Last update :- February 19th, 2012

Quick index:

Click on any image below to see a larger version.

Athlon XP2600+ @ 2412

Warning : if you overclock your processor or graphics card you will invalidate any warranty

You can already see from the title of the page that my system was running at 2412MHz as opposed to the default 2088MHz which only appears to be a modest 324MHz overclock but that doesn't explain the whole story. First I'll try and explain some of the theory with specifics to my motherboard (the EPoX 8RDA+) before going onto the actual results and how they were achieved. Finally, I'll include some popular benchmarks comparing the system before and after the overclock.


One thing that you have to note before overclocking your system is that all components are not made equal. Unless all your system components are picked especially (if you can find them) the level of overclocking cannot be guaranteed. In particular, the level of overclocking you can achieve depends upon the following components, in no specific order:

For instance, you may have a highly overclockable CPU (processor) but you won't see the best performance if the memory or motherboard chipset won't overclock very well. Therefore, when you overclock you cannot expect anything but only hope for some kind of performance gain.


If you remember from my earlier discussions with respect to overclocking, there are a number of things you have to consider:

  1. Clock multiplier
  2. FSB changing
  3. Voltage
  4. Cooling

From that list, all but cooling could be handled via the EPoX 8RDA+ BIOS (if you want more information on the BIOS for the 8RDA+ search for a review of this board). This was one of the main considerations when choosing the motherboard originally as I had always intended to do some overclocking at some point. Like my previous Abit motherboards, EPoX are also renowned for the overclocking potential of their motherboards - rarely using jumpers and mainly allowing adjustments via the BIOS. A lot of other manufacturers choose to either use jumpers or no overclocking features what so ever.

Let's look at each of these aspects of overclocking in turn - with particular reference to the system I had.

1. Clock multiplier

If you recall, the first Pentium II's and Celerons had their clock multiplier unlocked so you could simply adjust the clock multiplier and get a faster processor (CPU speed = Clock Multiplier x FSB). Intel and AMD changed that with the later generation Pentium III, Celeron II, Athlon and Duron. You could unlock the clock multiplier on the Athlon/Duron using the now famous "pencil trick".

I don't know about the Intel Pentium IV (as they're too expensive) but the AMD Athlon XP also has it's clock multiplier locked. Again, you can unlock the multiplier but they've made it more difficult. This, however, changed with the introduction of some nForce2 based motherboards such as the EPoX 8RDA+. I believe I'm correct in stating that this isn't the case with all nVidia nForce2 based motherboards (by all means correct me if I'm wrong) but the popular overclockers choices from EPoX, Abit and ASUS unlock the multiplier via the socket.

8RDA+ specific

Being an overclockers board, the EPoX 8RDA+ unlocks the clock multiplier of the Athlon XP in the CPU socket. Clock multiplier settings available are 3 to 4 (in 1.0 steps), 5 to 16.5 (in 0.5 steps) and 17 to 24 (in 1.0 steps).

2. FSB changing

The next option to try and change is the FSB (Front Side Bus) frequency.

For the range of AMD processors available at the time and their default FSBs see the table below:

Duron 100MHz
Athlon XP "Palomino" (up to 2100+) 133MHz
"Thoroughbred A" (up to 2600+)
"Thoroughbred B" (up to 2700+) 166MHz
"Barton" (up to 3000+)

The FSB also has a direct relationship to the other busses on the system.

The table below shows which processors support which bus speeds and the ratios for these busses:

Bus speed AMD AGP PCI
100MHz Duron 2/3 1/3
133.33MHz Athlon 1/2 1/4
166.66MHz Athlon 2/5 1/5

Due to the table above you have to beware when using non-standard FSB's. An AGP or PCI card my not work or may not be stable. In particular, the hard disk interface is based upon the PCI clock and corruption or damage can readily occur with a high PCI clock.

DDR memory is also available in different flavours and speeds to match your FSB requirements:

Name Type Clock DDR
PC2100 DDR266 133MHz 266MHz
PC2700 DDR333 166MHz 333MHz
PC3200 DDR400 200MHz 400MHz

In reality, the FSB is based upon a multiple of the PCI clock which is specified as 33.33.....333MHz (to as many decimal places as you can think of). That's why an FSB speed of 133MHz is actually 133.33MHz (4x33.33) and 166MHz is actual 166.66MHz (5x33.33).

The advantage of increasing the FSB is that if the rest of your system components can support it you end up with an overall system that is faster. For two processors clocked to the same overall speed, the system based upon the higher FSB will perform better overall. For example, a system based on a processor clocked at 8x166MHz (1333MHz) will perform better than one based upon a processor clock at 10x133MHz (also 1333MHz) because your memory and motherboard chipset are running at the higher FSB also.

8RDA+ specific

The good news with all nForce2 based motherboards is that the PCI clock is locked to 33MHz leaving the AGP clock as the only one to worry about. The EPoX 8RDA+ allows you to set the AGP clock indepdently of the FSB so that takes that out of the equation. Therefore it comes down to how high a FSB your processor, motherboard and memory will support.

FSB settings available on the 8RDA+ (29/01/03) are 100, 102 - 123, 125, 127 - 148, 150, 152 - 175 and 177 to 250 (all MHz).

The memory speed can be set to a % of the FSB speed with settings available of By SPD (memory manufacturer specified), 50%, 60%, 66%, 75%, 80%, 83%, 100%, 120%, 125%, 133%, 166%, 200%, Auto.

Now, you may think that with a processor based upon an FSB of 133MHz installed in a system with PC2700 (166MHz, DDR333) memory that it would be better to set the FSB to 133MHz and memory to 125% (166MHz). With the nForce2 this is not the case as it's designed to perform better when the FSB and memory frequency are synchronised. Therefore you would set the memory frequency to 100%. The situation improves even further when you use two (or more) sticks of memory (preferably of the same type and specification, if not from the same batch) in DIMM 1 and either DIMM 2 or 3. This is because it enables the dual 64-bit memory controller (ie, 128-bit) rather than the normal 64-bit - effectively doubling the memory bandwidth and therefore improving the performance.

Finally, with the 8RDA+ the AGP frequency can be set to Auto, 50, 66 - 87, 90, 93, 95, 97 and 100 (all MHz).

3. Voltage

If you do decide to try and overclock the processor, motherboard chipset and memory (and maybe graphics card) you're probably going to need to increase the voltages supplied to these to achieve a higher speed.

The specifications for the most common processors already covered here are available from AMD (it's best to right-click and choose "Save as" and you'll need Adobe Reader from here) and the core voltage is shown below:

Processor Speed/Model CPUID Vcore Max
Duron (Model 3) 650 to 950MHz ?? 1.6V 2.1V
Duron (Model 7) 900 to 1300MHz ?? 1.75V 2.25V
Athlon XP (Model 6) Palomino 1500+ to 2100+ ?? 1.75V 2.25V
Athlon XP (Model 8) Thoroughbred 1600+ (A) 680 1.6V 2.1V
1700+ (A) 680 1.5V 2V
681 1.5V 2V
1.6V 2.1V
1800+ (A) 680 1.5V 2V
681 1.5V 2V
1.6V 2.1V
1900+ (A) 680 1.5V 2V
2000+ (A) 680 1.6V 2.1V
1.65V 2.15V
681 1.6V 2.1V
2100+ (A) 680 1.6V 2.1V
681 1.6V 2.1V
2200+ (A) 680 1.6V 2.1V
681 1.65V 2.15V
2400+ (A) 681 1.65V 2.15V
2600+ (A) 681 1.65V 2.15V
2600+ (B) 681 1.65V 2.15V
2700+ (B) 681 1.65V 2.15V
Athlon XP (Model 10) Barton 2500+ ?? 1.65V 2.15V
2800+ ?? 1.65V 2.15V
3000+ ?? 1.65V 2.15V

To find out which processor you've got grab a copy of CPU-Z.

The Max column above is taken from the "Absolute Ratings" section of the "Electrical Data" in the above datasheets which states:

"....should not be subjected to conditions exceeding the absolute ratings, as such conditions can adversely affect long-term reliability or result in functional damage."

and in all cases is 0.5V above that for Vcore.

System specific

The 8RDA+ allows Vcore to be changed between 1.4V and 2.2V in 0.25V steps. Personally, I would only recommend setting Vcore to a maximum of the nominal + 0.3V or 0.35V rather than the full 0.5V specified above. The increased Vcore will definitely produce an increase in temperature and reduction in lifetime. Using 0.3V or 0.35V still allows some headroom for variation in the PSU rails under load conditions and inaccuracies of monitoring software.

The AGP slot and DIMM voltages are also adjustable for increased stability when overclocking. The AGP voltages available are 1.5V (default) to 1.8V in 0.1V steps, whilst the DIMM voltages available are 2.5V (default), 2.63V, 2.77V and 2.9V.

The voltage to the nForce2 chipset (Northbridge and Southbridge) cannot be changed in the BIOS so an alternative method of adjustment may be needed.

4. Cooling

With modern day processors (and associated components) running in excess of 1GHz they produce a lot of heat - especially AMD processors. This heat needs to be dissipated somewhere, first off the components themselves and then out of the system case.

The specifications of AMD's current range of processors are as follows:

Processor Speed/Model Tdie
Duron (Model 3) 650 to 950MHz 90°C
Duron (Model 7) 900 to 1300MHz 90°C
Athlon XP (Model 6) Palomino 1500+ to 2100+ 90°C
Athlon XP (Model 8) Thoroughbred 1600+ to 2100+ 90°C
2200+ to 2700+ 85°C
Athlon XP (Model 10) Barton 2500+ to 3000+ 85°C

Tdie from the above table is maximum temperature as measured by all of the above processors via an external temperature sensor in conjunction with an internal diode (with the exception of the Duron Model 3 which doesn't have one). Motherboard manufacturers are required to use this fact to shutdown the system once this temperature is exceeded:

"Thermal solutions must monitor the processor temperature to prevent the processor from exceeding its maximum die temperature."

Unfortunately, most motherboard manufacturers don't make this temperature available via monitoring software. Usually, an in-socket thermal sensor is used where the temperature is some 10°C below that of the thermal diode. If you can get a thermal sensor that is small enough to tape in the center of the processor on the underside, this is an ideal alternative as AMD quote temperatures measured in such a way are within 2°C of the internal diode. Personally I chose to have a system that runs at a maximum in-socket temperature of 55°C (equates to Tdie = 65°C) if the stated maximum is 85°C. I would prefer to have a system running cooler than that as the lower the temperature the more you can overclock. Often though, the Vcore has to be raised dangerously high before these levels are reached.

AMD produce a list of processor heatsink/fan combinations they recommend for their processors. They also produce a system cooling guide which recommends at least one case exhaust fan (80mm behind the processor), preferably an inlet fan (at the front or on the side) and a PSU with inlet and exhaust fans.

When you raise the voltage of or clock an IC (processor, graphics, motherbaord chipset, memory) at a higher speed than specified the temperature will increase as a result because it's being stressed more. Therefore you have to look at better cooling to get the temperature down.

If you want to overclock your system you will probably be able to acheive a mild overclock before temperatures start getting too high. The heatsink and fan supplied with retail boxed processors in most cases will not be up to the job. Therefore you have to consider a better solution. Additional case cooling with an inlet fan at the front of the case and an exhaust fan at the back of the case (behind the processor) will also help.

A variety of cooling methods are available with the costs proportionate. For extreme overclocking, using a Vapourchill or Prometia system will drop temperatures below 0°C but at a substantial price. For the next level of overclocking you can get reasonably priced water cooling (sometimes in conjunction with peltiers) and run your processor in the mid 20's to high 30's. With a low-cost, good quality heatsink and fan you can expect to acheive temperatures in the low 40's to low 50's under load conditions.

Other components such as the motherboard chipset and memory may also benefit from better cooling if they're being pushed beyond their stated specifications.

System specific

I chose the GlobalWin CAK4-88T copper heatsink and fan as it was rated for the AMD XP3000+ and above to go with the XP2600+ I was going to overclock.

When installing the heatsink and fan it's important to make sure there's a good contact between it and the processor and for it to be level. You should also use a thin layer of a good heatsink compound or thermal grease in between the heatsink and processor. Most web-sites recommend "Arctic Silver III" which is pictured below which I used.

Arctic Silver III

Arctic Silver III

My Thermatake Xaser II aluminium case provided additional cooling via two methods. First, the case being aluminium itself acts to some extents as a large heatsink when compared to a steel case. Secondly, it has 5 x 80mm fans each pushing 35CFM of air at 21dBA (very quiet) when running at 12V. Three of these are intake (front at the base, front in front of 3.5" hard disks, side above graphics card) and two are exhaust (directly behind the processor). These are controlled via a fan switch which has settings of H/M/L but I tended to leave it at H as the sound was unobtrusive.


At the outset of overclocking I set myself some targets. I decided I was only willing to set the Vcore of the XP2600+ to a maximum of 1.9V which was 0.25V above the 1.65V default. I also decided I wanted to keep the in-socket temperature to less than 55°C. Based upon these settings I was hoping to clock the processor at 2400MHz or higher, which I thought would be average when compared with other users systems.

You also have to decide what level of stability you want with your system under load. If you intend using your spare processing time for Seti@home, Folding@home or similar distributed computer programs where the data has to be 100% reliable you need to be able to run your system error free using a precise program such as Prime95 for at least 24 hours without error before contributing to these projects. As I didn't intend upon running this type of application I decided a system would be stable enough for me if it could run a few loops of 3DMark2002 SE without crashing or visual errors.

1. Clock multiplier

The XP2600+ has a default clock multiplier of 12.5 so the first thing I tried was to up this to 13.0. On re-boot I soon realised that multipliers above the default were locked as the BIOS complained of overclock failure. Dropping the multiplier to 12 resulted in the system booting okay so therefore the multipliers below the default 12.5 were available.

2. FSB changing

The 8RDA+ is only guaranteed to support an FSB of 166MHz for the processor and 200MHz for the memory.

The first thing I wanted to establish was the performance of the memory, irrespective of the processor/chipset. Therefore I download the popular Memtest86 memory test program and kept the processor FSB at 166MHz in the BIOS, whilst adjusting the memory frequency. The default memory timings (By SPD) for my TwinMOS PC3200 CAS2.5 memory were 8-3-3-2.5. I could successfully run the memory up to 222MHz (133% of FSB, DDR444) with timings of 7-3-3-2 at 2.9V.

Next up was the processor FSB. I decided I was only willing to run the system in synchronous mode with the processor and memory clocks the same. I also decided I would only be willing to drop the multiplier to a minimum of 12x during actual use so tested using this as a base. Initial tests showed the maximum stable FSB to be ~183MHz. Slightly disappointed with this I decided to pay a visit to the EPoX forum at AOA Forums to see if there was any advice on offer. Quite a few people were having success using an FSB of 200MHz or greater straight out of the box but quite a few were also struggling. This led me to the EPoX nForce2 FAQ site which detailed a number of possible modifications.

Note: All images below, except where specified otherwise, have been reproduced with kind permission of Paul at the EPoX nForce2 FAQ. For all pictures, click on the image for a larger version.

Vdd mod

The default Vdd to the motherboard chipset was 1.58V as displayed in the BIOS and via the motherboard monitor software. It was recommended that increasing this to around 1.84V with the addition of a 680ohm resistor might help. Being an Electronics Engineer anyway and having previously modified my Abit KT7A-RAID this was a straight forward modification so I did it:

8RDA+ Vdd location 8RDA+ Resistor loaction Finished modification
Vdd location Resistor loaction Finished modification

After completing the modification the BIOS stated that Vdd was 1.84V.

Northbridge cooling

The heatsink supplied for the nForce2 Northbridge can best be described as adequate for the job but not really ideal for the purpose of moderately high FSB overclocking. For an explantion why see the "Cooling" section at the EPoX nForce2 FAQ. Bearing this in mind I decided to fit the highly recommended Vantec Iceberq copper chipset heatsink and fan (using Arctic Silver II as the thermal compound), which was an ideal fit:

Vantec Icerberq - top Vantec Iceberq - side
Vantec Icerberq - top Vantec Iceberq - side

Southbridge cooling

The nForce2 Southbridge supplied on the 8RDA+ gets very hot to the touch but is supplied with no heatsink of any kind. I chopped up an old Pentium heatsink I had lying around and added it with thermal tape. Although adding a fan as well is a bit of an overkill I just happened to have one lying around so I added that as well.

MOSFET cooling

It was also recommended to add cooling to the MOSFETs on the motherboard which regulate the 12V and 5V as supplied by the PSU to produce the other system voltages (Vcore, Vdd, Vagp, Vdimm, 3.3V, etc). In my case I chopped up the larger BGA heatsink supplied with the Vantec Iceberq and the supplied thermal tape, not as shown below:

MOSFETs and Vreg before MOSFETs and Vreg after
MOSFETs and Vreg before MOSFETs and Vreg after

After carrying out these modifications I could get a stable system using either 12x196MHz (2352MHz) or 12.5x192MHz (2400MHz). Although the 196MHz FSB offers a higher overall bandwidth, I decided to stick with the lack of 4MHz and go for a higher overall processor speed by using the 192MHz setting.

3. Voltage

In order to achieve the 12.5x192MHz final setting the following voltages were used:

4. Cooling

The images below show the resulting temperatures, voltages and associated fan speeds taken from Motherboard Monitor 5 (remembering that both CPU and PSU fan speeds vary due to temperature based upon the components chosen in my system) at default settings, 2412MHz idle and 2412MHz under load. You can see that the maximum temperature acheived is:

MBM5 XP2600+ Idle MBM5 XP2600+ @ 2412 Idle MBM5 XP2600+ @ 2412 Load
MBM5 XP2600+ Idle MBM5 XP2600+ @ 2412 Idle MBM5 XP2600+ @ 2412 Load

In summary we have the following:

I'm slightly suprised by the CPU fan speed not varying too much when taking into consideration the specifications.


The images below shown the readings taken from WCPUID before and after the overclock and explain why I state that the system was running at 2412MHz (due to the rounding up of the FSB to 192.94MHz):

WCPUID XP2600+ WCPUID 2600+ @ 2412
WCPUID XP2600+ WCPUID 2600+ @ 2412


3DMark2001 SE

The first benchmark run was the popular 3DMark2001 SE graphics test and my Leadtek WinFast A250 TD GeForce4 Ti4400. This tests the capabilties of your system with an emphasis on the graphics and gives you an idea of how well (or stable) your system may be when running 3D games is differing levels of detail. The default resolution of 1024x768x16 was used:

3DMark2002SE - XP2600+, 275/554 3DMark2002SE - XP2600+ @ 2412, 275/554 3DMark2002SE - XP2600+ @ 2412, 302/654
XP2600+, 275/554 XP2600+ @ 2412, 275/554 XP2600+ @ 2412, 302/654

In summary we have the following:

As you can see, a jump of 1383 3D marks is quite an increase and will give performance boost when playing 3D games.


Next up was the PCMark2002 Home & Office benchmark:

PCMark2002 - XP2600+ PCMark2002 - XP2600+ @ 2412
XP2600+ XP2600+ @ 2412

In summary we have the following:

SiSoft Sandra 2003 (Build 2003.1.9.31)

System Summary

Sandra Summary XP2600+ Sandra Summary 2600+ @ 2412
Sandra Summary XP2600+ Sandra Summary 2600+ @ 2412

It would appear that I now had an XP3000+ (Model Number 3028 (estimated)) for the price of an XP2600+. At the time of writing this (10th April, 2003) you could buy an XP2600+ "Thoroughbred B" for £148.04 (inc. VAT) and an XP3000+ "Barton" for £409.96 (inc.VAT). It isn't the real equivalent of an XP3000+ though because the "Barton" has 512K cache compared to the "Thoroughbred B" with 256K cache so the Barton would offer better all around performance. The nearest "Thoroughbred B" price comparison is the XP2700+ at £211.81 (inc. VAT). Which ever way you look at it, it's quite a saving!

Using AMDs XP numbering scheme the XP3000+ is the nearest equivalent of a Pentium IV 3.06GHz at £482.84 (inc.VAT). The Pentium IV also has 512K cache, "Hyper Threading" and a better memory performance but I wasn't willing to pay those kind of prices.

CPU Arithmetic Benchmark

This benchmark demonstrates how the processor handles arithmetic and floating point instructions (don't ask me what it means).

Sandra CPU Arithmetic XP2600+ Sandra CPU Arithmetic 2600+ @ 2412
Sandra CPU Arithmetic XP2600+ Sandra CPU Arithmetic 2600+ @ 2412

In summary we have the following:

Not bad when you consider that the Pentium IV 2.8GHz with 512K cache costs £312.02 (inc. VAT) - that's £163.98 more.

CPU Multimedia Benchmark

This benchmark demonstrates how the processor handles multi-media instructions and data (don't ask me what it means).

Sandra CPU Multi-Media XP2600+ Sandra CPU Multi-Media 2600+ @ 2412
Sandra CPU Multi-Media XP2600+ Sandra CPU Multi-Media 2600+ @ 2412

In summary we have the following:

Okay, so the Pentium IV 2.8GHz with 512K cache costing £163.98 more blows the XP away in Floating-Point but the Integer difference is negligible.

Memory Bandwidth Benchmark

This benchmark demonstrates system memory bandwidth performance. Note: At the time SiSoft ddn't include support for the nVidia nForce2 chipset and it's dual-channel memory controller so these figures may be misleading.

Sandra Memory Bandwidth XP2600+ Sandra Memory Bandwidth 2600+ @ 2412
Sandra Memory Bandwidth XP2600+ Sandra Memory Bandwidth 2600+ @ 2412

In summary we have the following:

RDRAM was still the highest performance memory around but at a price premium and only supported for Intel processors. My guess would be that if the PC3200 memory on the nForce2 was clocked at the specified 200MHz the gap between it and the rival VIA KT400 would increase. If SiSoft included support for the nForce2 dual-channel memory controller the gap may widen further.

Cache & Memory Benchmark

This benchmark demonstrates system cache and memory performance. Note: At the time SiSoft ddn't include support for the nVidia nForce2 chipset and it's dual-channel memory controller so these figures may be misleading.

Sandra Cache & Memory XP2600+ Sandra Cache & Memory 2600+ @ 2412
Sandra Cache & Memory XP2600+ Sandra Cache & Memory 2600+ @ 2412

If you want to advertise on this page please E-mail me at paul_at_paulslounge_dot_co_dot_uk

Copyright © Paul Collins, 2001 - 2012
Paul's Lounge
All rights reserved

Valid XHTML 1.0 Transitional

Privacy Site Map Home About Paul Blog News Links
Computers Emulators Software Games Tips & Tricks Fishing Golf