This is the first article in a six-part guide book.
Short-Media LLC, as well its owners, staff, members, and authors, take no responsibility for the use of the information contained in this guide. Modify and configure systems using this guide at your own risk. We imply no guarantee or warranty by publishing this guide.
Anyone serious about PC gaming knows that the current king of CPUs is the AMD Athlon 64. Since its introduction in 2003, it has provided phenomenal performance, at much lower clock speeds than its Intel competitors. The Athlon 64 platform has brought many new features to the table and with these new features comes many new overclocking challenges. The purpose of this guide is to provide both theoretical insight into the ‘art’ of Athlon 64 overclocking, and to provide hands-on examples to assist you in your endeavours. There are numerous overclocking guides available online, but I hope that this guide will present a unique and fresh approach to a commonly misunderstood practice. The information I will present is an accumulation of knowledge that I have obtained over the past five years as a hardware and overclocking enthusiast. There is no clearly defined ‘proper’ way to overclock, but the theory and logical procedures that I outline will help you to avoid many otherwise inevitable headaches and frustrations.
This guide assumes some knowledge of computer hardware and software. However, it should be thorough enough for just about anyone to follow along and understand.
I will use the term ‘Athlon 64’ or ‘A64’ during the article quite frequently, but for all intents and purposes, any processor operating on the socket 754/939 platform is within the scope of this article. This includes, but is not limited to ‘Athlon x2’, ‘Sempron/Sempron 64’, ‘Opteron 939’ and ‘Mobile Athlon 64’ processors. The socket 940 platform is not within the scope of this article, however much of the theory still applies.
[#] Indicates division between articles.
- Hyper Transport
- Integrated Memory Controllers
- Cool n’ Quiet
- Power, Amperage and Voltage 101
- Heat and Cooling
- How to Overclock an A64 (in a nutshell)
- Overclocking the HTT bus, is it useful?
- A64 Mathematics
- A64 CPU Multipliers
- CMOS, Bus Locking, Memory Modules and Timings
- CPU Stepping Codes, CPU Selection, Mobile Athlon 64, and Hardware
Overclocking Tools 
Step 1: Getting Started 
- Qualify System for Overclocking
- Testing Configuration
Step 2: Finding the Maximum CPU Clock
- Finding the Maximum CPU Clock Speed
- Longer-term stability testing
Step 3: Finding the Maximum Memory Clock at the Best Possible Timings. 
- Knowing Your Memory
- Stability in the Operating System: Painting a very different picture
Step 4: Balancing Memory and CPU Clock 
- Selecting a safe maximum CPU and Memory clock
- Long-term Stability Testing
When AMD designed the Athlon 64, it was a big step in a new direction that really set it apart from its predecessors. The 64-bit instruction set was certainly its most unique feature. However, there are numerous other features, which are relevant to overclocking that we’ll discuss. New acronyms like LDT, IMC and HTT have scared many ‘old school’ overclockers. But, once you read a bit about these new features, you’ll realize that there is nothing daunting about these new platforms and you’ll be ready to get down to business.
AMD’s A64 platforms have abolished the ‘Front Side Bus’. The ‘Front Side Bus’ was essentially a data bus that carried data to and from system components and the CPU (usually by connecting the CPU to the north and southbridge chipsets). These chipsets provide connections to other buses, such as the AGP and PCI bus, and many other system components. All current Intel platforms, and pre-A64 AMD chips follow this basic model.
AMD decided to do things a little differently with the A64, and adopted ‘Hyper Transport Technology’. Hyper Transport is a high-bandwith, low latency computer bus that replaces the aging FSB. The Hyper Transport bus does essentially the same thing as the FSB, only much faster. Many people find themselves still calling it FSB, but for the sake of correctness, we’ll call it the HTT bus.
There is a common misconception that Hyper Transport Technology is a proprietary AMD technology. HTT was developed by the ‘HyperTransport Technology Consortium’. Hyper Transport, sometimes called LDT (Lightning Data Transport) has been used by many vendors, including nVidia and Cisco Systems. You may recall that nVidia used HTT to provide high bandwidth, low latency communication between the north and southbridge chipsets in their older socket A ‘nforce’ platforms.
The Hyper Transport bus operates using a multiplier system to derive its overall speed. The ‘base’ or lowest HTT frequency that the HTT Consortium defined is 200MHz. The overall operating clock speed that the HTT operates at is simply a multiple of that 200MHz base or ‘reference’ frequency. Many other clock frequencies are also derived from this 200MHz reference clock, such as CPU clock speed. Most Socket 754 A64s, for example, operate at an HTT speed of 800MHz. A clock multiplier of 4x was used to obtain this. So 200×8 = 800MHz. You may be asking why AMD lists an HTT speed of 1600MHz in the specifications for these processors. HTT is a ‘double pumped’ or ‘double data rate’ technology, much like DDR RAM. So, simply double the final result. 200×8 = 800MHz x 2 = 1600MHz. I’ll get more into A64 Mathematics later on.
Integrated Memory Controller (IMC)
The primary function of the ‘Northbridge’ in older ‘pre-A64’ platforms was to provide an interface between the system memory and the CPU. You may notice that most A64 mainboards only have one chipset, as opposed to the north/south pair seen on older Socket A boards. All memory read/write requests had to traverse the FSB before arriving at the northbridge.
AMD decided to incorporate an ‘on-die’ memory controller in all Athlon 64/Sempron 754 models. This memory controller operates at the same clock speed as the processor, which results in very low access times to main memory. Since this IMC is on-die, there is no external data bus that memory traffic needs to traverse either. This on-die memory controller has proven to make the A64 one of the best gaming processors available. Low memory latency often equates to higher numbers of frames per second during gaming. The IMC also poses some challenges and benefits when overclocking, and we’ll see why in later sections.
Much like the late Socket A platforms, the A64 is designed to use PC3200 DDR memory, which has a base speed of 200MHz (400MHz double data rate). Memory dividers can be used for slower PC2700/PC2100 memory without having to reduce the reference clock speed. There will be more discussion surrounding memory modules and dividers ahead.
Cool ‘n’ Quiet
Cool ‘n’ Quiet is another interesting feature that AMD has incorporated into its latest platforms. It essentially throttles down the processor to a fraction of its operating frequency by using lower multipliers and decreasing the core voltage when idle. This equates to much lower temperatures and power consumption. When the CPU load increases so does the clock speed and vcore. Cool ‘n’ Quiet is a feature that most overclockers disable. Having the processor automatically decrease voltage when running clock speeds out of spec can cause instant instabilities and other issues. Having more predictable behaviour from the CPU is very beneficial when looking for high overclocks. Cool ‘n’ Quiet technology, although not generally used by overclockers, is a blessing in disguise. We’ll see why later.
Next section: Overclocking Theory