AMD's plans have been changing too frequently lately. During the last half a year this company managed to pleasantly surprise us with the launch of the new Thoroughbred processor core revision with much higher supported clock frequencies, but they also managed to upset us quite a bit with the delay of the so long awaited Athlon 64 (ClawHammer). As a result, a new processor for the desktop market featuring x64-86 architecture is expected to start selling only this fall, and until that time the major player from AMD in this market segment will remain the good old Athlon XP. However, Intel Pentium 4 with the newly introduced Hyper-Threading technology and the upcoming transition to 800MHz bus have pushed AMD to make a few enhancements to their Athlon XP to help it compete with Intel Pentium 4 on equal terms. Now AMD Athlon XP processors will boast bigger L2 cache memory (it will be increased from 256KB to 512KB). The codename for the new Athlon XP processor core with the bigger L2 cache is Barton. So, today we are going to pay special attention to this core and its peculiarities and performance.
However, before we start, we would like to speak a bit more about AMD's current plans. So, we you know the launching of Athlon 64 has been postponed until September. There are two reasons for this move. Firstly, AMD is still experiencing some difficulties with the production of processor cores with x64-86 architecture. The CPUs, which are currently manufactured on AMD's Fab30 in Dresden can't boast such working frequencies, which could let them outperform the today's fastest Pentium 4 as well as Athlon XP processors. This way, it doesn't make much sense to introduce Athlon 64 CPUs now, as it may kill the sales of AMD Athlon XP processors, but at the same time will be unable to become a serious competitor to the upcoming Intel CPUs, such as the new Pentium 4 3.2GHz with 800MHz bus, which will come out in the second half of April already. Secondly, there is no software yet to use all the advantages of the new x64-86 architecture, which deprives Athlon XP of its one more trump. And thirdly, AMD has a Barton core, which can help the company to retain its positions in the market for another while, at least until Athlon 64 becomes a more competitive CPU.
However, the server version of x86-64 processor from AMD, Opteron, will be released in April already. The "pure" performance doesn't matter that much in the server market that is why the new Opteron processor working at some 1.8GHz has every chance to become a popular and demanded product. Moreover, server operation systems supporting x64-86 are already available, so dual-processor servers built with Opteron CPUs will definitely become a success.
As for the further development of the Athlon XP family, namely the Barton core, we are glad to state that today AMD has finally announced Barton based CPUs rates as 3000+, 2800+ and 2500+. Together with the new processors launch, AMD also announced the change of their old black-and-green Athlon XP logo. From now on they will have a new logo designed in the same way as all other company processor logos.
Closer Look: New Barton Core
As we have already mentioned above, today AMD Company announced new Athlon XP 3000+, 2800+ and 2500+ processors based on Barton core. This new core appeared for evolutionary reasons: the working frequencies of the previous generation Athlon XP processors manufactured with 0.13micron technology have already reached their maximum. For example, Athlon XP 2800+ announced last October hasn't yet become a mass product. At the same time more progressive manufacturing technologies, which could allow AMD to keep increasing the core frequencies of its CPUs are not ready yet. 90nm technology is expected to be put into service only in 2004, and SOI technology still needs good polishing. Therefore, AMD had to find some other way of increasing its processors performance. And bearing in mind that the absolutely new x64-86 architecture will be introduced only in Athlon 64 processors, AMD engineers decided to create the new Barton by making a few simple enhancements and changes to the very successful and well-designed Thoroughbred core.
There were two possible ways of enhancing the Thoroughbred: increasing the processor bus frequency, which has already happened to Athlon XP processors starting from 2600+ model, and increasing the L2 cache memory size. AMD implemented only the second option in those Barton based Athlon XP CPUs, which were launched today. Their L2 cache has grown from 256KB, which was typical of all previous Athlon XP processors, to 512KB. As for the further bus frequency increase, it is still too early to talk about. From time to time we come across some rumors about Barton based Athlon XP processors acquiring faster 400MHz bus, and it looks as if these rumors were not absolutely ungrounded. However, AMD hasn't yet made up its mind about the use of 400MHz bus. The company engineers are only investigating the possibility of this frequency increase. If the results of these investigations show that processor bus frequency increase up to 400MHz will not harm the stability of the Barton core and will be really efficient for the processor performance, then maybe we will one day see new Socket A processors from AMD on Barton core supporting faster 400MHz bus.
The increase of L2 cache improved the performance of Athlon XP processors quite significantly. Therefore the core clock frequencies of the first CPUs based on the new Barton core didn't grow up compared with the frequencies of the latest Thoroughbred based Athlon XP models. Now the new Athlon XP 3000+ works at 2.167GHz, which is the same frequency as that of Athlon XP 2700+ on Thoroughbred core. The new Athlon XP 2800+ works at 2.083GHz, and Athlon XP 2500+ - at 1.8333GHz. All the new processors based on Barton core support 333MHz bus.
To tell the truth, the rating system of AMD Athlon XP processors has lost quite a bit of its harmony. To make the whole thing clearer to you, we made up a table with the ratings, and corresponding bus frequencies and L2 cache sizes for this processor family:
|Multiplier||FSB = 133MHz||FSB = 166MHz|
|L2 = 256KB||L2 = 512KB|
|13.5x||2200+ (1800MHz)||2800+ (2250MHz)|
|13x||2100+ (1733MHz)||2700+ (2167MHz)||3000+ (2167MHz)|
|12.5x||2000+ (1667MHz)||2600+ (2083MHz)||2800+ (2083MHz)|
|11x||1700+ (1467MHz)||2500+ (1833MHz)|
There are no other differences between the new Barton and the older Thoroughbred except the larger L2 cache. To make sure that this is true, all you need to do is to cast a glance at the Thoroughbred (B Stepping) and Barton core architecture schemes.
As we see, Barton differs from Thoroughbred-B only by a few more transistors, which serve to implement a larger L2 cache. Even from the structural point of view both dies look absolutely identical (except the cache, of course).
The increase in L2 cache size resulted into the growth of the processor die size. At the picture below you can see Athlon XP (Thoroughbred) on the left and Athlon XP (Barton) on the right:
Summing up everything mentioned above, we would like to offer you another table, where we compare the key features of Barton and Thoroughbred revision B cores:
|Athlon XP Ratings||1700+ - 2800+||2500+ - 3000+|
|Manufacturing technology||0.13micron copper compound technology, Fab30, Dresden|
|Cache size||L1 = 128KB|
L2 = 256KB
|L1 = 128KB|
L2 = 512KB
|Transistors||37.6 million||54.3 million|
|Max. core temperature||85oC||85oC|
|Max. heat dissipation||68.3W||74.3W|
Keeping in mind that Barton is very similar to Thoroughbred, it appears not at all surprising, that the L2 cache organization in the new Barton has also been left unchanged. Just like the one in Athlon XP on Thoroughbred core, the Barton L2 cache remained associative with 16 fields and a 64Byte data string. As a result, the L2 cache of Barton is as fast as L2 cache of Thoroughbred. The results of the L2 cache performance tests for Athlon XP 3000+ on Barton core are given below:
And for a better comparison, here are the results of L2 cache speed tests in Athlon XP 2700+ on Thoroughbred core working at the same core clock:
Both processors spend the same number of clocks for cache memory access, and the bandwidth measurement differences lie within the measuring error. Therefore, summing up the above made statements we dare claim that Barton is none other but the same Thoroughbred-B but with a larger L2 cache.
We won't be mistaken if we say that the compatibility of Athlon XP processors based on the new Barton core with the already existing mainboards is a question that concerns many of you, guys. AMD does its best to keep the Socket A platforms alive for as long as possible, therefore, it is not at all surprising that most Socket A mainboards will have no problems working with the new Barton based Athlon XP CPUs. In fact, all the specific requirements that the new Barton imposes over the Socket A mainboards are the support of 333MHz system bus and the availability of the processor voltage regulator capable of producing up to 45A of electric current. AMD claims that over 50% of mainboards supporting Athlon XP 2700+ will be able to work freely with the new Athlon XP 3000+.
Of course, to make the mainboards recognize the new processors correctly, BIOS update is required. The list of mainboards checked by AMD for compatibility with the new Athlon XP 3000+ currently includes the following models:
- ABIT KD7 (VIA KT400);
- ASUS A7S333 (SiS 745);
- ASUS A7V333 v1.04 (VIA KT333);
- ASUS A7V333 v2.0 (VIA KT333);
- ASUS A7N8X (NVIDIA nForce2);
- ASUS A7V8X v1.04 (VIA KT400);
- Biostar M7VIP (VIA KT333);
- Biostar M7VIK (VIA KT400);
- EPoX EP-8K5A2 (VIA KT333);
- EPoX EP-8K9A2 (VIA KT400);
- Gigabyte GA-7VR v2.0 (VIA KT333);
- Gigabyte GA-7VAXP v1.0 (VIA KT400);
- Gigabyte GA-7VAX v1.1 (VIA KT400);
- Gigabyte GA-7VA v1.0 (VIA KT400);
- Jetway V333DA (VIA KT333);
- Jetway V333U (VIA KT333);
- MSI KT4 Ultra (VIA KT400);
- MSI MS-6596 (VIA KT400);
- MSI MS-6712 (VIA KT400);
- MSI MS-6382E (VIA KT333);
- MSI MS-6561 (SiS 745);
- MSI MS-6593 (VIA KT333).
Heat Dissipation: Thermal Conditions and S2K Bus Disconnect
Since the new Barton features more transistors and bigger die compared with its predecessor, the heat dissipation has also grown up a bit. However, there is nothing dramatic about it: in case of 20% die size increase the heat dissipation of the new Barton (working at the same clock frequency as Thoroughbred) got only 9% bigger:
|Model||Core||Frequency, MHz||Voltage, V||Typical Heat Dissipation||Max. Heat Dissipation||Max. Core Temperature, oC|
As for the maximum heat dissipation of the new Athlon XP 2800+ and 2500+ based on Barton core, it doesn't differ from that of the last Thoroughbred based CPUs at all. However, Athlon XP 3000+ is a much "hotter" CPU, showing heat dissipation close to that of Thoroughbred 2800+. This is exactly the reason why Barton CPUs with 2800+ and 2500+ ratings do not require any special cooling and feel quite OK with the same cooling solutions as the fastest Thoroughbreds. As for Athlon XP 3000+, this one requires more advanced cooling with the thermal resistance of no more than 0.57deg/W.
So far AMD recommends the following five coolers to be used with the new Athlon XP 3000+ (Barton) processors:
- Ajigo MF034-032;
- AVC 112C86FBH01;
- Dynatron DC1206BM-L/610-P-Cu;
- Fannertech Spire SPA07B2;
- Taisol CGK760172.
In fact, the listed coolers do not boast any monstrous look and feature no huge heatsinks and high-speed fans. Here is, for instance, a picture of Dynatron DC1206BM-L/610-P-Cu, which is most likely to be shipped with the boxed Athlon XP 3000+ processors:
The key specification of this cooler is not its big size, but the copper foot and a lot of thin ribs of the heatsink.
Moreover, with the launch of the new Barton based processors, AMD decided to start setting mainboard guys on the right track. A while ago AMD forced the mainboard manufacturers to implement a special CPU thermal protection scheme using the integrated thermal diode. If this hadn't been done, the mainboard wouldn't have been certified by AMD. As we see, the results are evident: most mainboards available in the today's market do have a CPU thermal protection scheme.
The second move on AMD's way had to do not with the CPU protection against burning, but with the temperature reduction during work. Now AMD Company will require all mainboards applying for certification to support S2K Bus Disconnect function, which will allow reducing the average consumed power and the heat dissipation of the CPU in most Windows applications without any performance losses. The S2K Bus Disconnect implementation implies the following. During HALT command (HALT means the CPU will be stopped because there are no instructions to be processed), the CPU can be switched to the corresponding waiting mode (Halt and Stop Grant) with lower power consumption and heat dissipation. However, Athlon XP also required System Bus Disconnect to be able to switch to the lower power consumption state. In fact, this should be implemented in the mainboard chipset and BIOS. But until recently, the BIOS of most mainboards used to be configured in such a way that Athlon XP never got to the lower power consumption state. As a result, even in idle mode the temperature of all Athlon XP processors remained pretty high.
Now the situation should change drastically and Athlon XP processors will become much cooler on those mainboards, which will support S2K Bus Disconnect. Many today's chipsets, such as VIA KT400, VIA KM400, SiS 746 and NVIDIA nForce2 do support S2K Bus Disconnect without any problems. There have already appeared the first mainboards, where the Bus Disconnect feature may be enabled in the BIOS. So far there are only five mainboards like that. They are: ASUS A7V8X v1.04, EPoX EP-8K9A2, Gigabyte GA-7VAXP v1.0, Gigabyte GA-7VAX v1.1 and Gigabyte GA-7VA v1.0. However, since they do not certify any more mainboards without the support of Bus Disconnect function, this list should very soon grow longer.
To illustrate everything mentioned above, and to show how much warmer are the Barton based processors than the Athlon XP ones based on the Thoroughbred core, we ran a few tests with the new Athlon XP 3000+ (Barton) and Athlon XP 2700+ (Thoroughbred). As you remember, these CPUs work at the same core clock frequency equal to 2167MHz.
We measured the processor temperature on two mainboard revisions: ASUS A7V8X 1.02 without Bus Disconnect support and ASUS A7V8X 1.04 with Bus Disconnect support. The tests were run in Windows XP operation system. The temperatures were taken from the thermal diode integrated into the cores of all Athlon XP processors.
First of all we measured the CPU temperature in Idle mode.
As we see, enabling S2K Bus Disconnect function has a great effect. The CPU temperature drops by 15oC in both cases: by Barton and by Thoroughbred based processors. However, Barton featuring a bigger number of transistors still appears about 6oC warmer than its rival.
Now let's check how both CPUs are going to behave under some workload. To warm the babies up we used a well-known BurnK7 utility.
In this case S2K Bus Disconnect function appears inefficient. In fact, this is not surprising at all. BurnK7 loads the CPU so heavily that the operation system simply lacks time to send the HALT command, when the CPU could cool down a bit. In other words it means that S2K Bus Disconnect doesn't work if the CPU is under steady workload. But this is a hypothetical situation. Most PCs used for office needs stay idle waiting for more data to be processed 95% of their working time. As for the temperature difference between Barton and Thoroughbred in BurnK7, it makes only 8oC.
In order to evaluate the average CPU temperature under regular workload, we checked the status of the tested processors during SYSmark2002. This benchmark models exactly the work of an ordinary user in typical office and digital content creation applications. The list of apps included into the SYSmark 2002 testing set looks as follows: Microsoft Word 2002, Microsoft Excel 2002, Microsoft PowerPoint 2002, Microsoft Outlook 2002, Microsoft Access 2002, Netscape Communicator 6.0, NaturallySpeaking v.5, McAfee VirusScan 5.13, WinZip 8.0, Macromedia Dreamweaver v4.0, Adobe Photoshop 6.0.1, Adobe Premiere 6.0, Macromedia Flash v5 and Microsoft Windows Media Encoder 7.1. The average CPU temperature during SYSmark2002 is presented on the diagram below:
And again the advantages of the new Bus Disconnect function are evident. When activated, the processor temperature drops down by 15-17oC. And this happens during the real functioning! However, Barton core still appeared warmer than Thoroughbred. The temperatures of these two cores working at the same frequencies differed by 6-9oC depending on the Bus Disconnect mode. If you are curious to see the temperature dynamics during SYSmark2002 test, please see the next diagram built according to the temperature log-file:
This way, this Bus Disconnect function allows reducing the processor temperature quite significantly without bothering the user and hindering proper system functioning. If the mainboard guys support this initiative and add the fan rotation speed control option, which would work depending on the CPU temperature, then we will have every chance to see the first quiet platforms based on high-performance AMD processors very soon.
Price and Availability
New Athlon XP processors based on Barton core will cost considerably more than their predecessors. The official price set for the new Athlon XP 3000+ will equal $588, for Athlon XP 2800+ - $375 and for Athlon XP 2500+ - $239. However, this doesn't at all imply that AMD is experiencing any problems with the Barton based processors manufacturing. The simple calculations show that with 200-mm wafers used at AMD's fab in Dresden, the production costs of Barton dies get only 20% higher than the costs of Thoroughbred-B dies, provided the production yields are the same. We don't expect the production yields of Barton dies to be lower than those of Thoroughbred-B, because they use the same manufacturing technology, and the dies are very similar to one another, which we have already mentioned in the beginning of our article. So, high price of AMD Athlon XP on the new Barton core can be explained only by some marketing reasons and hence can be easily reduced if the situation in the processor market changes. This is exactly the reason why AMD may start manufacturing Athlon XP processors on Barton core the ratings below 2500+ and they will not lose anything. Anyway, it is still too early to say if it ever happens or not.
We all remember that the announcements of the last Athlon XP processors based on Thoroughbred-B core were mostly "paper announcements". These announcements didn't imply that the CPUs would appear in stores. Quite a bit of time had to pass before the newly announced CPUs finally started selling. Sometimes, we had to wait for a few months even. But is it possible that the story repeats now with the new Athlon XP on Barton core? I believe that this is the question that concerns many of you.
Luckily, we have every reason to state that today's announcement is backed up not only by AMD's vital desire not to fall behind Intel, but also by a real opportunity to produce a sufficient amount of new processors. So, Athlon XP 3000+ and Athlon XP 2800+ are about to start selling in stores in the nearest future. As for Athlon XP 2500+ also based on Barton core, it will be available a bit later for marketing reasons. Probably, by the end of Q1 2003.
To evaluate the potential of the Barton core in terms of working frequencies increase, we undertook a number of experiments with the fastest Athlon XP 3000+ based on this core. As you remember, the nominal core clock of this processor is 2167MHz. however, you should also keep in mind that AMD is going to release one more processor model based on Barton core and rated as 3200+. That is why this core simply should have some clock frequency reserves. So our primary task now will be to discover how big these reserves are.
Before we pass over to the practical part of our overclocking session, we would like to point out one more thing. Since Barton core is quite similar to Thoroughbred-B in its architecture and structure, we expect it to be able to overclock up to the same top frequency as Thoroughbred-B does. In other words, since we know that the maximum frequency Thoroughbred-B based Athlon XP processors can work at equals 2.25GHz, the Barton based CPUs should be able to reach some similar limits.
As for the Barton clock frequency multiplier, it is organized in a similar way. The pieces we got for review this time features a locked frequency multiplier, however, we could easily unlock it by closing the last bridge in L3 group. Moreover, those mainboards, which know to unlock the multiplier by Thoroughbred based Athlon XP processors (these are NVIDIA nForce2 based ones in the first place), can cope with Barton multiplier unlocking in no time. It means that Barton overclocks just the same was as Thoroughbred does.
Athlon XP 3000+ overclocking was carried out by raising the FSB frequency. We managed to reach 175MHz FSB frequency by increasing the Vcore a bit (up to 1.75V). Further FSB frequency growth led to inevitable instability of the whole system during the basic tests.
The maximum frequency obtained equaled 2280MHz, which is only 30MHz higher than in case of overclocking the fastest Athlon XP on Thoroughbred-B core (2800+). This way, our forecasts appeared absolutely correct: maximum core clocks of Thoroughbred-B and Barton are very close to one another.
Testbed and Methods
These tests were intended to help us compare the performance of the new Athlon XP processors based on Barton core with the performance of the older Athlon XP CPUs on Thoroughbred core and with that of their competitors from Intel: Pentium 4 processor family. The platform for Socket A processors was built on NVIDIA nForce2 and equipped with dual-channel DDR333 SDRAM, because this particular chipset combined with this particular memory is the today's fastest configuration. As for Pentium 4 processors, they were tested with a E7205 based mainboard working with dual-channel DDR266 SDRAM. This combination ensures high performance without the slowly dying out RDRAM.
As a result, our testbeds looked as follows:
|Intel Pentium 4||AMD Athlon XP|
|CPU||Intel Pentium 4 3.06GHz (HT, 533MHz QPB)|
Intel Pentium 4 2.8GHz (533MHz QPB)
Intel Pentium 4 2.66GHz (533MHz QPB)
Intel Pentium 4 2.53GHz (533MHz QPB)
|AMD Athlon XP 3000+ (Barton)|
AMD Athlon XP 2800+ (Barton)
AMD Athlon XP 2800+ (T-bred)
AMD Athlon XP 2700+ (T-bred)
AMD Athlon XP 2600+ (T-bred, 333MHz FSB)
AMD Athlon XP 2600+ (T-bred, 266MHz FSB)
AMD Athlon XP 2500+ (Barton)
|Mainboards||MSI GNB Max (Intel E7205)||EPoX EP-8RDA (NVIDIA nForce2)|
|Memory||Crucial XMS3200 CL2 DDR SDRAM, 2x256MB|
|Graphics Card||ATI RADEON 9700 Pro|
|HDD||Seagate Barracuda ATA IV, 80GB|
All the tests were run in MS Windows XP Professional SP1, and the BIOS Setup of the mainboards participating was configured to grant highest performance possible.