Articles: CPU
 

Bookmark and Share

(2) 

Table of Contents

Pages: [ 1 ]

Usually the info about the launching of new Intel or AMD processors leaks to the press in advance. At least a few weeks prior to the actual launch, we are pretty much aware of all the specifications, working frequencies and prices of the new processors. The meticulous journalists delving into every detail see, to be very professional in digging out juicy facts about the upcoming products of the leading microprocessor companies. However, the today's announcement of the new Athlon XP from AMD is surprisingly unexpected. AMD has been concealing these drastic changes of their roadmap very successfully, so that the first hints started cropping up in the press only in the end of last week. Nevertheless, the fact is undeniable: today, on August 21, 2002, AMD officially announced two new models in the Athlon XP family. They are Athlon XP 2400+ and Athlon XP 2600+.

This harsh and unexpected demarche of AMD's was evidently aimed at only one single thing: the company is not going to give in and continues the struggle with its main competitor, Intel, for the title of the fastest x86 CPUs manufacturer. Of course, according to their preliminary plan, Athlon XP was supposed to reach 2600+ rating only by the end of the year. However, since the times AMD was working on this plan, Intel has already made a few significant changes to its own roadmap. If you remember, Intel was planning its Pentium 4 2.8GHz to come out only in the very end of the year. Now they are going to start shipping Pentium 4 3.06GHz in November already. Moreover, this CPU will boast HyperThreading technology, which theoretically allows (in case the applications are optimized for it, which is kind of too early to speak about) improving the CPU performance by 30% (as Intel guys claim). As a result, the final Intel roadmap looks as follows:

You will definitely agree that the launching of Athlon XP 2600+ in the end of the year could hardly look any convincing against the background of this roadmap. Especially since AMD cannot afford to let Intel take an indisputable lead in the high-performance processors market, because they need to make sure that the upcoming 8th generation Athlon XP processors (ClawHammer) will get onto fertile soil when they come out in last 2002-early 2003. This seems to be one of the reasons why AMD decided to revise its roadmap and speed up the launching of the new Athlon XP processors, just like Intel, especially since the Fab30 in Dresden using 0.13micron manufacturing technology allows it.

As a result, the today's announcement of the new Athlon XP 2600+ and Athlon XP 2400+, which can boast significantly improved performance, will be only the first move in this direction. Some other steps on AMD's part are to follow, which will help the company to retain a certain parity with Intel in the today's CPU market. According to some preliminary data, they are planning to launch Athlon XP 2800+ in October, which is most likely to be designed to support new 333MHz bus, thus allowing to take full advantage of the DDR333 SDRAM bandwidth, which is little by little becoming a new standard. Then, by the end of this-beginning of the next year, a new Barton core should come out, which will add a twice as large 512KB L2 cache to the features of Athlon XP. The first Barton based CPUs are expected to be rated as something like 3000+. However, around the same time, in 2003 new Athlon processors based on 8th generation architecture are expected to come out, so there is no need to worry about AMD's future in 2003.

But, let's return to today. The launched Athlon XP 2400+ and 2600+ indicate a breakthrough, which allowed AMD to outpace Intel, which processors can now boast the maximum working frequency of only 2.54GHz. However, you shouldn't forget that next Monday Intel will announce Pentium 4 2.66GHz and Pentium 4 2.8GHz and will again leave AMD behind, but the first word is said: AMD can be the first if they wish to. Especially since the new Athlon XP do not use 333MHz bus yet.

New Athlon XP: Not Only Higher Core Clock, But…

As you surely remember from our previous Athlon XP 2200+ Review, these processors couldn't boast very good overclocking potential. Our Athlon XP 2200+ working at 1.8GHz actual core frequency and using 133MHz FSB allowed increasing the bus frequency only up to 142MHz. This way the maximum core working frequency we managed to achieve then made only 1917MHz. In other words, Athlon XP 2200+ allowed only 6.8% frequency increase over the nominal, which is definitely not enough to reach the speed of Athlon XP 2400+. Nevertheless, as we see, AMD didn't have any difficulty announcing a new Athlon XP 2400+, as well as Athlon XP 2600+, which core frequency is much higher. How on earth did the engineers manage it?

The answer is very simple: to find it just take a closer look at the results shown by wcpuid utility, which displays the info about the CPU in the system. The screenshot below was taken from the older (here we mean not outdated, but the previous) Athlon XP 2200+:

And this is a screenshot I took on a system with a new Athlon XP 2600+:

So, the first thing I would like to draw your attention to is the fact that new Athlon XP have a new core stepping. If the first Athlon XP based on Thoroughbred core had 0 core stepping, then the new processors feature core stepping equal to "1". Of course, the processors of the new core stepping are designed to allow new Athlon XP CPUs to work normally at higher core frequencies.

It's true, in this case it is all about Thoroughbred core redesign, which allowed AMD to increase Athlon XP core clock frequencies much more. When the company shifted from Thunderbird to Palomino core, they have already done something like that, however, this was not the only measure undertaken, as they also introduced a few changes to the processor fie architecture. Now the architecture remained unchanged. AMD has simply optimized its Thoroughbred.

Speaking more precisely I would like to point out that the mentioned above redesign added a new metallic layer to the processor die reducing the spurious capacitance. Moreover, the core structure got special decoupling capacitors responsible for combating electromagnetic interference. Also some changes were made to the positioning of the components on the die surface. As a result, AMD managed to retain the same architecture, and at the same time make it possible to increase the core clock frequency as well as reduce the heat dissipation of Athlon XP on Thoroughbred core quite tangibly.

The next table allows you to compare the features of the new and the old core:

  Athlon XP 2200+
Thoroughbred, stepping 0
Athlon XP 2600+
Thoroughbred, stepping 1
Core frequency 1800MHz 2133MHz
Bus frequency 266MHz (133MHz DDR)
Technology 0.13micron
Cache size L1=128KB, L2=256KB (384KB total)
Nominal Vcore 1.65V 1.65V
Die size 80sq.mm 84sq.mm
Number of transistors 37.2 million 37.6 million
Socket Socket A
Max. core temperature 85oC 85oC
Max. heat dissipation 67.9W 68.3W
Typical heat dissipation 61.7W 62.0W

As you can see, the new core is slightly bigger. However, if the core frequency grows up a lot, the heat dissipation increases just a tiny bit, which is very important. This way the redesign proved highly efficient, no doubt.

Another change introduced in the Athlon XP family together with the launching of the new Athlon XP 2400+ and 2600+ is made to the formula used to calculate the processor performance rating. In fact, there is nothing to be surprised at. Athlon XP processors get their rating basing on their performance in some reference benchmarks (the list of these benchmarks is available on AMD's site). Of course, as the frequency grows up the performance gain cannot remain constant. Since the CPU is not working with other system components any faster, because the processor bus frequency remains the same, the performance gain provided by the same repeating increases in the core frequency gets smaller and smaller. That is why on making new Athlon XP processors AMD discovered that 66MHz frequency growth is not enough to increase the CPU rating by 100. That is why the old formulas showing the dependence of the rating on the frequency do not work any more. To make things clear to you I would like to suggest looking at the table below, which illustrates the dependence of the performance raying of the contemporary Athlon XP processors and their core clock frequency:

CPU Frequency, MHz
Athlon XP 1500+ 1333
Athlon XP 1600+ 1400
Athlon XP 1700+ 1460
Athlon XP 1800+ 1533
Athlon XP 1900+ 1600
Athlon XP 2000+ 1667
Athlon XP 2100+ 1733
Athlon XP 2200+ 1800
Athlon XP 2400+ 2000
Athlon XP 2600+ 2133

It's true that if the frequency had been calculated with the older formula, then Athlon XP 2400+ worked at 1933MHz, for instance. However, its actual frequency appeared higher. As for Athlon XP 2600+, its frequency could make only 2.067MHz if calculated with the old formula. In reality the working frequency for this processor is 2133MHz. Well, a really pleasing trifle I should say :)

Rating Frequency, MHz Thoroughbred core, stepping 0 Thoroughbred core, stepping 1
Vcore Max. heat dissipation Typical heat dissipation Vcore Max. heat dissipation Typical heat dissipation
1700+ 1467 1.5V 49.4W 44.9W - - -
1800+ 1533 1.5V 51W 46.3W - - -
1900+ 1600 1.5V 52.5W 47.74W - - -
2000+ 1667 1.6V
1.65V
60.3W 54.7W 1.6V 61.3W 55.7W
2100+ 1733 1.6V 62.1W 56.4W - - -
2200+ 1800 1.65V 67.9W 61.7W 1.6V 62.8W 57W
2400+ 2000 - - - 1.6V
1.65V
65.3W
68.3W
59.3W

62W
2600+ 2133 - - - 1.65V 68.3W 62W

New Thoroughbred core will be used not only in Athlon XP 2400+ and 2600+, but also in the previously launched Athlon XP 2200+ and 2000+. Unfortunately, there are no evident differences in the marking between the CPUs based on the older and the newer core, even despite pretty noticeably differences in terms of overclocking and heat dissipation. However, you still can easily distinguish between the CPUs on the old Thoroughbred core and a new one with the help of a few additional tricks. The table below contains electrical and thermal specs for the existing Athlon XP CPUs based on different steppings of the Thoroughbred core:

As you can see, there will be only two models based on different cores: Athlon XP 2200+ and 2000+. Moreover, most Athlon XP 2000+ and 2200+ processors built on the new core stepping will boast lower nominal voltage (Vcore=1.6V). These CPUs should be your choice once you decided you might be willing to do some overclocking one day. The nominal Vcore can be easily read from the CPU marking:

In conclusion to our story about the peculiarities of the new Thoroughbred core with 1 stepping, I would like to stress one more fact. New Athlon XP 2400+ and 2600+ support 15x and 16x clock frequency multipliers respectively. Unfortunately, now not all mainboards know to recognize this clock multiplier correctly, so I suggest checking if your mainboard can work with a processor like that or not, before buying yourself a new CPU. As for us, we ran the tests of the new Athlon XP processor on EPoX EP-8K3A+, which proved excellent. So, this is the first mainboard that for sure works fine with the newcomers.

Overclocking

I should say that the results shown by the previously tested Athlon XP on Thoroughbred core (0 stepping) can hardly be called impressive. This is one of the reasons why AMD introduced a new core with the stepping equal to 1. This core can work at higher core frequencies, so that we have every reason to expect more from the new Athlon XP 2400+ and 2600+.

I would like to remind you that the maximum frequency I managed to achieve during overclocking of Athlon XP 2200+ based on the older core revision was 1917MHz. The new processors proved much more efficient.

This time we used the new Athlon XP 2600+ for our overclocking experiments. Its default working frequency makes 2133MHz. We started raising the FSB frequency smoothly from the standard 133MHz and reached 145MHz. However, the system proved not very stable in this case and kept crashing during the tests and 3D games. To be fair I should point out that we system got started even at FSB equal to 150MHz, however, we couldn't even load the operation system in this case. The curious thing about it is that the increase in the processor Vcore failed to help improve stability. That is why for the sake of stability, we had to reduce the FSB frequency to 144MHz. Then the system performed absolutely impeccably.

As you can see from the screenshot above, the actual core clock of our guinea pig made 2316MHz. This way, we managed to overclock the CPU by 8.6%. This is a slightly better result than in case of Athlon XP 2200+ on Thoroughbred core with 0 stepping: as you remember, the CPU overclocked only by 6.8%. as a result, overclocking allowed our today's hero to reach the potential working frequency of the upcoming Athlon XP 2800+.

It is important to understand that overclocking is not just a fun thing. It also demonstrates to us the features of the processor architecture. So, I dare state that thanks to a new Thoroughbred core stepping AMD will be able to launch a 2800+ CPU even without speeding up the bus to 333MHz. So, AMD still has a few trumps left in the battle with Intel for the title of the fastest x86 CPU maker.

Testbed and Methods

Since the architecture of the Thoroughbred core (1 stepping) remained the same, it is only due to core clock frequencies that the new Athlon XP 2400+ and 2600+ will outperform Athlon XP 2200+. However, you should nevertheless bear in mind that if the difference between Athlon XP 2400+ and 2600+ makes 133MHz, Athlon XP 2400+ is 200MHz faster than Athlon XP 2200+.

During this test session we will try to figure out the performance of the faster AMD Athlon XP processors against the background of the fastest Pentium 4 processors from Intel. At the same time we will compare the performance of Athlon XP 2400+ and 2600+ with that shown by their predecessor, Athlon XP 2200+.

To make the testing conditions equal for all testing participants, w assembled DDR333 platforms. Athlon XP platform was based on the today's fastest VIA KT333 chipset, and Pentium 4 platform was built on i845G, which is the fastest DDR333 chipset for Socket478 CPUs, as to our experience.

As a result, we got the following systems:

  Intel Pentium 4 AMD Athlon XP
CPU Intel Pentium 4 2.53GHz (533MHz QPB)
Intel Pentium 4 2.4B GHz (533MHz QPB)
AMD Athlon XP 2600+ (2133MHz)
AMD Athlon XP 2400+ (2000MHz)
AMD Athlon XP 2200+ (1800MHz)
Mainboard MSI 845GMax (i845G) EPoX EP-8K3A+ (VIA KT333)
Memory 512MB PC2700 CL2 DDR SDRAM
Graphics Card VsionTek Xtasy GeForce4 Ti4400
HDD IBM DTLA 307015

All tests were run in MS Windows XP Professional. The mainboards were configured to show the maximum performance (we set the minimal timings).

Since there are much more benchmarks to discuss, we decided to split them all into several groups for your convenience.

Performance

Office Applications and Data Encoding



In fact, AMD Company doesn't recommend measuring the performance of its processors with SYSmark 2002 benchmarks set. No wonder, as many applications included in it use SSE2 instructions, which are not supported by Athlon XP processors. That is why Intel manages to perform much better here.

The sound compression into mp3 format is performed by Athlon XP pretty fast. The new 2400+ and 2600+ processors cane easily outpace the today's fastest Pentium 4 2.53GHz CPU.

And in case of video compression by DivX codec, Pentium 4 CPUs regain their leadership. The new fifth version of the codec we used for our tests supports SSE2 instructions. Besides, video processing algorithms are very sensitive to the data transfer rate between the processor and the memory. Of course, due to faster processor bus Pentium 4 processors outperform Athlon XP here, because their faster bus allows them to use to the full extent the bandwidth of DDR333 SDRAM. The ongoing Athlon XP processors, which will support 333MHz bus, have every chance to become a more serious threat to Pentium 4 here.

In archiving applications Pentium 4 again performs slightly better. Bearing in mind that this process is also very critical to the memory subsystem performance, we surely understand that the problem of getting Athlon XP processors a new faster bus is very acute.


Although PCMark2002 is a synthetic benchmark, I placed it in this particular section, because the algorithms its uses to tests the CPU performance include JPEG decompression, LZ77 compression and decompression, text retrieval and audiostream transformation.

The results obtained in CPU benchmark show that Athlon XP 2600+ is a bit faster than Pentium 4 2.53GHz. in fact, AMD could have taken this benchmark as a reference for the processor rating calculations.

As for the memory subsystem performance, Pentium 4 processors leave Athlon XP behind due to faster bus, even despite the similar memory used in both platforms. Although the memory bandwidth of Athlon XP system equals to 2.7GB/sec, the 2.1GB/sec processor bus prevents the CPU from taking full advantage of the memory bus. Pentium 4 CPUs have a much faster processor bus: 4.2GB/sec data transfer rate, which definitely tells positively on the work with the memory.

3D Games

Here Athlon XP is just a little bit behind Pentium 4 2.4GHz.

In Return to Castle Wolfenstein Pentium 4 is again a bit ahead. However, you shouldn't forget that this game is based on Quake 3 engine, which is pretty loyal to Intel processors.

Something of the kind can be observed in a popular helicopter flight simulator: Comanche 4.

In the new role game aka Dungeon Siege Athlon XP 2600+ nearly catches up with Pentium 4 2.53GHz.

And this is the second version of the shooter called Serious Sam: The Second Encounter. This game is known for its "pro-Athlon" mood. All tested Athlon XP processors on Thoroughbred core manage to beat Pentium 4 2.53GHz completely.

This time I also decided to test our CPUs with a new demo of the upcoming Unreal Tournament 2003 hit. And as we can see, the results appeared generally in Athlon's favour. In other words, the performance show corresponds exactly to the processor ratings of the CPUs tested. Athlon XP 2600+ proved the fastest in Unreal Tournament 2003.

3D Rendering

We investigated the 3D rendering speed in three popular testing packages: 3ds max 4.26, Maya 4.0.1 and new version of Lightwave - Lightwave v.7.5. In all the tests we timed the scenes rendering speed that is why smaller values on the diagrams correspond to better performance. To test in 3ds max we used the islands scene, for Maya 4.0.1 we used Maya-Testcenter rendertest methodology and in Lightwave we used sunset and raytrace scenes.

After the update to 3ds max version 4.26 came out, this test set acquired high-quality support of SSE2 instructions. Therefore, no wonder that Intel processor will work faster in this rendering test until 8th generation Hammer processors with the implemented SSE2 support come out.

Maya is not optimized for any processors that is why it shows that Athlon XP 2400+ is even faster than Pentium 4 2.53GHz.

I have already mentioned that Intel recommends Raytrace scene as a Pentium 4 performance test. It is not for nothing, I should say: the results speak for themselves.

When the performance is tested with the scenes not recommended by Intel, the results appear just the opposite. Athlon XP 2200+ working at 1.8GHz easily outpaces Pentium 4 2.53 working at much higher core clock.

Scientific and Professional OpenGL Applications

To test the performance of the new AMD CPUs in scientific tasks we resorted to ScienceMark 2.0 test (read more about this test here). On my part I would like to point out that this benchmark was independently optimized for Athlon XP as well as for Pentium 4. The diagrams for this test show the time each CPU required to complete the tasks, so smaller value stands for higher performance.



It is quite natural that Athlon XP family proves more successful in math1ematical modeling tasks. In fact, we haven't expected anything different. Only in molecule modeling, where high data transfer rate between the CPU and the memory is required, Pentium 4 platform can gain revenge from Athlon XP.







The benchmarks included into SPECviewper test set have always been a trump of Athlon XP processors. The reasons are the same: the algorithms used in these benchmarks are pretty outdated and do not have any SSE2 instructions. And intensive calculations are something Athlon XP can fairly be proud of.

We also tested the CPUs in one more benchmark - SPECapc for 3dx max 4.26. This new test allows estimating the performance in viewports in the latest version of 3dx max 4.26. The test uses four different scenes and animates them within the program environment. For more information about the test, look here.



The computational rating of Athlon XP is high, but during visualization higher data transfer rate between the processor and memory, as well as the support of new SSE2 instructions hand the laurels over to Pentium 4. As a result, however, Athlon XP 2600+ still fails to outperform Pentium 4 2.53GHz.

Conclusion

The today's launch of Athlon XP 2600+ and Athlon XP 2400+ allowed AMD to demonstrate that they don't have any problems with the 0.13micron manufacturing technology as well as with the Thoroughbred core. Having made a sudden dash forward the company managed to prove the world that Athlon XP still remains a worthy competitor to Pentium 4.

However, unfortunately, the increase in the core clock frequency is not so efficient without the support of faster processor bus. As a result, in many applications Athlon XP 2600+ yields to Pentium 4 2.53GHz in performance. In fact, Athlon XP proves the winner only in those applications, which require high computing power and do not involve SSE2 instructions. That is why it is quite evident that the transfer to faster 333MHz bus has already become essential. AMD also understands it, so we expect Athlon XP to get a faster bus very soon now, which will surely raise its chances in the competition with Pentium 4. 

Pages: [ 1 ]

Discussion

Comments currently: 2
Discussion started: 10/20/04 10:44:34 AM
Latest comment: 06/26/05 06:32:58 PM

View comments

Add your Comment




Latest materials in CPU section

Article Rating

Article Rating: 7.2727 out of 10
 
Rate this article:
Excellent
Average
Poor