal sensor and a TCC (thermal control circuit) unit that controls the supply of clock pulses to the processor. In this case, two operating modes are provided: automatic (Automatic mode) and on demand (On-Demand mode). Auto mode can be activated through the motherboard BIOS. In this mode, when the processor temperature rises to a certain value, the TCC unit is activated and generates pulses that block the supply of clock pulses, which actually causes a decrease in the processor clock frequency by 30-50% (in accordance with factory settings), increasing its idle time, which, in in turn, allows you to reduce the temperature. The TCC's on-demand operation is determined by the contents of the ACPI Thermal Monitor Control Register. According to its state, the TCC block can be activated regardless of the processor temperature, and the idle time of the processor can be varied more flexibly in the range between 12.5% and 87.5%. And, of course, the ability to turn off the computer if the processor crystal is catastrophically heated to 135 ° C has been implemented; in this case, the THERMTRIP# signal is issued to the system bus, initiating a power shutdown. Like all its predecessors, the new processor is built in accordance with the Intel NetBurst microarchitecture, which includes the following innovations:

400 MHz system bus;
Hyper-Pipelined Technology;
Advanced Dynamic Execution;
Execution Trace Cache;
Rapid Execution Engine;
Advanced Transfer Cache;
Streaming SIMD Extensions 2 (SSE2).

In a few words, we will describe these features of the architecture of Intel Pentium 4 processors. The 400-MHz bus (as it is also called - Quad Pumped Bus) allows, due to its special organization, physical level transmit 4 data packets per clock cycle over the system bus with a FSB frequency of 100 MHz. Thus, this 64-bit bus has a peak throughput of 3.2 GB/s, providing high-speed data exchange between the processor and other devices. The implementation of a 533 MHz Quad Pumped bus is expected soon, which corresponds to the operation of the system bus at a physical FSB frequency of 133 MHz, and, as one can easily assume, the data exchange rate on it will exceed the previously unattainable value of 4 GB/s. Hyper-Pipelined Technology involves the use of an unprecedentedly long 20-stage hyperpipeline (recall that the P6 family processors had half the pipeline). This approach allows you to significantly increase the processor clock frequency, although it leads to such a negative consequence as an increase in the reload time of the pipeline in the event of a branch prediction error. In order to reduce the likelihood of such a situation occurring, Pentium 4 processors use Advanced Dynamic Execution technology, which involves increasing the instruction pool to 126 (in the Pentium III, the instruction pool contained 42 instructions) and increasing the branch buffer, which stores the addresses of already completed branches, to 4 KB. This, coupled with an improved prediction algorithm, makes it possible to increase the probability of predicting transitions by 33% compared to processors of the P6 family and bring it to 90-95%. Pentium 4 processors implement a somewhat unconventional approach to organizing the L1 cache. Although L1, like most modern processors, consists of two parts: a data cache (8 KB) and an instruction cache, the peculiarity of the latter is that it now stores up to 12 thousand already decoded micro-operations, located in the order of their execution, determined based on predictions of branch transitions. The instruction cache of the Intel Pentium 4 processor with this organization is called Execution Trace Cache. The Rapid Execution Engine is two arithmetic logic units (ALUs) that run at twice the processor frequency. In the case of the processor we are describing, whose clock frequency is 2.4 GHz, this means that the ALU units operate at a frequency of 4.8 GHz, and given that they operate in parallel mode, it is not difficult to calculate that the processor can perform four integer operations per clock cycle (just over 0.4 µs). The second level L2 cache of the Pentium 4 family of processors is called Advanced Transfer Cache. Featuring a 256-bit bus running at core speed and advanced data transfer circuitry, this cache provides the highest throughput critical to streaming processing. As noted above, initially processors based on the Willamette core had a 256 MB L2 cache; the transition to 0.13-micron technology made it possible to increase the second level cache to 512 MB. This increase in L2 cache had a beneficial effect on processor performance, reducing the likelihood of an access miss. Pentium 4 processors implement support for an increased set of instructions for streaming SIMD extensions (Streaming SIMD Extensions), called SSE 2. In this set, 144 new instructions were added to the existing 70 SIMD instructions. These instructions enable 128-bit operations on both integer and floating-point numbers, delivering significant performance gains on a range of stream-processing tasks. There is only one “but” here - the code of the task being performed must be optimized and compiled accordingly.

With all the above improvements, the processors of the Pentium 4 model line are based on the same 32-bit Intel architecture (IA-32), and the new processor is no exception. As a result, the Pentium 4 2.4 GHz is optimized to work with 32-bit software and shows traditionally stable and high-performance work with such operating systems as Windows 98, Windows Me, Windows 2000, Windows XP and UNIX OS. We had the opportunity to test the operation of the new processor from Intel, using the following test bench configuration:

Intel Pentium 4 2.4 GHz processor;
motherboard MSI MS-6547 (based on SiS 645 chipset);
HDD Fujitsu MPG3409AH-E 30 GB with file system NTFS;
256 MB random access memory DDR SDRAM PC2700 (CL 2.5);
Gigabyte GF3200TF video card (GeForce 3 Ti 200, 64 MB) with nVIDIA detonator v. video driver. 27.42 (resolution 1024×768, color depth 32 bits, Vsync - off).

For testing we used an operating room Microsoft system Windows XP. The test results are shown in the table.

Perhaps someone will ask the question: how much can you increase processor performance and, in general, how necessary are they for modern personal computer such powerful central processors? To this we would like to answer that there will always be work for the central processor. Its computing power can be used by transferring the work of the logic of other computer subsystems to it, thereby reducing the cost of the latter. Some experts raise the question that with further increases in performance central processor it would be possible to shift the computing load of the graphics card processor to it (which has already been done in the past, but with completely different motivations).

In conclusion, I would like to note that the new processor from Intel - Pentium 4 2.4 GHz demonstrates stable operation and excellent performance in applications working with sound, video, 3D graphics, office applications and games, as well as when performing complex computing tasks . In a word, on the basis of this processor, high-performance stations for home and office can be created, capable of satisfying the most demanding user requests and solving problems that place the highest demands on the computing power of your personal computer.

ComputerPress 5"2002

“top” desktop processors at that time that crossed the 2-gigahertz mark. To date, both companies have a new model in their lineups, which means there is a reason to make another comparison or correct the shortcomings of the old one. Researching new models is always interesting if they differ architecturally, but today is not the case. Old cores, the next level of multiplication coefficients - these are the “new processors”. The “reverse” fact is worthy of attention: Athlon XP 2100+ is the last model based on the Palomino core, which was not even previously listed in the release plan and is covering the place until the release of the new Thoroughbred core.

Changes are coming for Intel processors too. Very soon there will be a transition to the 533 MHz bus, so the copy we have is also in some way a “farewell” one.

Well, let's try to make the most of this testing. Firstly, we can compare new model with the previous one, and evaluate scalability based on the difference in indicators in the tests. Secondly, you can put into operation the latest versions of the tests used and add new ones; fortunately, such articles are usually not used for intermediate comparisons. Finally, thirdly, completely useless and completely win-win attempts to identify the absolute leader in speed always remain relevant.

To solve the first problem, let’s add a 2.2-GHz model to the Intel Pentium 4 2.4 GHz, and AMD Athlon XP 2100+ Athlon XP 2000+, and we will test each pair on the same chipset. Based on the experience of the already mentioned large comparison, to solve the third problem we will select the three most interesting platforms for the Intel processor, and for the AMD processor we will limit ourselves to one, the fastest almost everywhere, VIA KT333 + DDR333. As for updating the test suite, please go to the results chapter.

Test conditions

Test stand:

Processors:
- Intel Pentium 4 2.2 GHz, Socket 478
- Intel Pentium 4 2.4 GHz, Socket 478
- AMD Athlon XP 2000+ (1667 MHz), Socket 462
- AMD Athlon XP 2100+ (1733 MHz), Socket 462
Motherboards:
- EPoX 4BDA2+ (BIOS from 05/02/2002) based on i845D
- ASUS P4T-E (BIOS version 1005E) based on i850
- Abit SD7-533 (BIOS version 7R) based on SiS 645
- Soltek 75DRV5 (BIOS version T1.1) based on VIA KT333
256 MB PC2700 DDR SDRAM DIMM Samsung, CL 2 (used as DDR266 on i845D)
2x256 MB PC800 RDRAM RIMM Samsung
ASUS 8200 T5 Deluxe GeForce3 Ti500
IBM IC35L040AVER07-0, 7200 rpm, 40 GB
CD-ROM ASUS 50x

Software:

Windows 2000 Professional SP2
DirectX 8.1
Intel chipset software installation utility 3.20.1008
Intel Application Accelerator 2.0
SiS AGP Driver 1.09
VIA 4-in-1 driver 4.38
NVIDIA Detonator v22.50 (VSync=Off)
CPU RightMark RC0.99
RazorLame 1.1.4 + Lame codec 3.89
RazorLame 1.1.4 + Lame codec 3.91
VirtualDub 1.4.7 + DivX codec 4.12
VirtualDub 1.4.7 + DivX codec 5.0 Pro
WinAce 2.11
WinZip 8.1
eTestingLabs Business Winstone 2001
eTestingLabs Content Creation Winstone 2002
BAPCo & MadOnion SYSmark 2001 Office Productivity
BAPCo & MadOnion SYSmark 2001 Internet Content Creation
BAPCo & MadOnion SYSmark 2002 Office Productivity
BAPCo & MadOnion SYSmark 2002 Internet Content Creation
3DStudio MAX 4.26
SPECviewperf 6.1.2
MadOnion 3DMark 2001 SE
idSoftware Quake III Arena v1.30
Gray Matter Studios & Nerve Software Return to Castle Wolfenstein v1.1
Expendable Demo
DroneZmarK

Pay	EPoX 4BDA2+	ASUS P4T-E	Abit SD7-533	Soltek 75DRV5
Chipset	i845D (RG82845 + FW82801BA)	i850 (KC82850 + FW82801BA)	SiS 645 (SiS 645 + SiS 961)	VIA KT333 (KT333 + VT8233A)
Processor support	Socket 478, Intel Pentium 4			Socket 462, AMD Duron, AMD Athlon, AMD Athlon XP
Memory	2 DDR	4 RDRAM	3 DDR	3 DDR
Expansion slots	AGP/ 6 PCI/ CNR	AGP/ 5 PCI/ CNR	AGP/5 PCI	AGP/ 5 PCI/ CNR
I/O ports	1 FDD, 2 COM, 1 LPT, 2 PS/2
USB		2 USB 1.1 + 1 connector for 2 USB 1.1	2 USB 1.1 + 2 x 2 USB 1.1 connectors	2 USB 1.1 + 1 connector for 2 USB 1.1
Integrated IDE controller	ATA100	ATA100	ATA100	ATA133
External IDE controller	HighPoint HPT372	-	-	-
Sound		AC"97 codec, Avance Logic ALC201A	PCI Audio, C-Media CMI8738/PCI-6ch-MX	AC"97 codec, VIA VT1611A
Built-in network controller	-	-	-	-
I/O controller	Winbond W83627HF-AW	Winbond W83627GF-AW	Winbond W83697HF	ITE IT8705F
BIOS		2 Mbit Award Medallion BIOS v.6.00	2 Mbit Award Modular BIOS v.6.00PG	2 Mbit Award Modular BIOS v. 6.00PG
Form factor, dimensions	ATX, 30.5x24.5 cm	ATX, 30.5x24.5 cm	ATX, 30.5x23 cm	ATX, 30.5x22.5 cm

Test results

We have already tried more than once to formulate criteria for an optimal processor test. Of course, the ideal is unattainable, but today we are taking our first step in its direction we are launching the project CPU RightMark(). For details and news of the project, we refer you to its website; here we will provide brief explanations that should help you understand the essence of the test experiment and its tools.

So, CPU RightMark is a test of the processor and memory subsystem, performing numerical simulation of physical processes and solving problems in the field 3D graphics. Very briefly, one block of the program numerically solves a system of differential equations corresponding to real-time modeling of the behavior of a many-body system, while another block visualizes the solutions found, also in real time. Each block is implemented in several versions, optimized for different processor instruction systems. It is important to note that the test is not purely synthetic, but is written using techniques and programming tools typical for problems in its field (3D graphics applications).

The block for solving a system of differential equations is written using the x87 coprocessor instruction set, and also has a version optimized for the SSE2 set (with loop vectorization: two iterations of the loop are replaced by one, but all operations are performed with two-element vectors). The speed of operation of this block indicates the performance of the processor + memory combination when performing mathematical calculations using double-precision real numbers (typical for modern scientific problems: geometric, statistical, modeling problems).

The results of this subtest show that the speed of working with x87 FPU instructions is higher in the Athlon XP, but due to support for the SSE2 set (naturally, absent in the Athlon XP), the Pentium 4 is much faster. We emphasize that this block does not use SSE commands, so the results of running the test in modes using SSE are omitted (they simply coincide with the corresponding MMX/FPU and MMX/SSE2). We note the almost perfect scalability of the test in terms of CPU frequency - here the influence of memory is almost reduced to zero due to effective caching and the nature of the unit's operation with intensive calculations with a relatively small amount of data exchange.

The rendering block, in turn, consists of two parts: a scene pre-processing block and a ray tracing and rendering block. The first is written in C++ and compiled using the x87 coprocessor instruction set. The second one is written in assembly language and has several options optimized for different instruction sets: FPU+GeneralMMX, FPU+EnhancedMMX and SSE+EnhancedMMX (this division into blocks is typical for existing implementations of real-time visualization tasks). The total speed of the visualization unit indicates the performance of the processor + memory combination when performing geometric calculations using single-precision real numbers (typical for 3D graphics programs, optimized for SSE and Enhanced MMX).

Again, the speed of working with x87 FPU instructions in the Athlon XP is significantly higher, but the use of SSE in calculations again puts the Pentium 4 ahead, despite the support of this set by Athlon XP processors. At the same time, in terms of performance per megahertz, both processors are almost on par, but in terms of total performance, the Pentium 4 gains a lead corresponding to its higher frequency. We emphasize that this block does not use SSE2 commands, so the results of running the test in modes using SSE2 are omitted (they simply coincide with the corresponding MMX/FPU and SSE/FPU). Let us note the excellent performance of the Pentium 4 + SiS 645 combination, obviously caused by the highest memory access speed and low latency. In general, the rendering process is accompanied by fairly active data transfer, which makes the contribution of the chipset and the type of memory used to the overall system performance significant.

The overall system performance is calculated using the formula: Overall = 1/(1/MathSolving + 1/Rendering), so the Pentium 4 gains a very significant benefit when using SSE2 in the calculation block physical model gives almost no performance gain without using SSE in the renderer block. But when performing calculations using SSE, the addition from turning on SSE2 is quite impressive. (Note that this characteristic is valid for specific selected testing conditions, but the test settings allow you to set almost any ratio between the time of rendering the physical model and visualization (by changing the screen resolution or calculation accuracy).) Since the Athlon XP does not support the SSE2 set, its performance quite obviously depends on the rendering speed scenes where it is inferior to the Pentium 4 when using the SSE set, although it remains the absolute champion in “pure” speed of operations using only MMX and FPU. Note that of the tested chipsets for Pentium 4, the i845D looks a little better than the i850 (probably due to the latter's higher latency), and the champion is the SiS 645 for the reason stated above.

A new version of the popular Lame encoder has been available for quite some time, but we haven’t had a chance to use it. As part of the preparation of this article, we tested both the old version 3.89 that we have used so far, and the latest officially available version 3.91. The results coincided completely (within the margin of error), which is quite consistent with the lack of mention of high-speed code optimization in the list of program innovations. (By the way, the encoder has been correctly supporting work with all available extended multimedia instruction sets and registers for more than six months now.) The test, as you can see, perfectly scales with processor frequency, since effective preliminary data caching is carried out here, but a number of questions remain regarding the rather low performance Pentium 4 on i850 and SiS 645. It seems to us that the most reasonable assumption is that such an impact on performance has Board BIOS: we have not yet seen the product from Abit in action, but the board from ASUS on the i850 is very familiar to us, and when used previous version firmware (once again we refer you to the past), such a decline was not observed. Athlon XP is still the leader in this test, and the 2000+ version is quite enough to win.

A new version 5.0 of the DivX codec was released quite recently, but given the enormous popularity of this product, it is not difficult to predict its active use in the near future, without waiting for new releases with bug fixes. Well, we are following popular wishes and moving on to using the DivX 5.0 Pro version. We also carried out similar testing with version DivX 4.12, and the results of comparing codecs are as follows: the encoding operation is accelerated quite noticeably - by more than a minute, regardless of the processor, chipset and memory type. Also note that DivX 5.0 Pro produces a slightly larger output video file. We have nothing to add to the comparison of the processors themselves in this test; everything has already been said in the previous article, but it’s worth paying attention to the good encoding scalability.

In WinAce archiving, as in MPEG4 encoding, the influence of the memory subsystem (due to the large volume of transferred data) approximately doubles the effect of increasing the processor frequency. Athlon XP is still better than its counterpart in this test.

In WinZip archiving, we only note a slight lag in the Pentium 4 on SiS 645 and complete equality in other cases.

The Winstones results look remarkably logical and understandable, but given the frequent inexplicable dips and spikes in these tests in the past, we will probably refrain from commenting.

Let me remind you that until now we had to say a decisive “we don’t believe it!” results of Athlon XP in the SYSmark test, since due to the crankiness of individual programmers, version WME 7.0, which is part of the applications of the Internet Content Creation group of this test, was not able to detect support for the SSE instruction set in Athlon XP. Fortunately, we are finally starting testing in an updated version of the SYSmark 2002 benchmark, which solves this problem.

Briefly about the differences in the test applications:

SYSmark 2001	SYSmark 2002
Office Productivity
Dragon NaturallySpeaking Preferred 5
McAfee VirusScan 5.13
Microsoft Access 2000	Microsoft Access 2002
Microsoft Excel 2000	Microsoft Excel 2002
Microsoft Outlook 2000	Microsoft Outlook 2002
Microsoft PowerPoint 2000	Microsoft PowerPoint 2002
Microsoft Word 2000	Microsoft Word 2002
Netscape Communicator 6.0
WinZip 8.0
Internet Content Creation
Adobe Photoshop 6.0	Adobe Photoshop 6.0.1
Adobe Premiere 6.0
Macromedia Dreamweaver 4
Macromedia Flash 5
Microsoft Windows Media Encoder 7.0	Microsoft Windows Media Encoder 7.1

As you can see, there are no replacements, only version updates. The algorithm for calculating the final points has not undergone any officially known changes, although we would suggest recalculating some proportionality coefficients.

It is interesting to compare the results of the old and new packages in the office subtest: firstly, some kind of correction factor was probably introduced, which led to a decrease in the performance of both sides. Secondly, obviously, due to the redesigned package Microsoft Office, Pentium 4 began to win in this subtest, although in SYSmark 2001 both processor platforms were on par.

In the content-creating subtest, the situation is even more interesting: due to the normal SSE recognition of the Athlon XP in MS WME 7.1, the AMD processor has improved, but the subtest of the new package includes a rewritten one to support SSE2 Adobe version Photoshop 6.0.1, so the Pentium 4 gets an even bigger boost.

As a result, SYSmark Pentium 4 moves from dubious leadership to obvious leadership. Also pay attention to how dramatically the performance of Pentium systems in this test increases with increasing processor frequency, and the almost absent similar effect for the Athlon system.

Rendering in 3DStudio MAX scales perfectly and usually shows no signs of dependence on memory speed, so we can only guess what they did in latest firmware BIOS for ASUS P4T-E by company engineers. The diagram clearly shows that rendering on the Athlon XP accelerates in proportion to the increase in processor frequency, but precisely due to the much higher frequency, the Pentium 4 2.4 GHz takes the lead in this test, although the speed of the 2.2 GHz model was approximately equal Athlon XP 2000+.

In general, there is nothing interesting in SPECviewperf: the results are almost equal everywhere, with a slight advantage of the Pentium 4, and only in the DX-06 is it noticeably ahead of the Athlon XP. Please note that the speed of the tests is practically independent of the speed of the processors.

When switching to a new Intel processor, the gaming benchmark makes a slight leap, but this does not help it even reach the results of the Athlon XP 2000+.

The addition of Return to Castle Wolfenstein, based on the Quake III engine, to the test games, naturally, did not change the situation in any way. Moreover, the relative indicators in these two games are almost identical. Let's add here DroneZ, which differs in the engine, but not in the nature of the results, and only the ancient Expendable remains not very good for the Athlon XP... Note that all games scale approximately equally well with the processor frequency, which also plays into Intel's hands.

conclusions

The farewell to the Palomino core was not very successful: it cannot be said that the Athlon XP lags so far behind its rival, and this lag does not occur everywhere at all, but the trends are obvious. Is it with a real frequency, or with a PR rating? AMD lags behind Intel in terms of magic numbers in the names of processors, and the performance increase with an increase in frequency (no matter how “inflated” it is considered for the Pentium 4) in most of our tests gives an advantage in absolute terms specifically the Pentium 4 line. Many applications finally “found out” about SSE support in the Athlon XP, which gave some boost, but this is a dead end, but optimization for SSE2 is still far from complete, and the further the more applications will switch from “ AMD camp" to "Intel camp".

However, Palomino still leaves its post in decent condition. The gap between the latest model and its existing competitors is by no means catastrophic, the price is attractive, and we are more And It will be interesting to watch AMD's attempts to regain leadership with a new core.

Tray Processor

Intel ships these processors to Original Equipment Manufacturers (OEMs), and the OEMs typically pre-install the processor. Intel refers to these processors as tray or OEM processors. Intel doesn't provide direct warranty support. Contact your OEM or reseller for warranty support.