Testing the first dual-core processors

0 179
Avatar for Geri
Written by
2 years ago

The appearance of the first desktop dual core processors reshaped the whole IT industry. Compared to the previous generations, this was a new paradigm in computing, and also required the programmers to learn new tricks. To understand why this major shift in the basics were necessary, we must understand the events that forced the industry to this direction. In this article, i will also test and compare the first generation of dual core processors to their single core counterparts, and to the second generation of dual core solutions.

The demise of 32 bit single-core chips

The AMD Athlon appeared on the market in 1999, and the Pentium4 was released in 2000. These new processors introduced a new era of computing in the desktop environment. These processors were much faster than their predecessors, the new generation of Athlon processors reached 1 GHz in late 2000, and the Intel Pentium4 got released in November, at 1.4 and 1.5 GHz frequencies. The performance of these processors were enormous compared to the Pentium 1 and Cyrix processors just a few years ago. Not only because the higher clock speeds, but due to architectural changes, that allowed these processors to execute even more instructions per clock cycle than any previous processors.

AMD released a new generation of Athlon called AthlonXP in the end of 2001, and Intel released new models of the Pentium4 az 2 GHz in 2002 January. These new high-end processors conquered the market in the upcoming years. Initially AMD and Intel failed to position these products to low-end and mid-range, because the Athlon and Pentium4 required new processor sockets. Due to this, the masses of people continued to use the older generations of computers, and it took a few years for this new technology to reach mass adoption.

The processor technology was not the only one that developed rapidly. In 1999, a typical high-end desktop PC had 64 MByte of memory, meanwhile in 2001 the price of a brand new 256 MByte memory module was only about $30. From the 1-2 GByte hard disks of the late 90s, a typical PC from 2001 got a 20GB hard drive. The technolgy was developing rapidly, it seemed that nothing can stop it.

And then, stagnation for years

Intel and AMD was not able to increase the clock speeds any more. Despite of all the efforts, the speed of the chips didn't increased further. Even nowadays, most of the chips are running around 2 GHz. In the early 2000s, however, developers thought that x86 processors will reach 20 GHz within a few years. In 1997, a typical high-end desktop processor was running at 200 MHz, and in 2002 the flagship chips from both Intel and AMD was running at 2 GHz. Thinking the industry will reach 20 GHz within another five years, was a realistic expectation.

The limits of the transistor technology

The electrons can't travel faster than the speed of light. Therefore, having smaller chips translating to higher clock speeds, as the electrons have to travel smaller distances. For some reasons, however, the silicon based chip technology doesn't scales well beyond 2 GHz, even if they continue to decrease the size of transistors. Above these clock rates, the chips are becoming significantly more power hungry, exceeding the realistic power envelope. Despite all of the attempts, decreasing the transistor sizes and porting the processors to smaller and smaller chip manufacturing lines, the manufacturers were not able to increase the clock speeds of the chips significantly ever again, up to this day.

The struggles of Intel

Intel got hit harder than AMD. AMD AthlonXP processors were designed for lower clock speeds from the beginning. However, Intel planned to scale the Pentium4 design up to at least 10 GHz with the first revisions of the chip, and up to 20 GHz with later revisions. They have designed the processor in a way that would have allowed them to scale the clock speeds very rapidly, in the expense of the execution efficiency within one clock. This number is called IPC, or Instructions per Clock - a number to measure, how many instructions a processor can execute within each clock. This is not a discrete number, and changes under every algorithm and workloads. Both the AthlonXP and the Pentium4 CPU can execute typically from 2 to 3 simple instructions per clock, but AMD was typically a little bit closer to 2.5, meanwhile Intel was closer to 2 instructions per clock. Modern superscalar processors can execute more than one instructions per clock, because they have multiple instruction pipelines inside them, starting from the Pentium1. Processors like the previous Pentium1 and Cyrix processors were able to execute one to two instructions per clock, and modern i7 and Ryzen processors from 2021 can execute 3 to 8 per clock. This mean't that the Intel Pentium4 was a little bit slower than the AthlonXP on the same clock speeds. Of course, Intel was not able to forsee this problem, and discovering it was a shock for the corporation. AMD was walking in the same shoes, but they were lucky to have a CPU which is more efficient in regards of IPC.

Silent research for four years

After the release of the original Palomino based Athlon XP in 2001, which allowed up to 1733 MHz clock speeds, they have released the new Thoroughbred based Athlon XP which exceeded two GHz, but otherwise it was identical. By the end of 2003 they have released another revisions of the chip, called Thorton and Barton, refreshing the models in 2004, Barton got more L2 cache, but they was unable to increase the clock speed and the performance significantly since 1999.

Of course, AMD and Intel understood that they have to pull out a rabbit from the hat, and that was the Athlon64. As the memory sizes grown, by the end of 2003 everyone was able to afford multiple GBytes of memory, if he wanted. On 32 bit, however, due to the limitations of the 32 bit integer numbers, only 4 GByte can be represented, which was actually limited to 3 GByte under most of the systems. This meant that new instruction sets have to be introduced to replace the x86.

Itanium and x86-64

Intel noticed this first, and they have released a 64 bit processor architecture called Itanium, which was not x86 complaint, and the backwards compatibility to run x86 programs were allowed by software-based solutions. The architecture was meant for servers, but they eventually wanted to release desktop units as well. The architecture failed miserably. Instead of CISC or RISC, they have used VLIW based instruction set for this processor, and the compilers never was enough efficient to reach even the speeds of x86, despite of the Itanium having enormous TDP envelope.

AMD learned from the mistake of Intel, and they have decided to add 64-bit capabilities to x86, keeping the backwards compatibility. The new instruction set, called AMD64, sometimes referred to x86-64, or mistakenly x64, can run the existing 32 bit applications as well, meanwhile it allows to exceed the 4 GByte memory limitations. The new instruction set debuted in their Opteron lines, and later on, they have released a new chip called Athlon64. This instruction set was somewhat faster than the original x86, so software developers was able to get a few ten percent speed increase, if they were willing to compile for the new instruction set. This was the first real improvement in the technology since the release of the Athlon in 1999, and the industry understood, how significant this step is. The Athlon64 gained popularity, despite of requiring new type of motherboards, as they have switched from Socket A to Socket 754 (and then to Socket 939 shortly afterwards, when they decided to put the memory controller directly to the processor).

Intel's capitulation to AMD

After the disaster of Itanium became clear, Intel released multiple generations and iterations of the Pentium4, and by the end of 2003 they reached the absolute limits of the technology. Every desperate attempts to scale the chip in performance have failed. Intel agreed to add the new 64 bit instruction set of AMD into their processors, making it clear that they lost the control over the chip evolution. Intel asked one thing in exchange: they asked Microsoft to delay the 64 bit version of Windows XP till they ship their own 64 bit capable chips in quantities.

Intel had another failed secret weapon

As previously discussed, their processor was less efficient than the solution of AMD, as they were able to execute less instructions per cycle. To fight this, they had a technology called HyperThreading. This technology was used in their servers. This allowed one processor to run two tasks at the same time, despite of having only one core. The operating system saw the chip as two individual processors. This conception was not alien from the server world, as dual processor motherboards, having two separate processor sockets, were a thing ever since the Pentium 1. The technology is called SMP, simmetric multiprocessing. However, only the server programs and server operating systems were able to utilize more than one core. End users only have access to these operating systems since the end of 2001, as Microsoft included multiprocessor support in the Windows XP by default. Linux supported SMP, but kernels in Linux distributions intended for desktop usage, usually had this feature disabled as well.

Everything is about to change - forever

Intel annouced to add both 64 bit support and HyperThreading to their new Pentium4 lines in 2004, but this only materialized in the forms of the Pentium4 model 3.2F, 3.4F, 3.6F and 3.8F processors by the end of the year. Intel annouced a new socket called LGA775 for the new line of processors, which also offered compatibility for the faster DDR2 memory. This new processor generation was called Prescott, but sadly most of the modells had the 64 bit support blocked inside them due to manufacturing defects. As for HT, initially they have previously experimented with HyperThreading on some earlyer high-end Pentium4 chips, and they also had it in some of their server processors. Intel advertised this generation of CPU with its higher clock speeds, not with its new significant technology. Intel was only reliably produce 64 bit variants of this chip in the upcoming years.

Intel Pentium 4 F generation, and 5xx series

The new Prescott line was not super expensive, the power consumption of the chips were typically below 100w around 3 GHz. The cheaper models were available around $200 which wasn't cheap by any means (we are speaking about 2004-2005ish prices), but indeed acceptable for a new high-end generation of processors. The chips gained attention enough from the programmers, who have started to optimize the programs for multithreaded environments. Intel promised 30% speed increase in multithreaded programs with the HT technology. Intel even annouced new models of these chips up to 2006.

Theads and processes

Previously, only server-type of workloads had to use multiple threads. Consumer level applications typically used one threads. Only true programming languages support threads, with the assistance of the operating system, such as C. In the universities, they have only started to discuss the topics of multithreading in 2005, but these discussions weren't deep enough to teach the pupils how to do it efficiently. There was a lot of confusion, how they worked. Some people thought that the processors will throw the data from one to the another, like some kind of manufacturing line. Some people though that games will can be able to separate tasks between cores, like one core will do the character animation, another the collision code, third will process with the rendering and so on, but that isn't possibe either, because you cant just calculate animation as you render the objects, as it will cause a lot of polygon flickering on the screen as you modify the data of an object you are just rendering simultanously. So the best method turned out to just cut the tasks in half, so one thread is processing the animation of half of the objects, another thread is processing the animation of the another half, then when both job is ended, other modules can be processed as well. But programmers needed years to realise this, so the speed adventages of the first multithreaded programs were very small.

AMD screams internally

Whith the apperance of 64 bit HT capable Intel processors, AMD was about to lose the controller role on the chip development. AMD and its fans were portraying the HT technology as inefficient, but they understood the dangers of it. As more and more programs supported multithreaded operation, and the Linux and Windows operating systems got patched to be able to do it more properly, AMD understood that they must follow the footsteps of Intel. Both manufacturer started to experiment silently with dual core chips on the server market. AMD annouced multiple times that they could release a dual core CPU to the consumer market any time, but Intel can't do it right now. Intel's reply was basically: HOLD MY BEER.

The first dual core CPU on the market

Intel announced the world's first (desktop) dual-core processor, the Pentium D in 2005 may 26. AMD announced its own dual core Athlon64 x2 processor one day later. Both corporations tried to keep the prices around $200, even if that meant lower profit margins. Of course these were paper launches, but by the end of summer, people was able to buy these chips. After 5 years of stagnation in technology, the ghost of multi-core processors escaped from the box, and software development changed forever. Of course, compatible motherboards were also needed to run these chips, software and operating systems had to be patched a little bit, so there was no sudden chatharsis especially for those who only ran one application at a time.

Carrying out this test

I ended up with some first-generation dual core CPU solutions with motherboards in the previous years. Luckily, i was unable to sell them. Well, luckily for the test... because i was unlucky for not being able to earn some profits on it. I took out the systems from their sarcophagous, put memory modules into them, tested if they still work and boot without problems, and cleaned them, if needed. I had to order some hardware to be able to carry out this test, but luckily i already had most of the hardware i needed to carry out this test.

The Intel test systems

Intel Pentium4 524

The Pentium4 524 is based on the Prescott core, and its a 64 bit Intel chip, which also supports HyperThreading. So this is not yet a dual core chip, but the operating system detects it as two separate chips due to its virtual secondary core. The chip runs at 3.06 GHz, and it was a re-release of early Prescotts later on, in 2006. Despite its clock speeds, it only consumes 84w. Its a 90nm chip, and has 1 MByte of L2 cache memory on board. The processor uses the LGA775 socket, and it will work in almost every LGA-775 based motherboards. The HT can be disabled, in that case, the processor will be operate as a single core chip. The chip will be therefore measured twice. This chip is going to indicate two things. One, how efficient the virtual secondary core of early Intel processors were. And another one is, how efficiently a program is multithreaded, like, if its even attempting to use multiple cores, or not. This practicular chip is manufactured in 2006, but this design is the oldest in this test, so despite its manufacturing date, this will feature the earlyest solutions before the real dual core models.

Intel PentiumD 925

The Pentium D line is the first dual-core solution of intel for the desktop market. This specimen of PentiumD was also released in 2006, one year later of the original PentiumD release, and it only differs from the original model in the larger cache memory (4 MB vs 2 MB). I have choosen this chip because its also 3 GHz, so it would be easyer to compare it to the previous Pentium 4 chip which also runs around this. Rumors were saying it consumes too much and emits too much heat. This is not true. Despite its two cores, its still consumes only 95W. This was made on 65nm, the initial PentiumD 820 was manufactured on the 90nm node, but also comsumes 95W on a few 100 MHz lower clock speeds. I am very curious to see the worlds first dual core processor architecture in action.

Intel Core2Duo E6300

The Core2Duo E6300 has nothing to do with the Pentium4. Its the second generation of Intel's dual core processors, therefore, it was named Core2. Intel developed this chip from their previous chip generations, and it supports 64 bit, and it has two cores. The Core2Duo was meant to fix the problems with IPC, and offer even better performance on the same clock speeds than AMD does. This chip runs at 1.83 GHz and it was released in the end of 2006. Intel claimed its about 40% faster than Pentium D from smaller power consumption. We will found out, if this is true, or not. It only consumes 65w, and it has 2 MByte of L2 cache. This model was very popular and cheap in the upcoming years - the initial release price was about $180. Weaker models were released with the Pentium Dual Core brand name, with only 1 MByte L2 cache. Not all LGA-775 motherboard supports the E6300, and most of the early boards, if they do, will require a BIOS upgrade, which the manufacturers released later on. This meant that some people was forced to continue to use the PentiumD even if they wanted the Core2Duo instead, so Intel had to keep the older D line alive for a while parallelly to this.

Operating system

As an OS, the Windows 7 64 bit was used in this test. I don't use Microsoft operating systems for almost a decade. It was a terrifying experience to handle this system. It takes about 2-3 minutes to load, compared to 20-30 seconds of a comparable Linux distribution. It killed itself on every time i changed the motherboard, so i had to completely reinstall it every single time.

picture: The never ending automatic repair procedure

The user interface is super low quality, this was the time when Microsoft decided to leave the world of Desktop, and create a content consuming operating system, for users without brains. I have only used this operating system for this test because its era-correct, otherwise i would not recommend using any Microsoft operating system for any purpose, so please use Linux instead, it will run perfectly on all of the systems listed in this article.

The graphics card

The nVidia GeForce 8800 GT was released in 2007, its an upgraded (downgraded) version of the 8800GTS from 2006. Its a peroid-correct video card for this hardware, it was very popular for gamers all across the world, even in the early 2010. The card has 512 MByte video memory, but i have downclocked the card from the stock 600/1500/900 MHz memory clocks, to 433/1200/633 MHz, and also downvolted it from its bios from 1v to 0.9v to decrease the power consumption, the heat output, and preserve the life of the card itself. The GeForce8 generation was the first card with unified shader architecture, the high end models used 256, 320 or 384 bit wide memory bus interface, and was equipped with 320, 512, 360, 768 or 1024 MByte video ram, althrough there are barely any difference between performance in games above the 512MB model. (Originally, i wanted to use a Radeon x740XL for this test, which is a very old and weak graphics card compared to the 8800GT, however, luckily i just got a 8800GT for the test).

The Intel test rig

The motherboard i will use for the Intel machines are an LGA775 based motherboard from MSI, called G41M-P26, and it supports processors starting from the Pentium4 up even to some Core2Quads up to 95W, so it will be fine for this test. This motherboard has an integrated video card, which will not be used in this test, because this test is only focusing on the processor power. As a video card, the 8800GT will be used, however the motherboard also has an integrated Intel GMA based IGP on its own. This motherboard has only two memory slots, luckily i was able to find two 1GB memory sticks for it. The sticks were running at DDR2-533 rating with the Pentium4, but the motherboard switched to DDR2-800 whith the Core2Duo and Pentium D processors. The sticks were running in dual channel mode, despite of using random memory modules without various speeds and latency parameters.

Motherboard: G41M-P26

Socket: LGA775

Processors: Pentium 4 524, PentiumD 925, Core2Duo E6300

Memory: 2x1 GB

Video card: GeForce 8800GT

Hard disk: 40 GB Maxtor

Operating system: Windows7

The AMD test systems

AMD Athlon 64 X2 4400+

This chip was released in 2005 may, but actual samples reached consumers by the end of the summer. The chip has two Athlon64 cores inside, and runs at 2.2 GHz. The socket it uses, it the Socket 939. The initial release price was about $580 of this chip, which is about 3 times of the price of the initial prices of Intel. AMD justified the higher pricing with the higher performance compared to the PentiumD. We will see how AMD's claim will keep up against reality. Its important to note that these early dual core AMD chips use the DDR ram, and not the DDR2 which the Pentium4 chips in this test will use. This chip has its own integrated memory controller. Similarly to the Pentium D, it has a good 90w power draw under full load. It has 2 MBytes of L2 cache (2x1MB).

As this chip still uses DDR memory, it was hard to find proper models. I was able to find four pieces of 512 MByte memory modules, however one of them turned out to be ECC, and it didn't wanted to when it was inserted into the motherboard together with the other non-ECC modules. Then i found some 256 MByte sized modules as well, but none of them wanted to start up in the motherboard for some reason. After wiping the memory modules with alcoholic disinfectant, two of the 256 MByte chips luckily decided to start up in the motherboard. This means that 512+512+512+256 MByte modules will be used in this test. They are different modules with different clock speeds, but the bios allows to force everything to run on DDR-400. This, however, didnt worked. Besides DDR-333, the bios offered my to clock all the memory chips at DDR-366, which is not an actual memory standard, and in reality, it translates to 183 MHz memory chip clock. Luckily, this worked (it gave a good 5% extra performance for the CPU).

The motherboard for this CPU is made by Gigabyte, and the model name is GA-K8NF-9. This motherboard has no integrated video, but it will be used with the GeForce 8800GT anyway.

Motherboard: Gigabyte GA-K8NF-9

Socket: 939

Processor: AMD Athlon 64 X2 4400+

Memory: 512+512+512+256

Video card: GeForce 8800GT

Hard disk: 40 GB Maxtor

Operating system: Windows7

AMD Athlon 64 X2 4800+ (AM2)

This chip is not just the newer and faster clocked version of the previous CPU, its actually a somewhat new design, it was released by AMD in winter 2006 as a quick response against Intel's new Core2Duo line. The chip uses AMD's new AM2 socket. They have decreased the prices of these new AM2 chips to the $200-$300 price ranges to be able to compete with the Core2Duo. Finally, AMD has switched from DDR to DDR2 memory as well, and this chip has its own DDR2 memory controller. They have increased the clock speed of this CPU to 2.5 GHz, but they have decreased the size of the L2 cache to just 1 MByte (512 KByte per chip). Probably they was thinking, the newer and better memory backend will allow then to spare on the L2 cache size, we will see how this will turn out. The chip was released in the last days of 2006, and the new AM2 platform itself was meant to replace the 939 platform as a whole.

The motherboard i have for this chip is an ASUS M2A. I found four 512 MByte DDR2 memory module for this motherboard, luckily this motherboard also have 4 memory slots, so this is going to result 2 GBytes of memory for our test. This motherboard has a terrifyingly weak GeForce 6200 integrated video card, and despite of this being a GeForce, its hard to tell if its actually better than the GMA950 on the LGA775 boards or not. The memory sticks i found, were rated usually at DDR2-533, which means 133 MHz memory chip clock. Despite my attempts to rise the memory speeds, the system was not stable at higher rates, so the memory was used on its stock 533 ratings. The dual channel memory mode was also refusing to activate for some reason.

Motherboard: ASUS M2A

Socket: AM2

Processor: AMD Athlon 64 X2 4800+ (2.5 GHz)

Memory: 512+512+512+512

Video card: GeForce 8800GT

Hard disk: 40 GB Maxtor

Operating system: Windows7

Let's begin the test

I personally have never saw these systems in action. Previously i have tested if they can boot, to check if they are okay, before i sell them, but thats all. I was aware of their existence, but i peronally ignored the release of these chips, because i have used Socket A from 2004 up to like 2013. I don't know what to expect, there are too many contradictary informations in the internet about this era. Such informations saying that the first gen dual cores are very inefficient due to they use the system ram for implementing interchip communication (Athlon64 x2), or they run too hot (Pentium D). I have already determined the later rumor to be false, even under full load, the temperature of any of the processors in this test, were about the similar as the temperature of my hand (by judging from the heatsink).

FFMPEG video encoding to h264

The new video standard of the age was h264, online videos started to use this format for 720p and 1080p video contents. The codec is more CPU demanding than the MPEG1 or MPEG2, both when encoding or decoding it. In exchange, it gives about 3-4x smaller files for the same video quality. The h264 encoder used in this test is the 64 bit version of mmpeg. First, the Pentium4 was tested:

We can see, that the second virtual core, instead of helping the performance, its just made the things worse. The HT-less variant finished the conversion within 2:25 sec, but the HT variant needed 2:35 fps for this task. Its a little bit disappointing. Now lets see the real dual core processors as well:

The PentiumD is able to unleash its potential here. The dual core version of Pentium4 with the giant cache offers super performance, the its faster more than twice than the single core original P4 processor. The CPU finished this test within one minute, which is reasonable, as the length of the converted video is about half minute.

The Socket939 based Athlon64 x2 also finished the test within one minute. Very impressive performance compared to the single core Pentium4 chips, but its a tiny bit slower than the PentiumD. It seems AMDs bold claims about superior performance compared to the Pentium 4 doesn't materializing in this test.

Intels new generation of dual core processor, the Core2Duo, is performing about 15% faster than the Pentium D, despite of running far less clock speeds. It seems Intel processors like the h264 encoding more than AMD.

The AM2 socket based Athlon64 x2 is slower, and finishes the test within 66 seconds. This is about 10% slower than the performance of the older Athlon64 x2, which is quite disappointing. It seems lowering the size of the L2 cache hogs this processor seriously, especially with these lower-clocked DDR2 memory modules, despite AMD clocked the chip notably faster. AMD offered this generation to fight against the new Core2Duo chips, and according to this test, but its notably slower than the older PentiumD and the Socket939 variation of the chip.

Video playback performance

In this test, i have used VLC media player, and disabled hardware video decoder acceleration to check, how these processors can handle the h264 video playback on their own. The video i used was using the 720p resolution, and was encoded with ffmpeg with the default presets.

Even the non-HT Pentium4 can play back the video without stuttering, all CPU-s from this era was able to pass this test without any problems. This time, AM2 variant of the Athlon64 x2 was faster a bit, requiring less CPU utilization to play the video, but the AMD cpus can't compete with the Intel counterparts. Both the Pentium D and Core2Duo beats both AMD Athlon64 and uses significantly smaller CPU utilization to play the video. The dual core processors from AMD can barely keep up even against the Pentium4's virtual dual core.

The Core2Duo requires even less CPU usage to play back the video, although it still hovers around 23%.

Compression

7zip is a popular crossplatform compression and decompression program. The 64 bit version was used. It can utilize the second core. In the test, the same 100 MByte file was compressed accross all the systems, with default/normal compression.

We can see that the Pentium4 variation with its virtual second core is faster than the non-HT Pentium4. The virtual core gives about 50% performance improvement, which is a very massive speed-up, allowing the file to be compressed just within 93 seconds. Lets see what is the case with the first real dual core processors:

The PentiumD needs only 42 seconds to finish the job. The Socket939 based Athlon64 x2 finishes the task within 55 seconds. Once again, Intel beats AMD by around a good 20%. AMD's claims about being significantly faster than the Pentium4 based Pentium D starting to fall apart quickly.

The Core2Duo processor requires 39 seconds to finish the task, which is only barely faster than the Pentium D, but as the Core2Duo eats less power than the Pentium D, it can be still considered a notable advantage.

The AM2 based Athlon64 x2 suffers from a terrifying speed drop under 7zip. Don't forget, this CPU was released against the Core2Duo, but it fails to compete even with the single core Pentium4. Its the slowest in this test, requires more than 3 minutes to finish the compression. This is probably not just due to the weaker cache, something else must be a contributing factor as well - probably a hardware bug in the branch prediction engine, that causes anemic performance under this compression algorithm. Very disappointing to see such weakness of the newer generations of the Athlon64 x2 cpu, this unfortunately makes this CPU less ideal for office environments.

Maker4D battle

I have bench-marked Maker4D, my own free 3D RPG game maker engine, because i wanted to test how it performs anyway. A 3D battle in game made with Maker4D was bench-marked, with 4 characters in total, in 1024x512 window. Although this is a game engine, it uses software 3D rendering, therefore it doesn't utilizes the graphics card to compute the 3D graphics, so i threat it as a CPU test and not as a traditional gaming test.

First, the Pentium4 CPU was tested, with its second virtual core:

The HyperThreading indeed was faster even here, it helped the renderer to go from 20 fps to 24 fps, pushing it over the limits of relatively fluid frame rates. The 20% performance increase is very impressive from this early virtual dual CPU technology. Lets see the real dual core solutions:

The PentiumD can achieve two times more fps than the single core Pentium4, reaching a constant 40 fps. Maker4D can utilize two cores very efficiently, and the PentiumD can unleash its claws. The 939 version of Athlon64 x2 can only reach 35% only, which is a victory for Intel once again.

The Core2Duo pushes the fps above 50 fps, hovering around 51 fps, showing its superior performance in rendering compared to the previous generations of processors.

The AM2 version of the Athlon64 x2 this time is faster than the older 939 version (probably because when i have designed my software renderer algorithm, i was designed in a way to avoid dirting too much cache memory). It still can't reach the speed of the Pentium 4 based Pentium D, not to mention the Core2Duo.

Cinebench

Normally, i don't do synthetic benchmarks, as those benchmarks don't reflect real-life usability in most cases. A friend asked me to do a Cinebench R15 benchmark anyway, because he was interested to see the end results, as he can easily compare it to his own results. So i agreed to do the testing of this program, but its important to note that Cinebench doesn't reliably measures ray tracing performance. A ray tracer producing this kind of graphics with this quality, should run on multiple frame per seconds, but Cinebench is multiple thousands times slower, and takes minutes to compute a frame on these chips. This means Cinebench lacks the actual optimizations of ray tracer algorithms, the code quality probably resembles the code of absolute beginners, this isn't a professional quality of algorithm for sure, so i am asking other benchmarkers as well, DO NOT use this program for measuring computing performance.

Anyway: lets see how the HyperThreaded Pentium4 performs under this test:

The second virtual core gives a good 15%-20% performance boost, similarly as how other algorithms scale on this Pentium4 based virtual dual core technology. Lets see the real dual-core processors now:

In this software, the Athlon64 based technology runs circles around the Pentium4 based chips, this is probably a very FPU-heavy test. Too bad it has nothing to do with reality. Cinebench shows superior performance for AMD chips, but in real world usage, dual-core Intel processors were significantly faster in every tests so far. Bribed test writers, or application-specific optimizations in the processors?

As we can see, the earlyer Socket939 variation of the Athlon64 x2 outperforms the AM2 variant once again. There is one possible explanation of this, and its the scene acceleration structure of Cinebench, which is unfriendly to the newer Athlon processor's smaller L2 cache. I am surprised Cinebench even uses some kind of acceleration structure, but that would explain the speed drop on the new AM2 model.

The games

Most of the games i have tested, came out after 2007. These games are newer than the processors, but they are within the lifetime of the processors, as people used these hardware actively up to 2013. I haven't checked previously, which one of these games can utilize multiple CPU cores, or which one are limited to one core. The composition of these games should resemble the reality regardless, althrough i found these games very boring and bad. Its a mistery why people played these games, they aren't entertaining. Really. I wanted to test more games, but i haven't found more game demos. As studios stopped releasing demo versions at this time, thinking, people will buy their low quality AAA games anyway. These AAA games are indeed the lowest quality and boring games i have seen in my entire life, i seriously wondering how they even managed to sell a copy from these.

Batman: Arkham Asylum

This game is very CPU demanding for some reason. The game is super boring, it gave me a feeling that its more like some kind of animated film than an actual game. I pulled down the settings to low before i tried to run this game.

The game ran unplayable on the Pentium4 chips, the HT version with virtual dual CPU manages to reach 11 FPS, meanwhile the single core version can only reach 9 fps. This game doesn't looks like its actually using those processor cycles for something useful, its just unoptimized, and super bad.

The Pentium D scaled very well, it reached 22 fps, this however didn't offered a good experience, as the game sometimes fell below this, making the game very unpleasant to enjoy, as the game was not designed to be very enjoyable at low frame-rate.

The Socket939 version of Athlon64 x2 loses against the Pentium D once again. The 19 fps sometimes jumps up to 21 fps, sometimes falls far below. Probably this platform would reach playability with a little overclock and fine tuning in the settings, but this type of experimenting is not the topic of this test for now.

The Core2Duo finally reaches 29 fps, which only falls to 25 fps under heavyer action. We can call this game playable at this point. Impressive performance from the new Intel generation, its around 30-40% faster than the Pentium4 on far smaller clocks, and it also beats AMD into the ground.

The newer AMD2 version of Athlon64 x2 is notably slower than the predecessor, not to mention the Core2Duo it was released against... It only manages to reach 16 fps, which is feels very unplayable in this "game". Once again, the newer Athlon64 lags behind, and the speed deficit is enough to push the game from borderline playability to totally unplayable depths, offering around 40% slower performance than the Core2Duo.

Quake4

The Quake4 uses OpenGL, and its the last game of ID software before it bankrupted. Another totally boring game, that caused the final demise of that corporation. The first scene wants to be very epic, when some house sized meduza alien walks away. This is probably the only feature of this game, otherwise this is a WASD techdemo with guns, that looks like something from 2002. The fps was measured at the meduza alien scene scene.

Quake4 is ugly, but at least its not CPU demanding. It was managed to reach perfectly fluid frame rates on all of the systems above. It seems its unable to utilize multiple cores, as the FPS barely differed between the single core and HyperThreaded Pentium 4 cores, and across the dual core processors whatsoever.

The Socket939 based Athlon64 x2 system was again almost 20% faster than its newer iteration, the AM2 based Athlon64 x2, which predicts a terrifyingly very bad gaming performance from the newer Athlon64 generations. The AM2 variant of the Athlon64 x2 barely manages to keep its nose ahead against the Pentium 4 processors.

The Socket939 based Athlon64 x2, alongside with the Pentium D and the Core2Duo topped out at a solid 63 fps. I have attempted to turn off the vsync, thinking that maybe thats the problem, but it didnt helped to increase the performance any further. Maybe there is some limitation in the engine, or maybe there was something wrong with the graphics drivers. Either way, this three CPU scored a constant 63 fps.

Need for speed: Shift

This game meant to be some kind of epic car racer game with epic camera effects. On startup, you have to skip various camera animations around the car, which the game developers tought to be epic, but they are actually just really boring and lame. The game tries to create the impression like the player is really inside the car, when driving it. It works, but more like an irritating way than an actually enjoyable way. The game is only enjoyable on large monitors, on small ones, the track appears to be too small, and it feels like you watching the game in a mouse-cinema. The developers of this game clearly didn't knew how a game should look like, but at least the graphics looks enjoyable.

The HyperThreading abilities of the Pentium4 giving no speed-up here, despite of the game really supports multi-threading. The game is unplayable on the single core P4 chips, barely reaching 9 fps, and it really requires a more modern processor with at least two real cores.

The Pentium D achieves a solid 25 fps, which sometimes fells below 24, but usually runs above 25 fps, making the game-play totally fluid. This CPU is indeed more than two times faster than its single core variation, offering very good scaling.

The first Socket939 version of the Athlon64 x2 was able to run this game at 26 fps and above, the game felt rock solid and fluid, everything was perfect. This time, this processor was able to beat its Intel counterpart with one fps, this is basically the only real-life test where the Socket393 based Athlon64 x2 was able to keep up the pace with the Pentium D.

The Core2Duo is about 30% faster than the Pentium D, the performance is hovering around 33 fps, making it totally playable even when heavy collisions and action is taking place on the screen. The Core2Duo is once again beating the competitors by huge numbers, showing AMD its place.

We can't say the same from the newer AM2 version of the Athlon64 x2, which was able to achieve only 14 fps, which is almost 50% worse than the performance of the first Athlon64 chips. Its almost 3 times slower than its main competitor, the Core2Duo. At this point, probably the combination of the small L2 cache, the bigger latency of the cheap DDR2 chips, and some sort of strange bug in the branch prediction units resulted unplayable frame rates, the chip can only compete with the single core Pentium4 chips and the game is totally unplayable at this point.

Crysis

But can it run Crysis? Yeah. Sadly, they can. Crysis is merely more than a tech demo, designed to sell their game engine, which isn't even a game engine, just a rendering engine techdemo, fine tuned to render grass, water, and forests. The studio tries to compensate for the lack of story, gameplay with pre-rendered videos, which you can't skip. The game doesn't even looks that nice. Its a mystery what keeps the Crysis and Far Cry franchise alive, beyond irrational fanboyism of some loud but clueless gamers. All of the games look like the same, they lack story, action, and everything that would turn them into a real game. But, lets benchmark it anyway.

The game runs totally playable on a Pentium4, so they at least developed this game engine carefully, to avoid hogging the processors too much. The HT version of the Pentium4 gives a good 10% speed increase, accelerating the game up to 38 fps.

When running the game on real dual core processors, we can see only some minimal speed increase, so Crysis can't really profit too much from the extra cores, the performance of the game engine is however enough on single core systems as well. The Socket 939 version of Athlon64 x2 reaches 40 fps, beating the Pentium4 based machines with a few FPS.

The Pentium D is barely faster than its single core Pentium 4 counterparts, which means that this game is really unable to profit too much from the extra processor cores.

The Core2duo reaching 58 fps, it almost collides to the monitor sync rate, offering a good 25% speed improvement compared to its predecessor, the Pentium D.

The AM2 version of Athlon64 sets a new negative record, its even being beaten by the Pentium4, and compared to the older 939 based model, its about 45% slower. The tiny sized L2 cache with the combination of a flawed branch prediction engine and low quality DDR2 memory modules can be the reason of the anemic performance.

Conclusion:

Pentium4 HyperThread 524

The latest Pentium4 chips with the HyperThread virtual dual core technology, are indeed noteworthy. If the motherboard supported this chip, upgrading to the newest HT capable 64 bit Pentium4 processors indeed giving a notable boost in performance. It wasn't helping too much in games, but it gave a nice improvement for office-type of usage, where compressing, rendering, and integer based workloads and browsing the internet can get an extra 15-30% performance from this chip. The chip has about 10-20% overclock potential as well.

Pentium D 925

The PentiumD is Intels dual core Pentium4 chip, and is very impressive. AMD Fanboys laughed this chip, but Intel made them to eat their hat. It beat the Athlon64 in almost every real life task from gaming to office type usage. The second chip is giving a 100% performance boost, when its being fully utilized. When running more demanding games, the second core of the Pentium D will help the chip to reach playable or even fluid frame rates, and demanding office software will scale.

The horror stories about the pathetic Pentium4 which boils like a water heater and slower than anything else in the universe is not true, the chip outperformed the equivalent AMD counterparts in basically every test. The chip was not even that hot, with the thin-sized aluminium stock Intel cooler (meant for 60w processors), it was still touchable under full load.

The chip has about 10-20% overclocking potential, if proper cooling is applied (for example, the more thick stock Intel cooler, which was meant for 90w CPUs). The chip can reach 3.3 to 3.6 GHz when overclocked.

Athlon64 x2 for Socket939

The instant answer from AMD to the PentiumD and Pentium4 HT was clearly not as brutal as AMD planned. When AMD-s throne was threatened by the newest iterations of Pentium4, the mutant dual core Athlon64 with its large cache memory was enough from AMD to keep up with Intel for a few more months, but the ground slipped out from the feet of AMD. AMD never thought that a Pentium4 based solution will force them on its knees, and its happened out of the blue. This chip gave enough time for AMD to prepare the launch of the new generation of the family.

The overclock potential of this chip is about 5-10%, so its not too high, the 2.2 GHz model can reach about 2.4 or 2.5 GHz.

Core2Duo

The Core2Duo was Intels new chip, this time focusing on stronger IPC and it was based on Intels earlier superscalar designs. The chip is a success, Intel was able to beat AMDs Socket 939 based Athlon64 solutions, and it was able to beats its own PentiumD chip by 20-30%. On the same clock speeds, its almost twice as fast as the Pentium D, which meant that this chip had very good potential, as this chip only runs on 1.86 GHz, yet it offers this brutal performance.

The overclock potential of this chip is about 30%, despite of running at 1.86 GHz at stock, it can easily reach 2.4-2.5 GHz on stock voltage.

Athlon64 x2 for AM2

This chip is bad. Its full of hardware bugs, it was released against the Core2Duo, but both its predecessor, the Socket 939 based Athlon64 x2, and the Pentium D also beats it. Just within months, AMD lost its throne and its leading role on the CPU market, and was forced to compete with this very weak solution. Beginning from this point in the history, AMD was not able to offer any CPU for the high-end market for more than 10 years.

The overclocking potential of this chip is about 20-25% without rising the voltage, but trying to find proper memory modules for it is also necessary to get more usable performance from it, due to the tiny cache sizes.

Summary

9
$ 5.31
$ 5.26 from @TheRandomRewarder
$ 0.04 from @Metalhead33
Avatar for Geri
Written by
2 years ago
Enjoyed this article?  Earn Bitcoin Cash by sharing it! Explain
...and you will also help the author collect more tips.

Comments