AMD’s Athlon has been enjoying its new Thunderbird core for a little over a month now and it has been blessed with the company of its younger brother, the Duron, for a couple weeks as well.  In spite of the length of time that we’ve known about both of these CPUs, it seems like we’re discovering new things about AMD’s flagship and their value processors almost daily.

The most recent discovery happened to be made by the very source that originally cracked the code on how to overclock the original Slot-A Athlons.  Tom’s Hardware Guide, one of the founders of the online hardware enthusiast community that we have today, published an extensive article on how the ASUS A7V allowed for the multiplier on both the AMD Athlon (Thunderbird) and the Duron to be adjusted. 

Even more recently, the Guide published information from AMD relating to the fact that only “engineering samples” of Thunderbird and Duron processors would allow their clock multipliers to be altered by the A7V and final shipping parts would not be as overclocking friendly. 

While we have yet to prove that the limited number of Socket-A Thunderbirds and Durons currently on the market do in fact fall in line with AMD’s claim that they are completely multiplier locked, we did manage to discover that ASUS’ A7V wasn’t the only motherboard capable of manipulating the clock multiplier on the engineering sample Thunderbird and Duron CPUs we have been using in the lab for the past few weeks. 

The first Socket-A motherboard we ever received was VIA’s KT133 reference board, which we got our hands on weeks before we even had a Thunderbird to play around with.  The reference KT133 board interested us because it did feature clock multiplier adjustments on the board itself, but upon testing the board with our Thunderbird CPUs we quickly realized that the multiplier settings had no effect on our seemingly clock locked CPUs. 

Shortly thereafter, we received a handful of other Socket-A motherboards, the Gigabyte GA-7ZM, the FIC AZ-11 and its OEM counterpart, a motherboard from a Compaq Thunderbird/Duron system.  Little did we know that the AZ-11 we received back in May featured the same settings that ASUS’ A7V was discovered to have. 

The First AZ-11

We actually have two AZ-11 boards, the first was a pre-release engineering sample we requested ahead of time in order to include it in our original Socket-A Motherboard Preview.  On this board, as you can see from the picture below, there was an unusual set of jumpers that were physically hacked off of the motherboard after production.

Looking at the bottom of the board, we also noticed that the jumpers were obviously placed on the board after the fact since the leads of this particular jumper block stuck out much further than the rest of the leads on the back of the board. 

Click to Enlarge

At the time, we thought nothing of it, since AMD had already told us that the CPUs were completely multiplier locked, although we did think it was odd that FIC would go to the trouble to saw off that block of jumpers before sending us the motherboard. 

After reading about Tom’s Hardware Guide’s overclocking success with the ASUS A7V (which we will be reviewing shortly), we decided that maybe it was worth looking into those mysterious jumpers on the AZ-11. 

We noticed that the four pins on the AZ-11 could potentially correspond to the four Frequency ID pins Tom described in his article.  By assuming that the position 1-2 on our 4 x 3 jumper block corresponded to the Athlon/Duron’s FID setting of 1 and assuming that the remaining 2-3 setting on our jumper block corresponded to an FID setting of 0, we went to work.

The AZ-11’s jumpers were labeled, from top to bottom: OVERRIDE2, OVERRIDE1, OVERRIDE 4, OVERRIDE 3.  These four jumpers, each with three pins, conveniently correspond to the four FID pins, making OVERRIDE1 the jumper that controls FID0, OVERRIDE2 the jumper that controls FID1, OVERRIDE3 the jumper that controls FID2 and OVERRIDE4 the jumper that controls FID3.

Since the top of the pins were cut off we took advantage of the fact that the leads were so long and simply placed jumpers on the bottom of the board instead of on the top where they would normally go. 

Click to Enlarge

Lo and behold, the AZ-11 was able to adjust the clock multiplier of all of our CPUs using these jumpers.  Unfortunately, this pre-release AZ-11 board wasn’t as stable as we would liked it to be in order for us to run a full suite of benchmarks at all available clock speeds…

Shipping Revision of FIC’s AZ-11 to the Rescue

Not too long after our trip to Computex in Taipei, Taiwan, FIC managed to send us a final shipping revision of the AZ-11, which was considerably more stable than the pre-release engineering sample we originally previewed. 

The first thing that we noticed about the final board was that the set of OVERRIDE jumpers that we were blessed with on the prerelease board were no longer present.  Fortunately, the contacts were still on the board which gave us the hope that with a bit of soldering we might be able to get that functionality back on this shipping board.

Using our trusty soldering iron we soldered a total of 12 pins we removed from an older motherboard no longer in use into the 12 contact points on the AZ-11 board where the pins would normally be. 

The newly soldered pins

After making sure that the pins would not short each other, we fired up our AZ-11 test system, and as we had hoped, the clock multiplier functionality was still there. 

Limitation Number Two: No Core Voltage Adjustment

With our modified AZ-11 paving the way for a fairly thorough investigation of the various clock speeds of Thunderbirds and Durons, we were ready to start. 

Overclocking our Thunderbird samples shouldn't be too difficult since these CPUs are already running at a decently high core voltage; unfortunately, the same doesn’t hold true for the Duron.  On most motherboards this wouldn’t be an issue, since most motherboards allow you to manually adjust the core voltage of your CPU, but on the FIC AZ-11, including the final shipping revision of the motherboard, you’re not offered this feature.

There is a jumper on the AZ-11 that is labeled "MAGIC TUNNER" which is supposed to allow you to manually adjust the core voltage of the CPU.  Unfortunately, setting this jumper to the enabled position had no noticeable effect and did not allow us to increase the core voltage of our processors any.  Hopefully, FIC will correct this in future revisions of the motherboard or by a simple BIOS update; the latter seems like a very likely option. 

But for us, without any core voltage adjustment options, overclocking the Duron was not possible.

Using the Golden Bridges

Tom’s latest article indicates how to physically modify the Thunderbird/Duron in order to adjust their clock multiplier and/or core voltage setting using the “Golden Bridges” that are present on the CPUs themselves.

Thunderbird core voltage settings

Using this newly found information we modified our Duron CPU to operate at the maximum allowed core voltage of the current revision of the Duron processor, 1.85v.

The modification was done by using a conductive pen to connect the two points of the “Golden Bridges” present at position L7 on the Duron CPU. Since the tip of our conductive pen was entirely too big to do a precision job, we simply used the tip of a pin to connect the two points on the "Golden Bridges."

We started by first increasing the core voltage to 1.70v, which only required one modification, then later moved up to 1.80v, which required a second modification and finally 1.85v, which forced us to connect all of the “Golden Bridges” present at L7.

In spite of all of this, our Duron 700, running at 1.70v and finally at 1.85v would not overclock past 700MHz. We could easily underclock it to 600 and 650MHz but even running it at 750MHz was out of the question. This anomaly was present even with our 800MHz and 1GHz Thunderbird parts, the 800 wouldn't boot at any clock speeds faster than 800MHz and the 1GHz part that we ran at 105MHz x 10.0 (1050MHz) wouldn't boot at 100MHz x 10.5 (1050MHz).

The only explanation we could come up with for this was that somehow the clock multiplier was being limited by the original clock multiplier of the CPU. For example, the Duron 700's "Golden Bridges" indicated a 7.0x clock multiplier and while it would work with 6.0x and 6.5x multipliers anything higher than 7.0x wouldn't work. The same situation existed with the two Thunderbirds we tested.

Since we have such a small sample size we can't say anything definite about the status of clock multipliers on all CPUs, we purchased a few retail chips and will test those as soon they come in and report on our findings.

While we are aware that manipulating the "Golden Bridges" that correspond to the FID and BFID pins on the CPU we have yet to successfully find a tool that can be used to cut a connected "Golden Bridge." We will continue to experiment with this as well.

The Purpose & the Test

What was the point of all of this?  Since it may not be possible, especially with future revisions of the Thunderbird & Duron, to overclock your CPUs as we have been able to with our engineering samples, why on Earth would we spend the time doing this?

In our original Thunderbird and Duron reviews we were limited to only a handful of clock speeds we could provide you with for comparison purposes, simply because we only had a few CPUs. 

Now, with the ability to clock the CPUs at any speed, we can provide you with a more thorough illustration of the performance of the Thunderbird as well as the Duron at all available clock speeds.

So if you're debating whether to go with a lower clocked Thunderbird or a similarly priced yet higher clocked K75 based Athlon, these benchmarks should help you out since we're comparing every commercially available Thunderbird & Duron.

Note: The BX platform running at 133MHz FSB and the Celeron running at 100MHz FSB are both overclocked platforms and are not sold in that form on the retail market. Those performance numbers are only included for purposes of remaining thorough with our comparison.

Under Content Creation Winstone 2000 you can see that the Thunderbird, clock for clock, is faster than the older Athlon which we already knew. So if you're choosing between getting a lower clocked Thunderbird or a faster K75 Athlon the above chart should help you out a bit.

You can also see that the Duron at 600MHz is coming in around the same speed as the Athlon 600 and the Pentium III 600 on a BX platform. If you're not sure about the overclocking potential of the Duron (or whether or not you're even going to be able to overclock it) and you currently have something like a Pentium III 600E, you may want to stick with your current setup.

Under SYSMark 2000 the Thunderbird's advantage over the regular Athlon is much more noticeable than under Content Creation Winstone 2000 because of the benchmark's L2 cache dependency. On the same note, the Duron falls somewhat short of an equivalently clocked Athlon in a couple of the tests.

Quake III Arena falls in line with what SYSMark 2000 told us from the previous page, a Thunderbird running at 950MHz should give you performance just slightly greater than an older 1GHz Athlon. However here it takes a Duron 650 to come ahead of an Athlon 600.

At 1024 x 768 the benchmark is memory bandwidth limited by our GeForce2 GTS card which is why the benchmarks come out so close to one another.

The UnrealTournament benchmark reflects what we've already seen in Quake III Arena.

Expendable is much more memory bandwidth dependent than the previous benchmarks while at the same time it does appreciate a fast L2 cache which gives the Thunderbird its advantage here.

Content Creation 2000 under Windows 2000 illustrates the performance benefits running Windows 2000 offers over running Windows 98SE. Clock for clock there is about a 10 - 15% increase in performance over Windows 98SE for the exact same system.

High End Winstone 99 prefers the Athlon and especially the Thunderbird because of its powerful FPU as well as its fast exclusive L2 cache.

Quake III Arena under Windows 2000 once again illustrates the performance improvement Windows 2000 offers over Windows 98SE. The improvement isn't as great in Quake III Arena as it is under Content Creation Winstone 2000 but it is a tangible improvement none the less.

The Thunderbird & The Duron do very well under SPECviewperf, the only thing holding them back here is the KT133 platform. Using either of those CPUs on the forthcoming AMD760 could prove to be even faster.

As we've proven time and time again, the distributed computing performance of CPUs running the RC5 client isn't dependent on memory bandwidth or L2 cache speed, rather just the raw power of the CPU itself.

Final Words

Now that you finally have a better picture of the performance of the Thunderbird and the Duron you can hopefully make your buying decisions with greater ease. We will keep you all updated as to how our attempts at further overclocking our Thunderbird & Duron processors goes as well as the retail parts we have on the way.