by Ilya Gavrichenkov
10/23/2000 | 12:00 AM
The AMD Athlon CPU launched in summer of 1999 has changed the situation in the processor market just drastically. AMD, which had been always catching up with Intel, finally managed to run neck and neck with the microprocessor giant and now has become Intel's fully-fledged competitor. All this became possible due to a new K7 processor core with a totally new architecture. At the same time, the problems and failures in the CPU market, which have been streaming down on Intel all the year round resulting into grave CPU shortages, helped AMD to strengthen its position in the processor market.
In the beginning of this summer AMD undertook another move forward to further conquering the processor market and introduced another CPU - Duron, aimed at the Low-End sector. Duron was initially intended to perform as a competitor to Intel Celeron processor and as we can see it succeeds. Really, the cost of AMD Duron CPU is very close that of Intel Celeron while the performance is much higher. Why? Well, first, it features a faster 200MHz EV6 bus, and second, it can boast a larger effective cache size. However, Duron also has its weak points. Here we should mention the lack in low-cost mainboards supporting this CPU as well as pretty complicated overclocking methods. From this viewpoint Celeron looks much more advantageous: there is the whole bunch of cheap mainboards on i810 chipset for Celeron, and overclocking by means of increasing the FSB frequency is as easy as ABC. Moreover, Intel Celeron is quite overclocking friendly and allows getting a 1.5 times higher working frequency in a wink of an eye.
However, now the things are getting better for Duron. A few manufacturers have already introduced their low-cost products for this CPU thanks to SiS, which announced a cheap integrated chipset for Socket A mainboards, SiS 730S. As for overclocking, this is exactly what we are going to discuss in this article.
As you remember, it is pretty hard to overclock Duron CPU, as well as Athlon, by means of increasing the FSB frequency. The reason is the EV6 bus, which works in DDR mode transferring the data along both signal edges. That is why if the FSB frequency is 100MHz the data is transferred along the bus at 200MHz overall frequency and any slightest increase of the nominal FSB frequency immediately results into system instability. Therefore, the only acceptable way to overclock your AMD Duron processor is to change the CPU clock frequency multiplier. This clock multiplier is fortunately locked not in the processor core, like in case of Pentium III or Celeron, so, you may unlock it if you make a certain effort. For instance, if you take a slot Athlon, you will be able to change the clock multiplier by resoldering a couple of elements on the processor board.
And what about socketed processors? Let's find out…
Well, all Socket A CPUs, Duron and Thunderbird, feature two signal sets dealing with the clock multipliers. The first one, called FID, is a four-bit signal, which tells the multipliers to the chipset. The second one, called BP_FID, is also a four-bit signal, which tells the multiplier to the processor core. We don't know why AMD couldn't be happy with one signal set combining the functions of both, but FID has nothing to do with the processor core and isn't connected to it in any way.
As a result, when the system is booting, the mainboard gets the CPU clock multiplier from FID signal while the processor core gets it from BP_FID. In addition, if the values received by these two system components do not match then everything crashes because of no synchronization.
Besides, AMD considered the implementation of different FID and BP_FID directly inside the processor core to be too expensive that's why they are partially generated on the processor surface. This is exactly the reason why we see some resistors and golden bridges on top of the processor.
In particular, L3 and L4 bridges are responsible for BP_FID signal and L6 - for FID. During the manufacturing process certain bridges are broken, which sets the CPU clock frequency multiplier. Without going too deep into details, we would like to point out that each of the four L6 bridges stands for the corresponding bit status in FID signal. Namely, the basic 3x multiplier receives additional 0.5x, 1x, 2x and 4x (in case of working frequencies over 1.1GHz, FID signal is generated in a bit different way). As for the eight L3 and L4 bridges, all of them are responsible for BP_FID, which is clocked in a similar manner. However, in this case the nearest pair of bridges stands for each bit.
After we have got some idea of how the Socket A CPU clock frequency multiplier is set, the only way to overclock our CPU seems to be the change of the tiny golden bridges configuration on the processor surface. But put your file away and don't hasten to ruin your poor CPU: AMD left a loophole in the multiplier locking, so let's be civilized and find out the "peaceful" overclocking method.
When we spoke about a loophole we meant that the bridges setting BP_FID are located not only on the processor surface and inside the core but also down on the Socket A. So, the mainboard turns out capable of influencing the BP_FID! Besides, the mainboard can also change the FID signal since it goes from the CPU to the chipset North Bridge directly via the mainboard.
Some mainboards do allow changing the CPU clock frequency multiplier. However, unfortunately, far not all the boards can boast this cool feature, mainly because it is quite hard to implement. That's really complicated because the mainboard has to modify not only BP_FID signal sent to the CPU but also FID signal sent to the chipset North Bridge. Note that both signals are electrically different. So, now we know only about three mainboards available in retail, which allow modifying the CPU clock multiplier. They are:
Besides, some mainboard manufacturers promised to introduce their new mainboards based on VIA Apollo KT133 and featuring this peculiarity. Among the first ones should be MSI and FIC.
So, now we will consider only those mainboards, which allow changing the CPU clock multiplier, because they simplify the whole overclocking procedure just impressively.
However, this is only the beginning of our story. It turned out that sometimes you cannot achieve the desired aim by simply installing the CPU into Socket A and setting the required multiplier in the BIOS Setup or with the jumpers. Unfortunately, only the first shipments of Socket A Athlon and Duron processors allowed this easy overclocking. AMD has also found the way to eliminate the "loophole" in the multiplier locking and resorted to it as soon as "good" mainboards appeared. The thing is that there remained one more set of bridges, L1, on the way from BP_FID pins on the Socket A to the processor core. These bridges are situated on the CPU surface. The scheme below shows the way all this stuff works:
So, AMD started breaking the L1 bridges, which brought the overclockers' efforts to naught. In other words, overclocking Athlon and Duron CPUs by means of changing the clock multiplier on the board turned out vain.
|L1 Golden Bridges||L1 Golden Bridges|
But do not give way to despair! Who may prevent you from restoring the broken L1 bridges? If you succeed, the mainboards, which allow changing the CPU clock multiplier, will get back their lost possibility. Moreover, there is no need to care about other bridges as well as about restoring the broken contacts in case of failure.
So, here is a summary of the entire above mentioned stuff. If we take a mainboard, which allows changing the CPU clock frequency multiplier and close four L1 bridges on the CPU surface, then we will get a very good overclockable system. Now let's discuss how you could close the golden bridges.
As we have already found out, it is necessary to electrically close the L1 golden bridges in pairs in order to unlock the CPU clock frequency multiplier. There are three ways to do it.
The first and the easiest way is to cover the broken bridges with a thin layer of lead. Just take a pencil with a 0.5mm graphite lead or sharpen an ordinary pencil. If you take a closer look at the bridges (with the help of a microscope), you will notice that the bridges were burned through. This is most likely to have been done with a laser, because the ceramic packaging of the CPU got darker in places. Instead of the bridge you can now see a small groove. So, now you have to fill this tiny groove between the bridge edges with lead. For our experiment we took a clutched pencil and covered the bridges carefully, blowing away the lead bits from time to time. Here is what we managed to achieve:
The second way is to close the bridges with some conductive adhesive. Though "conductive adhesive" may sound a bit scary, it's not a rare thing. Just go to the nearest electronic shop and get a small tube of this adhesive. Below you can see a photo of the adhesive we got in our local (Russian) "CHIP & DIP" shop for only $3.90.
Though the tube is really tiny, it was just enough for my needs. I took a piece of a thin wire and dropped a bit of adhesive onto the burnt bridges. Note that you should degrease the CPU surface beforehand for better results.
When working with the adhesive you should be very quick, because it is based on some dissolvent and hence dries rapidly. After you finish, make sure that new adhesive bridges do not meet. We used a thin scalpel to be sure that everything is alright with them.
The third method is a bit more complicated. You can solder up the bridges with a thin soldering iron. But if your hands are trembling with impatience then you'd better not do it. If you worry too much about your precious CPU, then ask some technical guys for help to avoid risks. Here is what it should look like afterwards:
This method provides the most reliable closing of all. However, your CPU will never regain its initial looks. That's why if you decide on this method then say good-bye to your warranty.
So, tell us: is it really so hard to unlock the CPU clock frequency multiplier by AMD Athlon and Duron processors? We believe, it's as easy as ABC. All you need is just your desire and a couple of simple tools. Here they are:
What will this give us? How much will the performance grow? Stay tuned and you will see yourself!
The tests for AMD CPUs were run in the system configured as follows:
The testbed for Intel processors was configured as follows:
To make the Duron CPU work normally on ABIT KT7 mainboard we had to reflash the BIOS with a new UL version.
As for overclocking, we tested AMD Duron 600MHz, which overclocked up to 1GHz, AMD Athlon 700MHz, which overclocked up to 900MHz, and AMD Athlon 1000MHz, which overclocked up to 1.2GHz. At the working frequency higher than that, the system appeared very unstable, which was very likely to be caused by the CPU overheating. ABIT KT7 is designed in such a way that you can't install Thermaltake Chrome Orb cooler for Socket462. As for the Thermaltake Golden Orb for Socket 370, we didn't dare use it because we were afraid to damage the CPU surface.
Well, we are not magicians and we cannot work wonders that's why of course, we increased the core voltage when overclocking. In other words, we could overclock our Duron processors up to 800MHz at the nominal 1.6V and then we had to increase the voltage by 0.075V for every further 100MHz. So, when our Duron worked at 1GHz the Vcore was equal to 1.75V. To tell the truth, the CPU got quite warm and didn't want to work properly without a normal cooler. We managed to save the situation with the help of a large cooler taken from the boxed Intel Pentium III 866.
Note that the performance benchmarks were run on the CPUs with not very common frequencies, which we obtained by means of overclocking or downclocking of the processors mentioned in the testbed configurations. Namely, we had an AMD Athlon 600MHz and 650MHz for Socket A (though you can come across Athlon 650MHz in some shops, the Athlon family starts from 700MHz officially) and Duron CPUs working at 800-1000MHz. This allowed us to get a clearer picture of the dependence between the CPU performance and working frequency.
And now let's take a general look at our benchmarks:
As you can see, Duron working at 900-1000MHz performs almost level and the growing core frequency doesn't seem to make the CPU any faster. So, we can conclude that the CPU architecture (i.e. high clock frequency multiplier and low (only 100MHz) FSB frequency as well as small cache) aren't that efficient at the working frequencies close to 1GHz. As for Athlon performance in this test, the higher gets the clock frequency, the faster appears the CPU. That's why it will really make sense to increase its working frequency over 1.2GHz.
And this diagram shows the dependence of the CPU performance on the core frequency for AMD Duron processor in Quake3 test:
|Quake3 Arena, fps||89.5||92.6||96||98.1||100.1||102.9||105.5||107.1||107.7|
|Performance per 1MHz, fps||~0.15||~0.14||~0.14||~0.13||~0.13||~0.12||~0.12||~0.11||~0.11|
You can see clearly that the higher gets the CPU frequency, the smaller appears the relative performance gain. In other words, every further megahertz adds less to the overall percentage. So, it will hardly make any sense for AMD to launch over 1GHz Duron CPUs as they are, without any architectural modifications. And another architecture means another CPU, that's why it is no longer the good old Duron…
Well, the practical conclusion we can driver from here sounds as follows: don't try overclocking AMD Duron over 1GHz. It won't have any tangible effect but the risk of burning the CPU will get too huge, because of the system overheating.
And now let's compare the performance of Duron and Athlon processors at the given frequencies, so that we could see how greatly the first one falls behind its faster brother (in percents):
|3D Studio MAX 3.0||9||9||12||11||11||12||13||13||14|
It seems surprising but Duron stably falls 12-13% behind Athlon at any frequency. The similarity of the CPU architectures explains the stability we observe and the lag itself results from the different L2 caches.
Now we are going to dwell on every benchmark we carried out.
This is a synthetic benchmark aimed at finding out the strong points and bottlenecks of the CPUs. If you are interested, you may download it from here. It runs, all in all, about a minute and a half and provides very stable results. The higher is the value, the better is the result.
As you can notice, the results shown by Athlon processor increase faster than those shown by Duron. No doubt, Athlon owes its success to a larger L2 cache. Besides, the results achieved by the CPUs differ by only 10% at lower frequencies, while at higher frequencies the difference makes over 17%. Well, not so good for Duron...
This test belongs to the 3DMark2000 benchmark set. It makes the CPU execute a number of instructions typical of the modern 3D games. Of course, the game itself is the best gaming test you can think of, however, running a synthetic test like that may be also of certain interest to us.
Here the performance grows together with the frequency and the performance difference between AMD Athlon and Duron CPUs makes around 10% all the time. There is a certain deviation at lower frequencies, however, it is negligible. The cache size doesn't matter that much at lower frequencies, because the CPU requires new data more rarely.
For the test file we selected Anisotropic Wheel. There were three frames rendered. The diagram below shows the time required for the rendering of one frame, i.e. the smaller is the value, the better.
This benchmark is a perfect way to prove that Athlon and Duron feature identical cores and to show that the performance depends only on the core clock frequency. The major workload falls upon the FPU and the Athlon L2 cache, which is four times larger than that of Duron, ensures a 10-11% performance gain over the latter at any given frequency.
We installed 1.16n patch and activated the following settings:
As usual, we used demo001 and every test was run three times. After that, we chose the maximum value out of the three.
Again, Athlon's advantage over Duron is not that impressive though it increases with the growth of the working frequency. And you will never call this test a synthetic one, will you?
Besides, we can't help pointing out that AMD processors overclocking doesn't have any serious influence on the performance in Quake3. We have already mentioned several times in our reviews that Quake3 is very sensitive to all the slightest changes that happen with the FSB. However, this referred to an Intel based platform, where all overclocking operations touched upon the system bus. Now we overclock AMD CPUs by increasing the core clock frequency multiplier and as you can notice, it doesn't have any tangible effect. Of course, there is some effect, but it is considerably lower than in case of Intel CPUs overclocked via FSB frequency increase.
|AMD Duron||89.5||107.7 (+20%)|
|AMD Athlon||95.2||117.9 (+24%)|
Note that the same performance gain (22%) is achieved when overclocking Pentium III 600 up to only 800MHz (with a 133MHz FSB).
We used an original game version without patches. The resolution was set to 640x480x16 and the minimal desired framerate was equal to 1. Again all the tests were run three times each and the best result was taken for the final diagram.
The situation has hardly changed. The performance difference between Duron and Athlon processors makes about 10% all the time. As for the deviations, we have to blame the unstable benchmark for that. Though we ran the test three times for each CPU at the given frequency, the results sometimes differed greater than we had expected.
This test was run with default settings three times for each CPU at the given frequency. We archived a 100MB array of 50 files (mostly Unreal Tournament textures). The results illustrate the time required for archiving, hence the smaller is the value, the better is the performance.
Nothing new as usual. Larger L2 cache by Athlon guarantees its victory over Duron, so that it requires 13-14 seconds less than its rival.
Here we used the same test array as in the previous case. The benchmark was run at "Normal" compression mode with a 256KB dictionary. Again, the lower result means the faster performance.
Well, what can we say here? This difference between AMD Athlon and Duron seems to be constant. No doubt that it is Duron's smaller L2 cache, which is to blame for the CPUs failure. The performance difference between the two processors doesn't depend on the CPU core clock frequency: it is only the work of chipset with the memory that matters here. Since we overclocked the CPUs by changing the CPU frequency multiplier, the core got faster and the bus between the CPU and the memory turned into the major bottleneck of the system.
Well, let's find out how could we benefit from Athlon and Duron overclocking. When we overclocked the CPUs from 600MHz up to 1GHz, the frequency got 67% higher. And what about the performance in different applications? Take a look here:
|Frequency gain, %||Performance gain, %|
|600->1000||Duron 600->1000||Athlon 600->1000|
|3D Studio MAX 3.0||51||62|
This table shows how the core clock frequency increase influences the performance of our testing participants in each particular application. Take for instance the calculating tests, such as CPU Mark 99 and 3D MAX, which do not interact with the memory that much, or the games, such as Quake3 and UT, which do not care at all about the core clock growth. As you can also notice, 3DMark2000 beautifully "emulates" the CPU workload in games, and WinRAR is more memory-dependent than WinZip.
What is the real use of Athlon and Duron overclocking by changing the CPU clock multiplier? Of course, it makes sense for those of you who deal with complex calculations: engineers, 3D-animators, etc. If you are dedicated gamer, overclocking your CPU will provide you the performance gain about 40% of the clock frequency increase. Anyway, this is quite a good outcome, especially bearing in mind that the owners of older Athlon based systems didn't have any possibility to overclock their CPUs some time ago.
But things are still not so wonderful, as they might seem...
Let's compare the CPUs from AMD with those from Intel working at the same clock frequencies:
The chart shows that Celeron 600 yields to Duron 600 quite significantly when working at the nominal FSB frequency (66MHz). However, as soon as it gets a 100MHz "doping" the gap between them turns much smaller. As for Athlon and Pentium III, they run neck and neck at all the given frequencies.
Here Celeron with the 66MHz FSB falls just dramatically behind its rival. It can somehow compete with AMD Duron only if overclocked up to 100MHz. However, the latter working at 850MHz again manages to make a significant dash forward and brings Celeron's efforts to naught. As for their elder brothers, Intel Pentium III with 100MHz FSB is just a bit ahead of AMD Athlon with the same bus. But with the FSB overclocked up to 133MHz Pentium III leaves its opponent far behind.
Here (remember that the smaller numbers denote higher performance) Intel Celeron is again in the very end and even working at 850MHz it can't beat Duron 600. Athlon defeats Pentium III at the given frequency. And as for Duron 850MHz, it simply flies!
Once again Intel Celeron 600 appeared lagging far behind: its 66MHz FSB seems to slow for this competition. At higher working frequencies, namely at 850MHz, Celeron has finally managed to get a bit ahead of its AMD competitor. As for Pentium III, it retains the lead over Athlon at 600MHz as well as at 800MHz (133MHz FSB). This test once again gives us to understand that FSB frequency is one of the main conditions for success.
As for Unreal Tournament, it longs for higher FSB frequency almost as much as for higher clock frequency. The overclocked Duron looks quite nice here. And so do Intel processors, except Celeron 600MHz.
Remember that the lowest values achieved in this test stand for the highest performance. Therefore, according to the diagram, Intel CPUs prove quite fast and attractive. In fact, we have always suspected WinZip to be somehow "optimized" for Intel processors...
Well, here Intel CPUs left their competitors far behind. We dare suppose that this test is even more optimized for Intel than the previous one. In fact, we think that the AMD CPUs architecture shouldn't be blamed for looking somewhat slower. It is mostly WinRAR organization that influences the results of this test. But on the other hand, it may also be the processor-to-cache bus of AMD CPUs, which is four times as narrow as that of Intel Pentium III or Celeron: only 64bit.
In conclusion, we would like to offer you the today's approximate prices on the CPUs tested:
|Intel Celeron 600||$80|
|Intel Pentium III 600E||$155|
|Intel Pentium III 800||$240|
|AMD Duron 600||$55|
|AMD Athlon 700||$120|
|AMD Athlon 800||$150|
As you see, AMD Duron is extremely cheap! In general, AMD processors are considerably cheaper than those from Intel. Especially bearing in mind that AMD CPUs get more and more popular. Low cost and popularity of AMD CPUs have also influenced the cost of the Socket A mainboards quite tangibly. So, it is hardly doubtful that attractive price and "interesting" peculiarities of AMD processors can make the company flourish even more with the time.
If we looked at our testing participants from the customer's viewpoint, we could state that:
All in all, AMD developed a cool CPU, which can not only compete with Intel Celeron on equal terms, but even defeat it.
Well, we have to admit that overclocking Slot A Athlon CPUs was a much harder task to complete some time ago, because it required special devices and more effort. And now, you can see from this article, that socketed AMD Duron and Athlon overclocking has become as easy as ABC and even a commencing overclocker can do it without any problems.
Moreover, as we showed Duron models working at lower frequencies allow overclocking by nearly 1.5 times, which is quite impressive. However, in fact, it is not the initial core clock frequency that influences the overclockability of the CPU. If you have followed our news carefully, you should have noticed that we found out the dependence of the CPU overclocking potential on the week, when it was manufactured. The second line of the CPU marking contains the number of the manufacturing week for this particular model. We have recently updated our Duron overclocking results and added a couple of new CPUs. Here is what we've got now:
|CPU Marking||Week of |
|Most Probable |
No doubt, that AMD CPUs have left Intel a bit behind in terms of CPU overclockability. However, we don't think this "overclocker's paradise" will last long. AMD, as any other CPU manufacturer, is not very enthusiastic about overclocking its processors that's why it may lose patience one day and undertake something to stop this. So, hurry up and enjoy, before we have to write another overclocking guide for you :-)