Introduction
I certainly agree that Tom’s Hardware Guide may not be the strong promoter of overclocking anymore that it once used to be. I decided to keep overclocking at a quieter and more reasonable level, simply because we, as the largest PC hardware publication on the Internet, are aware of the responsibility for our readers. Pushing people into doing something that could possibly damage their computer system or at least cost them important data is something we prefer to leave to others. However, running your processor beyond its official spec does not necessarily implicate any damage whatsoever, as long as you stay within reason. Overclocking can be a very cost effective way to improve the performance of your system, and getting involved with it will most likely improve your knowledge as well, because you might have to try out and learn several things about your system until you will finally succeed.
Tom’s Hardware has got a rather long tradition to inform its readers on how to modify their systems to run their processors beyond spec and thus faster. We were the first to report about the bus speed increase to 83 MHz in Pentium class systems in November 1996, we disclosed the ‘B21’-trick for Celeron processors in 1998 and last year we published all the details to produce the famous ‘gold finger’ overclocking devices for SlotA Athlon processors. Following this tradition we are proud to be the first who can supply you with information on how to overclock AMD’s brand new Athlon/Thunderbird and Duron processors for SocketA or Socket462.
Duron and Thunderbird
You are certainly aware of it. In the last three weeks AMD released the successor of the successful Athlon processor, still called Athlon, as well as the new low-cost Duron processor. The new Athlon is using the so-called ‘Thunderbird’ chip, which places the processor core as well as the second level cache onto one die, making the new Athlon faster as well as smaller. This ‘Thunderbird-Athlon’ comes with a new package, called ‘SocketA’, which plugs into a Socket with 462 pins vs. the ‘SlotA’ cartridge solution used by its predecessor. Duron is based onto the same architecture as ‘Thunderbird’, with the only difference that it comes with only a quarter of Thunderbird’s L2-cache. If you want to learn more about those two AMD-CPUs please read those two articles:
AMD Processors vs. Intel Processors
In the last 9 months we have learned that AMD has finally reached a position where it can supply processors that are fully competitive to the counterparts from Intel. When AMD’s Athlon processor was released in August 1999 it was even significantly ahead of any Intel product and it took AMD’s arch enemy no less than three months to come up with a solution that could properly compete with Athlon. Since AMD’s Thunderbird release on June 5, 2000 both processor makers are pretty much on par, so that it is up to the customer to decide whose solution he prefers. Basically, AMD’s new Thunderbird is superior to Intel’s ‘Coppermine’ Pentium III in floating point intensive applications, while Pentium III can score points with software that makes heavy usage of Intel’s proprietary ‘streaming SIMD extensions’, short ‘SSE’. Since last Monday (June 19, 2000) AMD is now also in the position to attack Intel’s low-cost processor ‘Celeron’. At the same price AMD’s new Duron processor is significantly faster than the Intel’s Celeron.
The hardware cracks out there are not quite satisfied yet though. Running AMD vs. Intel processors while keeping them within spec may be one thing, but would the new AMD-CPUs be able to live up to the performance achieved by overclocked Intel processors? So far Intel had the almost untouchable image as the producer of the most overclockable microprocessors worldwide. Especially the attractively priced Celeron used to be overclocker’s best friend. AMD processors were known to be much closer to the limits, making them rather touchy CPUs for the tweakers around the world.
Now, since AMD is finally also offering processors in a ‘one-chip’ solution, things might have changed. Could it be that the new AMD-processors can finally reach the good overclocking results achieved with Intel-CPUs? Would Thunderbird and Duron be able to possibly even surpass Pentium III and Celeron?
Overclocking Duron and Thunderbird – The Theory
Quite a few weeks ago we were in the lucky situation to receive the full data sheets of ‘Thunderbird’ and ‘Spitfire’, today known as ‘Duron’. In those data sheets we found the following promising data:
Well, it’s not too hard to see that this is the list of all ‘multiplier’ or ‘clock divider’ settings for Thunderbird and Duron. The pins FID[0-3] are obviously responsible for the multiplier used by the processor. Thus finding the pins would mean finding the way to overclock those CPUs.
Now obviously it’s not too hard to do so, since the data sheet naturally contains a description of each CPU-pin. This is where they’re at:
Unfortunately locating the pins alone doesn’t quite settle the story. You need to find out what to do with them. The following information helps us with that.
Overclocking Duron and Thunderbird – The Theory, Continued
Seemingly the processor tells the system (in this case the north bridge), which clock multiplier it would ‘like’ by driving the FID-pins. Then the system initializes the processor, using the ‘Serial Initialization Packet Protocol’ = SIP Protocol. This shows that the FID-pins are no input, but output pins, which differs significantly from the BF-pins used by Intel processors in the past, which were input-pins. The AMD-processors get initialized by the SIP-protocol, and whatever this protocol tells the CPU is what it is going to do. Therefore the multiplier for an AMD SocketA processor can be chosen either by modifying the data send by the SIP-protocol, which could be done by the BIOS, or by supplying the north bridge with your own FID-settings, which requires a ‘cut’ between the north bridge and the actual FID-pins of the CPU.
We are already working on an easy solution to do the latter. However, none of you might ever need to do that, because there is a much easier solution. Read on to find out why.
The Voltage Selection for Thunderbird and Duron
All die-hard overclockers know it. The key to a successful overclocking is choosing the right core voltage. This is valid for Thunderbird and Duron just as much. While Thunderbird has a default voltage of 1.7 V, Duron is using 1.5 V. If you want to overclock the two processors successfully, you need to raise the voltage.
I don’t really think that there is much sense in providing the information on which CPU-pins are responsible for the core voltage, because most motherboards will most likely give you an option to change the voltage either in the BIOS or via jumpers. However, you should be aware of one important limitation.
As you can see, the maximal voltage that can be chosen is 1.85 V. Please realize that when you are trying your luck with Thunderbird or Duron. Thunderbird can only be supplied with a core voltage of 0.15 V above its spec (which is less than a 10% raise), unless you want to change the whole motherboard power supply circuits. Duron is better off. You can add some 0.35 V to the 1.5 V default voltage, which should be good enough for most reasonable clock speeds.
Overclocking SocketA Processors – It’s Easier Than You Think
You might remember that QDI has already announced that their SocketA-motherboard will support overclocking on the day of the Thunderbird launch. However, so far nobody was able to test their board yet. We were particularly lucky that Asus supplied us with a A7V-motherboard at Computex, which is Asus SocketA-solution . This board was already used with Thunderbird as well as Duron and turned out to be the fastest and most reliable SocketA-platform that has been tested by Tom’s Hardware so far.
Asus did not tell us much when we received this motherboard and we certainly didn’t receive a manual for this pre-release product either. This is why we were more than delighted when we found out that the A7V allows the change of the processor voltage as well as the adjustment of the processor multiplier! The first is done with jumpers, the latter can be achieved with dipswitches. Believe it or not! Without knowing we received the first Thunderbird/Duron overclocking platform!
The fact that Asus was right away able to include overclocking features on their early SocketA-solution proves the point that other motherboard makers will eventually also be able to equip their boards with these nifty settings. Thus I suppose that you don’t really need to fiddle around with the FID-pins. As soon as Asus releases the A7V you have all you need to overclock your Thunderbird or Duron processor.
Overclocking Thunderbird and Duron
As soon as I had found out about the overclocking features of the Asus A7V I started to run Duron at higher speeds. I was very impressed with the result! I’ve got four Duron processors, one 650 MHz type and three Duron 700 types. After raising the voltage to some hefty 1.85 V all four Durons reached 950 MHz without a glitch!!!
The results with Thunderbird weren’t too shabby either. My 1 Ghz TBird reached 1.1 GHz, also at 1.85 V. More than that was unfortunately not possible, but the 1.1 GHz were rock stable. I ran all my benchmarks with it, including the usual 3-times Sysmark2000 run which takes no less than an hour.
We are also in the possession of a pre-release Athlon/Thunderbird 800. This CPU would only reach 900 MHz reliably, but that could be due to its pre-release status. I personally believe that most Duron or Thunderbird processors will reach a stable 900 to 950 MHz in clock speed.
Testing Overclocked Thunderbird and Duron Processors
We wouldn’t be Tom’s Hardware if we would not supply you with a good amount of performance data that shows the benefit of an overclocked SocketA-CPU. To waste as little time as possible I decided to run four benchmarks at each clock speed of Duron and Thunderbird. The benchmark that took longest, and which is responsible for the late release of this article is of course BAPCo’s Sysmark2000, which takes some hefty 1 – 1.5 hours to run. Additionally I ran Quake 3 Arena’s Demo001 at the ‘NORMAL’-setting, 3D Studio Max 2 and finally I made a new addition to our benchmark suite. I wanted to do this for ages, but something always kept me from succeeding with it. Finally I can now offer my readers a Linux kernel compilation benchmark, which shows the integer performance of a CPU very well.
To give you a good comparison with Intel processors I ran all the benchmarks on Intel’s Pentium III 1 GHz and an overclocked Celeron 1 GHz (original Celeron 667) as well. The platform used for Pentium III was Asus’ new CUSL2 i815E/Solano2 motherboard and Celeron was plugged into an Asus CUBX BX-motherboard.
Benchmark Setup
Platform Information | |
Graphics card for all tests | NVIDIA GeForce 2 GTS Reference 200MHz Core, 333MHz DDR-RAM 32MB |
Hard Drive for all tests | IBM DPTA-372050, 20.5 GB, 7200 RPM, ATA66 |
SocketA System |
|
Motherboard | Asus A7V, ACPI BIOS 1001, June 2000 |
Memory | 128 MB, Enhanced Memory Systems PC133 HSDRAM CAS2 |
IDE Interface | Onboard ATA66 |
Network | 3Com 3C905B-TX |
Celeron 1 GHz System, FSB 100 MHz, Multiplier x10, Celeron 667 |
|
Motherboard | Asus CUBX, ACPI BIOS 1006 beta 03, June 2000 |
Memory | 128 MB, Enhanced Memory Systems PC133 HSDRAM CAS2 |
IDE Interface | Onboard ATA66 chip |
Network | 3Com 3C905B-TX |
Pentium III System |
|
Motherboard | Asus CUSL2, BIOS 1002, Intel 815E chipset |
Memory | 128 MB, Wichmann WorkX MXM128 PC133 SDRAM CAS2, settings 2-2-2, 5/7 |
IDE Interface | Onboard ICH2, ATA100 capable |
Network | 3Com 3C905B-TX |
Driver Information | |
Graphics Driver | NVIDIA 4.12.01.0522 |
Driver for VIA Chipsets | 4in1 4.22 AGP-driver 4.03 |
ATA Driver | Promise Ultra66 driver rev. 1.43 Intel Ultra ATA BM driver v5.00.038 |
Environment Settings | |
OS Versions | Windows 98 SE 4.10.2222 A Screen Resolution 1024x768x16x85 SuSE Linux 6.4, Kernel 2.2.14 |
DirectX Version | 7.0 |
BAPCo Sysmark 2000 | Patch 3, Resolution 1024x768x16x85, 3 Runs, previous Defrag |
Quake 3 Arena | Retail Version command line = +set cd_nocd 1 +set s_initsound 0 Graphics detail set to ‘Normal’, 640x480x16 Benchmark using ‘Q3DEMO1’ |
Linux Kernel Compilation | ‘time make dep clean bzImage’, Kernel 2.2.14, THG-conf. file |
3D Studio Max 2 | Rendering file ‘ktx-rays.max’ to 640×480, time reported by software |
Office Application Benchmark
Duron still shines against Celeron even if both are overclocked! A Duron 900 is already faster than one of the rare Celeron 1 GHz in Sysmark 2000. Thunderbird may not be quite as fast as Pentium III on a Solano platform, but a 1 GHz Thunderbird at 1050 MHz can already skip the GHz Pentium III, which is virtually unoverclockable.
3D Gaming Benchmark
In Quake 3 Arena the picture looks even better for Duron. Once the little AMD processor is overclocked to 850 MHz, which should be possible with any Duron, it can top a Celeron overclocked to 1 GHz! Thunderbird needs the 1050 MHz once more to leave Pentium III 1GHz behind. However, please be aware of the fact that it is virtually impossible to overclock Coppermine over 1 GHz right now.
Linux Kernel Compilation
To make the understanding of this chart a bit easier I decided against the publication of the time it takes to compile the kernel. ‘Smaller is better’ is something many people have trouble to understand. Instead I divided 3600 seconds by the time it took to compile the kernel and published this number. Thus in the above chart ‘larger is better’, which should please everybody.
It was very interesting to see the big difference between Thunderbird and Duron. Obviously the kernel compilation under Linux likes a lot of L2-cache. That’s why Duron lags quite a bit behind Thunderbird. However, Duron 950 is still faster than Celeron at 1 GHz. Thunderbird looks very strong here. At 950 MHz it’s already a bit faster than Intel’s Pentium III at 1 GHz.
Floating Point Performance
You are already used to me dividing 3600 seconds by the time it takes to render the ‘ktx_rays.max’ file, so that this chart also goes by ‘larger is better’. You can see that the FPUs of Thunderbird and Duron are identical and so is the bus speed of both processors. Therefore we cannot expect any difference between the FPU-performance of the two. We also know that the FPU of the latest Intel processors is much weaker, so you can see that a Pentium III at 1 GHz is just as fast as a Duron 700 (which costs a fifth) in this benchmark.
Conclusion
After this highly successful story, which looks extremely promising as well as so much easier than what Athlon-overclocking used to be before, we’ve got all reason to welcome AMD’s new processors even more. Here are the facts why:
- With a full-featured SocketA motherboard the overclocking of Duron and Athlon/Thunderbird should be a piece of cake.
- As you could see in the excerpts of the data sheets above, it seems extremely unlikely that AMD could revoke this ‘overclockability’ of the two SocketA-processors easily. As long as the SIP-protocol initializes those two processors we’ve got no reasons to worry. A change of that would require a serious redesign of both processors.
- Duron 650 and Duron 700 seem to go up to 950 MHz, which increases performance significantly. Even Intel’s fastest Celeron processors are hardly able to reach 1 GHz and even if they should, they are still slower than a Duron overclocked to 950 MHz.
- Thunderbird can go up to 1.1 GHz, which is the highest clock speed that we have tested on any processor so far. With some additional cooling Thunderbird might reach even more. Intel’s Pentium III is just about able to reach a few MHz more than 1 GHz if you are extremely lucky. Thus AMD is ahead of Intel once more.
Well, I hope that all overclockers are pleased with the outcome of this article. There isn’t really much that should keep any of you from going for one of those two new AMD processors. The only problem that I can see right now is the shortage of SocketA-motherboards with VIA’s Apollo KT133 chipset. In a few weeks the situation should improve significantly. I am sure that most motherboard makers will follow Asus’ example and include overclocking features into their SocketA-solutions.
AMD might hate this article, but I feel that it supplied the final reason why virtually anyybody should be happy with the latest AMD processors now. The wind that blows into Intel’s face has become another bit stronger …
Please follow-up by reading the article Overclocking AMD’s Socket Processors