Having tested the first mainboards based on the DDR chipsets for Socket A platforms, AMD-760 and ALi MAGiK 1, we noticed that the use of DDR SDRAM, which featured twice the bandwidth of the regular PC133 SDRAM, provides a not very significant performance boost. When we tested Gigabyte GA-7DX mainboard based on AMD-760 chipset we managed to get a 15% performance gain, while in case of Iwill KA266-R based on ALi MAGiK 1 chipset the increase made only 10% over the performance shown by the regular mainboards on VIA KT133 chipset.
Why did the performance grow that slowly? The answer to this question is very simple. The thing is that memory bus bandwidth plays the major part only in a limited number of applications processing large amounts of data. In all other cases, the latency (it's a period of time between the data request and its actual fulfillment) appears much more important. Take, for instance, the performance of the Pentium III systems with RDRAM based on i820 chipset. Even though RDRAM features higher bandwidth than PC133 SDRAM, the systems with SDRAM beat those with Rambus memory almost in all applications. Why? Because of the RDRAM's high latency.
And what about the latency of DDR memory? Take a look at the chart below to find out the answer:
As you can see from this picture, PC133 SDRAM CAS2 features the lowest latency of all the today's memory types used. PC2100 DDR SDRAM CAS2 features a bit higher latency. It is important to understand that DDR SDRAM technology doesn't imply that the memory bus bandwidth is simply doubled due to the fact that the data is transferred along both signal edges. Unfortunately, the today's technologies do not allow reducing the memory access time in the same proportion. As a result, it takes DDR SDRAM the same time to start responding to data requests as in case of the ordinary SDRAM. That's why some applications running on systems with the DDR memory are hardly any faster than if they were running on systems with the ordinary SDRAM.
Moreover, most PC2100 DDR SDRAM available in the today's market is unfortunately only CAS 2.5. The graph above doesn't show the latency of this memory, however, if we perform some simple arithmetic calculations, we will be able to conclude that it appears closer to the latency of PC1600 CAS 2 than to PC2100 CAS 2. So, you can easily imagine a situation when a system equipped with PC133 SDRAM works faster than a similar system equipped with PC2100 DDR SDRAM. Everything depends on the application. If the application requires accessing large data packs in consecutive order, then the twofold advantage of the DDR memory will provide a significant performance increase. However, if the application does not access the data consecutively, the positive effect made by the high bandwidth of the DDR memory will vanish and it will be the latency of the memory subsystem that will contribute most to the overall system performance.
In order to make everything clear to you, let us offer you a very simple example. Imagine a running competition between two athletes. The first one (representing the DDR memory) runs twice as fast as his rival, however, he starts running later than his rival. Who wins the race depends on the distance the same way as the performance in this or that application depends on the type of this application and on the way the memory requests are carried out.
In the future, when the programs are processing more and more data, DDR SDRAM will prove more and more efficient. However, now you can easily find such applications, which care much more about the memory latency rather than its bandwidth.
This is one of the reasons why it is too early yet to give up PC133 SDRAM. One more thing, which may prevent DDR memory from getting widely spread in the nearest future is its cost. Even though manufacturing DDR DIMM modules is only 5-10% more expensive than manufacturing PC133 SDRAM DIMM modules, those pieces available now in retail cost almost twice or even three times as much as the regular SDRAM modules do. Moreover, it is still pretty complicated to buy these DDR modules, since not every store offers them.
Well, everything mentioned above as well as the fact that DDR SDRAM is forecast to win only 30% of the memory market by the end of the year means that it would be still unreasonable to give up developing and manufacturing chipsets with PC133 SDRAM support. As far as Pentium III and Pentium 4 platforms are concerned, Intel will promote the chipsets for them supporting PC133 SDRAM. The recently launched i815EP for Pentium III is getting more and more popular day by day, and Brookdale, which should also support SDRAM is due in Q3 2001. Speaking about AMD Athlon platform, VIA has taken this role upon itself and upgraded its KT133 Socket A chipset having added 266MHz FSB support to it.
As a result, new core logic appeared: VIA KT133A, which consisted of the enhanced VT8363A North Bridge and the already familiar to you VT82C686B South Bridge. We won't dwell on the features of the new chipset, because they are just the same as those of the well-known VIA KT133 one. The only difference between these two chipsets is the 266MHz FSB support in VIA KT133A, which is required by all the new AMD Athlon-C processors working at 1GHz, 1.13GHz and 1.2GHz and will be required for the upcoming Athlon processors based on both: Thunderbird and Palomino cores. The two North Bridges are so similar to one another that they were even constructed pin-compatible. It means that most mainboard manufacturers do not have to redesign their older mainboards based on VIA KT133 chipset. They will simply change the North Bridge and offer the world new products supporting 266MHz FSB.
EPoX EP-8KTA3 belongs to a well-known family of Socket A mainboards from this manufacturer. Here are its main specs:
|Supported CPUs||AMD Athlon/Duron (200/266MHz)|
|Chipset||VIA Apollo KT133A (VT8363A + VT82C686B)|
|FSB Frequencies||95, 100, 102, 104, 106-114, 116, 118, 120, 124, 127, 130, 133, 136, 140, 145, 150, 155, 160, 166MHz|
|Overclocking Friendly Features||Vcore, Vagp and Vio changing|
|Memory||4 168-pin DIMM slots for PC100/PC133 SDRAM|
|Expansion slots (AGP/PCI/ISA/AMR)||1/6/1/0|
|Additional features||Integrated P80P Debug (POST) controller|
|BIOS||Award Modular BIOS v. 6.00PG|
We would like to point out right away that there is also another modification of the same mainboard: EP-8KTA3+, which is also equipped with an integrated ATA/100 RAID controller, HPT370. Besides, the mainboard we had at our disposal featured a layout for the second Flash memory chip that's why we suppose that EP-8KTA3+ will also support something like DualBIOS technology, even though the official mainboard specs do not mention anything of the kind.
EPoX turned out one of the first mainboard manufacturers to introduce its products based on VIA KT133A chipset. Nevertheless, EPoX didn't choose the simplest way, even despite the pin-compatibility of KT133 and KT133A, which we have already mentioned in the article. 8KTA3 features an absolutely new mainboard design. Therefore, there are a few noticeable changes compared to the previous product, EP-8KTA2. Namely, the new board got an additional DIMM slot, P80P POST controller and the possibility to get an integrated IDE RAID controller.
Also thanks to the new VIA KT133A chipset used, EP-8KTA3 supports AMD Athlon/Duron processors with 200MHz FSB as well as new Athlons with 266MHz FSB.
A very important peculiarity of EP-8KTA3, which distinguishes it from all the previous mainboard models as well as from many competing products by other manufacturers, is the CPU power supply. Taking into account the upcoming launching of Thunderbird processors with the frequencies up to 1.5GHz, it was redesigned in such a way that it could support much more power-hungry CPUs. The new power supply layout uses relatively big 4700uF capacitors reducing the noise pollution of the signal.
However, everything was not that perfect. Since all the elements used in the power supply circuit of the mainboard are of pretty large size and located close to the processor socket, massive coolers appear very hard to install onto the CPU on EP-8KTA3. In particular, you won't be able to use any of the popular Chrome Orb or Super Orb coolers. Moreover, even installing a CPU into the processor socket appears quite a hard thing to do, because the capacitors placed too close to the socket will prevent you from accessing it that easily.
In fact, this happened, because EPoX tried to keep the small PCB when installing all those controllers, so that to make it fit for any ATX case. That's why all EP-8KTA3 components are pressed very close to one another.
The memory slots are situated right at the processor socket. We should draw your attention to the fact that EPoX EP-8KTA3 appeared nearly the first Socket A mainboard equipped with 4 DIMM slots. However, it doesn't at all mean that the mainboard can support up to 2GB PC133 SDRAM. In reality, the third and the fourth DIMM slots are shared and if used simultaneously, allow installing only single-side memory modules. The matter is that VIA KT133A chipset supports only 6 memory banks that is why the maximum supported memory makes only 1.5GB in any case.
Like with mainboards based on VIA KT133, the memory used with EPoX EP-8KTA3 can be clocked for the frequency equal to that of the FSB or FSB+PCI, which makes sense for the CPUs with 200MHz front side bus. So, VIA KT133A appeared the first chipset supporting 266MHz FSB, which does allow using the asynchronous memory bus to the full extent. As we saw by AMD-760 and ALi MAGiK 1 chipsets, the memory bus and the processor bus frequencies are fixed with each other. This gives us some hope that DDR Socket A chipset from VIA, the upcoming KT266, which is due in early spring will also allow DDR SDRAM and EV6 processor bus to function asynchronously.
The universal AGP slot on EPoX EP-8KTA3 supports 1.5V and 3.3V AGP 2x and 4x graphics cards. It is equipped with an original retention mechanism locking the installed AGP-cards in the correct position. Unfortunately, this slot is placed so close to the DIMM slots, that in case the graphics card is installed, you will be unable to insert/remove the memory modules.
The mainboard features 6 PCI slots, which is quite enough to satisfy any pretentious user. We were also very pleased to see that there is an ISA slot on the board, which has become a very rare thing recently, because most mainboard manufacturers prefer to equip their products with AMR or CNR slots rather than with ISA targeting their products for the OEM market in the first place.
As for the general mainboard layout, the lack of free room around the processor socket is not the only drawback of the EP-8KTA3. Besides that, EPoX engineers didn't stick to the ATX specification requirements when placing the power supply connector as well as the IDE connectors onto the PCB. Even though the IDE connectors are located very conveniently for Tower cases, it is unacceptable for Desktops. Moreover, the power supply cable going to the connector, which is located almost in the middle of the PCB, will hardly arouse any positive emotions.
We should separately mention the integrated P80P Debug (POST) controller on EPoX EP-8KTA3, which is a typical feature of many other EPoX mainboards aimed at progressive users. The main working principle of this controller is based on the following. In the beginning, during the execution of each initialization the BIOS puts to port 0080h a code uniquely determining the initialization type and the device being initialized. In case the operation succeeds, BIOS starts initializing the next device. If the device initialization fails, then the next procedures do not start and the BIOS either gives up or repeats the attempt. Anyway, the two-digit indicator in the lower left corner of the mainboard shows the code of the last device initialized. With the help of a special code table provided in the user's manual you can find out, which device caused the failure. So, this POST controller allows simplifying the assembling and debugging of the system. In fact, EPoX's approach provides much more information on the occurring problems than the D-LED or voiced diagnostic systems, even though it is not so illustrative for an inexperienced user as the alternative solutions.
For the South Bridge, EPoX EP-8KTA3 uses VIA 686B, which means that the mainboard supports ATA/100 interface. The same South Bridge of 8KTA3 is responsible for the hardware monitoring. The mainboard supports two thermal diodes. The temperature is taken from the diode located in the middle of Socket A that's why it is not quite adequate to the real CPU temperature. The chip also monitors 5 voltages and the rotation speeds of 2 coolers, though there are 3 cooler connectors altogether on the board. The South Bridge also implements AC'97 software sound controller.
As for the mainboard BIOS, EP-8KTA3 uses Award Modular 6.00PG with pretty standard settings.
EPoX EP-8KTA3 North Bridge is equipped with an active cooler. There is a regular heatsink and a fan like those, which you may see on the graphics cards. It's hard to say whether it tells on the mainboard stability that greatly, however, all mainboard manufacturers are little by little starting to equip their North Bridges with a fan, which is very likely to be quite justified.
We have to point out to you that throughout all the tests EPoX EP-8KTA3 proved very stable and reliable and didn't have any problems with the CPUs supporting 200MHz FSB as well as with those supporting faster 266MHz FSB. Although, there is nothing to be surprised at: EPoX mainboards have always been very stable. Besides, the reconstructed power supply circuit seems to have made its contribution to the mainboard's stability due to large high-quality capacitors.
EPoX has been long devoting a lot of time and effort to make its products attractive for overclockers. EPoX tried to make its EP-8KTA3 one of the most overclocking friendly mainboards. Well, let's see whether they succeeded or not.
The first advantage worth mentioning is the fact that all CPU configuration settings as well as Vcore settings are changed via BIOS Setup. Of course, this is very convenient. On the screenshot below you can clearly see what overclocking options our today's hero offers:
Like any other worthy mainboard, 8KTA3 allows changing the clock frequency multiplier by those AMD Athlon/Duron CPUs, where it is unlocked (see this article to find out how you could unlock your clock multiplier). Moreover, you can also change the FSB frequency.
We find it useful to say that it seems to be popular nowadays to make the mainboards support changing the FSB frequency with 1MHz increments. A lot of mainboard manufacturers are now introducing this opportunity in their products. However, unfortunately, EPoX doesn't belong to them. So, 8KTA3 allows setting only discrete values from the following list: 95, 100, 102, 104, 106, 107, 108, 109, 110, 111, 112, 113, 114, 116, 118, 120, 124, 127, 130, 133, 136, 140, 145, 150, 155, 160, 166MHz. Note that in this case the PCI frequency can be obtained from the FSB frequency if the latter is divided by 3 in the interval up to 120MHz and by 4 in the interval between 124 and 166MHz. Even though the range of supported FSB frequencies includes 27 values, around 133MHz and up there are not enough values for "fine" overclocking.
Speaking about the practical part of this issue, we should say that the mainboard loses its beautiful stability as soon as the FSB frequency exceeds 150MHz. That's why you will hardly ever use the "allowed" 155MHz+ FSB frequency on your EP-8KTA3 mainboard.
Moreover, EPoX made it possible to change the CPU Vcore, AGP voltage and Vio. Vcore can be changed with 1.025V increments increasing the voltage up to the maximum of 1.85V for Athlon and 1.75V for Duron CPUs. Note that increasing the Vcore by 0.1-0.15V may be not enough for extreme overclocking, however, unfortunately, EP-8KTA3 power circuit doesn't allow increasing it any higher without the necessary mainboard modification.
As for Vio and Vagp, these may be useful for the CPU overclocking with the FSB frequency increase. Therefore we are glad to announce that EPoX EP-8KTA3 enables you to increase Vio up to 3.7V, and Vagp - up to 2V.
So, EP-8KTA3 offers a pretty nice set of features for CPU overclocking by means of increasing the clock frequency multiplier and the FSB frequency. It is even easier to overclock processors playing with the FSB frequency on this mainboard, because it doesn't require unlocking the CPU clock multiplier. Since 8KTA3 is based on a new VIA KT133A chipset, nothing prevents you from increasing the FSB frequency that's why no matter which overclocking method you select, you will be able to achieve almost the same results in either case.
Testbed and Methods
As we got the mainboard based on VIA Apollo KT133A chipset at our disposal, we paid special attention to its performance. Formerly we could only assess the gain in performance that we acquired by combining a 266MHz processor bus and DDR SDRAM. But now EPoX EP-8KTA3 gives us a chance to clear out how much this gain is determined by each of the two factors - the wider memory bus and the faster CPU bus. We compared the performance of a system based on VIA Apollo KT133A core logic with a 266MHz processor bus with that of systems built on both available DDR chipsets working with PC2100 and PC1600 DDR SDRAM. Besides, we also added the performance rates of a platform based on VIA KT133 with a 200MHz processor bus to the list.
Moreover, when testing we noticed some difference between the results of EP-8KTA3 working with a 200MHz CPU bus and ABIT KT7, which is used in a KT133 based system. So, we decided to add the performance results obtained in a system based on VIA KT133A with a 200MHz CPU bus to our diagrams, so that to show this difference. In theory, VIA KT133 and VIA KT133A chipsets should demonstrate similar performance when working with a 200MHz processor bus. The inequality of their results takes place only due to the individual peculiarities of the mainboards tested.
All in all, we tested seven platforms with the following configuration:
|AMD-760 (266MHz FSB) |
PC2100 DDR SDRAM
MAGiK 1 (266MHz FSB)
PC2100 DDR SDRAM
|VIA KT133A (266MHz FSB) |
|AMD-760 (200MHz FSB) |
PC1600 DDR SDRAM
MAGiK 1 (200MHz FSB)
PC1600 DDR SDRAM
|VIA KT133A (200MHz FSB) |
|VIA KT133 (200MHz FSB)|
|CPU||AMD Athlon 1GHz (266MHz FSB)||AMD Athlon 1GHz (200MHz FSB)|
|Mainboard||Gigabyte GA-7DX||Iwill |
|EPoX EP-8KTA3||Gigabyte GA-7DX||Iwill |
|EPoX EP-8KTA3||ABIT KT7|
|Memory||256MB PC2100 DDR SDRAM||256MB PC133 SDRAM||256MB PC1600 DDR SDRAM||256MB PC133 SDRAM|
|Graphics Card||Creative 3D Blaster Annihilator 2 Ultra (NVIDIA GeForce2 Ultra)|
|HDD||IBM DTLA 307015|
As usual, we focused our efforts on performance of VIA KT133A based platform in real applications. First let us check the results of EPoX 8KTA3 with VIA KT133A core logic in office applications, and then we'll see how things stand with games. For a better comparison we will also look at the performance of a system based on AMD-760 (since it's the fastest DDR chipset) and that of a regular PC133 SDRAM system built on VIA Apollo KT133 chipset.
This benchmark shows some really curious results of the system's work in typical office applications. Business applications appear to be sensible to the memory latency, while they don't demand a lot from the memory bandwidth. As a result, the performance of KT133A in this case is even better than that of AMD-760.
What a striking change we can see! We have only replaced the first set of applications with content creation ones (like, for example, Dreamweaver and Photoshop), and we get an absolutely different situation. Since content creation applications work with a large data packs, the DDR systems look more impressive in this benchmark. The system built on KT133A fails to surpass even AMD-760 with PC1600 DDR SDRAM. However, if we compare the achievements of KT133 and KT133A chipsets, we will see that a faster processor bus alone yields a 7% performance gain, while the system with AMD-760 and PC2100 DDR SDRAM gives a gain of 11%. Thus, there is no great difference between the performance of KT133A and AMD-760.
Another test that demonstrates the average performance in 2D applications reveals the difference of only 2% between KT133A and AMD-760. At the same time KT133A is 6% faster that the common KT133. This way, the conclusion is that the performance in office applications depends much more on the frequency of the CPU bus, rather than on the memory bus bandwidth. Let's have a closer look at the results of SYSmark2000:
One more time we can observe KT133A boasting higher results than ALi MAGiK 1 systems with both PC1600 DDR SDRAM and PC2100 DDR SDRAM.
Again VIA Apollo KT133A based platform performs pretty well. AMD-760 is only 1.5% ahead.
This time we paid special attention to the professional 3D-modelling software called 3D Studio MAX R3. To estimate the performance of our review participants we checked how long it took them to render Anisotropic Wheel scene at the resolution set to 800x600. So, the smaller value denotes faster performance. Surprisingly enough, almost all the platforms demonstrated equal abilities, since the major workload during the scene rendering in 3D Studio MAX R3 was laid upon the processor FPU and not upon the system memory bus. As for ALi MAGiK 1 with PC1600 DDR SDRAM, its worst result made no surprise.
A single glance at the diagram is enough to make sure that performance is influenced by both system bus frequency and memory bandwidth.
We have also checked how fast the tested systems are at encoding mp3-files in a popular AudioCatalyst application. For this purpose we used a specially created 100MB wav-file. It turned out that encoding is one of the few operations, where DDR SDRAM brings higher performance gain than a 266MHz bus. In this test AMD-760, as well as ALi MAGiK 1, is 9% faster than KT133A, and there is hardly any difference between KT133 and KT133A chipsets at all.
Now let's check how the tested systems feel in gaming applications, but first we shall turn to the results of the synthetic 3DMark2000:
The situation is the same again: systems with 266MHz FSB are far ahead of their 200MHz rivals. VIA KT133A in 266MHz systems is 10% faster than the regular KT133, and AMD-760 with PC2100 DDR SDRAM adds another 4% increasing the gap.
As the resolution grows, the performance of the graphics subsystem becomes still more important. So, the type of the memory doesn't tell that tangibly on the total performance of the system. In the long run, all the systems come to similar results here.
When testing systems built on chipsets supporting DDR SDRAM, we mentioned Quake3 as one of the programs, where these chipsets provide the greatest performance gain. Now we can see that VIA KT133A with PC133 SDRAM provides a 12% increase in performance as opposed to the regular KT133. In the meanwhile, AMD-760 with PC2100 DDR SDRAM proves 19% faster than KT133, i.e. we get clear evidence that memory bandwidth matters a lot in Quake3.
Higher resolution leads to a less notable difference in the results. However, the trend is still the same as in the previous case.
Quake3: Team Arena works with more complicated textures, so there is a bit slighter divergence of the results than in case of the common Quake3. The platform built on KT133A is 12% ahead of KT133, but it is 5% slower than the AMD-760 based system.
At the resolution of 1024x768x32 the results are absolutely the same as in the regular Quake3.
In Unreal Tournament the performance of KT133A is really outstanding: the platform based on this chipset is even faster than that based on ALi MAGiK 1 with PC2100 DDR SDRAM. However, if we look at the performance of AMD-760, using DDR SDRAM with a 266MHz CPU bus results in a 5% gain unlike just a shift to a faster bus.
The results stay almost the same when the resolution grows.
MDK2 requires not so much from the memory bandwidth as Quake3 and Unreal Tournament. Still, the difference between the performance of AMD-760 and KT133A counts to 3%. However, it looks just negligible as opposed to the impressive 11% of KT133A's dominance over the regular KT133.
The results here repeat what we saw in Quake3 at higher resolutions.
Our last gaming test is a demo version of Mercedes-Benz Truck Racing. This truck racing simulator is a vivid representative of the latest generation games that use a lot of textures and a complex geometric and physical model. The diagram created according to the systems' results in Mercedes-Benz Truck Racing shows pretty clearly that all the platforms with 266MHz CPU bus turn out much faster than those with 200MHz one. Surely, it is incorrect to say that the speed of the processor bus is always the basic determinant for the Athlon systems performance. For instance, AMD-760 with PC1600 DDR SDRAM is one of the leaders, too. However, the platform based on VIA Apollo KT133A with faster processor bus and a memory with lower latency demonstrates high results throughout all the tests. It is hard to say, how much its success is determined by the latency of PC133 SDRAM and by the 266MHz bus, since the system starts working asynchronously if switched over to 200MHz. Naturally, in this case the performance is lower.
As our tests showed, modern applications are still more sensible to the memory latency than to its bandwidth. The tests run in real applications reveal that the performance of KT133A, which allows using processors with a 266MHz bus, is closer to the performance of the systems built on AMD-760 chipsets with DDR memory, than to that of KT133 based systems with PC133 SDRAM.
That's why systems built on the discussed VIA Apollo KT133A may become a brilliant alternative to DDR SDRAM systems. For the time being DDR memory is quite expensive and it is scarcely available, and the difference in the performance of KT133A and AMD-760 is no higher than 5%. So, judging by the price-to-performance ratio, KT133A with PC133 SDRAM looks much more attractive.
As for ALi MAGiK 1, the tests condemn this chipset revision as a complete failure: the system based on ALi MAGiK 1 with PC2100 DDR SDRAM nearly always lagged behind that built on VIA KT133A with PC133 SDRAM.
While Athlon processors officially supporting 266MHz bus are still unavailable, the mainboards based on KT133A can be successfully used without them. As the practice shows, all the CPUs with a 200MHz bus can work well with a faster bus. Although you may need to lower the multiplier in this case. Nonetheless, the results confirm that at the same resulting frequency the CPUs with a faster bus show considerably higher performance. For example, for 1GHz Athlon used in a testsystem the difference climbed up to 7-10%.
Speaking about EPoX EP-8KTA3 mainboard, which was used in our a system built on VIA KT133A chipset, it has brought delightful results. This board can boast marvelous stability and excellent performance. All this ensures high performance even with a 200MHz bus. Of course, this mainboard is no absolute perfection, but for many of us it can be the right thing to choose. EP-8KTA3 is sure to win the hearts of overclockers, who may find 8KTA3 good enough to satisfy all their needs.