never stopped me before, good luck. Not sure how to split this up, wouldn't feel bad if you took out anything regarding choosing CPU for new build, there's less options for PII these days anyway. (actually, if you would just edit all my posts all the time, everyone would be happier. =P)02/01/2012 10:46 PMPosted by MÿxxïIf you're feeling VERY wordy and would feel better just making a word document
AMD Athlon 64. Athlon II, Phenom and Phenom II CPUs overclock similarly to prior AMD processors. A main referance "bus"(often called the FSB, but it technically isn't) with a default speed of 200, having a series of multipliers for cpu clock and actual bus speeds. Phenom II "Black Edition" (commonly "BE") processors have an unlocked CPU multiplier, they are made for overclocking and are obviously the most desirable. It is much easier to overclock with a BE processor since you can leave the main reference clock close to 200 and not worry about memory OC at all. The non-BE Phenom or Phenom II's (or any of the Athlon line) are still able to be overclocked, but it takes more work to set up and you need a more "enthusiast-friendly" motherboard to adjust all the multipliers - or else you will be overclocking the memory too much, which is harder to get stable and usually not necessary.
If you are building from scratch for overclocking, it is relevant to know how AMD "classes" and numbers their CPUs. (as far as I know they have always done it this way) All AMD CPUs having the same core type and stepping (ex. Deneb core C3-stepping) are the same chip, produced in large runs and then tested. Seeing as how millions of transistors are formed on the chip, they do no all turn out equal. The CPUs are tested similarly to what we do with overclocking, to find which chips will run what speed with what voltage. Chips that do better are given higher model numbers and sold set at higher clocks. The most popular Phenom II's for overclocking are probably the 955 and 965 x4 Black editions having the later "C3" revision(125w). That is not to say others are bad, these were just the best ones during the height of Phenom II overclocking's popularity and are also as much as you need to spend to reach top performance. The vast majority of C3 Deneb x4s run at 125w and these are normally recommended for a quad core build. 1055T and 1065T are popular x6 Phenom II processors, although 6 cores is typically seen as overkill for gaming since few games take advantage of more threads(chips with less cores are typically clocked higher to begin with). Higher numbered models are technically a better chip, but are most often already pushed pretty far and have similar max overclocks to slightly lower models. Many vendors are running out of the lower numbered chips since AMD has moved on to producing the FX-series chips.
You need a good CPU heatsink for overclocking, even without touching voltages. The stock AMD heatsinks that come with Phenom and Phenom II CPUs are decent, but won't keep you safe over standard Phenom II speeds. The Cooler Master Hyper 212+ is very popular and fits on AM2(+) and AM3(+) sockets - for the money, this is about the best cooling you can get without going to liquid cooling or something huge(er). There are also water cooling options that come prefilled and intended for internal mounting that are easy to set up. If you're building for overclocking, you want good cooling and an unlocked multiplier.
Generally for the beginner, leave the clock at 200, leave the voltage on default or "Auto" and adjust the CPU multiplier, .5x at a time(or smallest increment), until the system becomes unstable, then back it off. Before you start, get a program that lets you monitor your temperatures. HWmonitor is a commonly recommended free program, although there are many more advanced monitoring programs like AIDA64 if you want to pay a little money(AIDA64's "stability test" is not as good as anything I will mention for that). Per AMD, the "Maximum operating temperature" of most Phenom IIs is 55C - 62C. That is not to say that your CPU will instantly become blue-smoke at those temps, but to avoid damage you should make sure your processor never reaches them! Searching for your CPU's model number on http://www.cpu-world.com/ will tell you many things including max temp and supported voltage. Programs like CPU-Z, AIDA 64 and others will tell you your chip's exact model and stepping if you don't know it.
<<< 2/14 edit >>>
Stability in your Overclocking
There are many ways to check stability. Booting into windows is the first test, but does not mean you're in the clear! Monitor your CPU temps throughout any testing, if it gets too high, stop! Being able to run benchmarks like PCmark or 3Dmark without crashing shows some degree of stability. Better yet, specific programs can me used to stress your CPU and memory. MemTest86 is not as good as it used to be for checking memory overclock (and it takes many tests over a very long time). There are stress tests like Prime95 which make your processor continuously do math at 100% to stress it, using variable amounts of system RAM. I found Prime95 to be the most demanding test you can run. It might be dangerous to run Prime for days on end like some people do - usually 15-20 minutes is good enough and surviving an hour counts as "solid" to most people. "Linx" or "Intel Burn Test" can also be used for this, completing several passes of these extremely harsh tests will stress your system at least as much as Prime will and are usually faster. For stress-testing, make sure the program you use is set to run as many threads as you have cores. Hyper-PI is a program that will run multiple instances of the popular "SuperPi" benchmark simultaneously on multiple cores, computing 32M completely on all cores is a decent test. Used correctly, all of these programs are more "stress test" than they are "benchmark" and can help verify your overclock as stable and cool. Maximizing your CPU via overclocking
When overclocking, the above tests will eventually reveal instability by returning errors or crashing the system. When you discover the limit of your stability, voltage is usually the first thing to look at when you want to go further. With a simple multiplier overclock, the CPU voltage(vcore) is usually the only relevant one. If raising this value by a small amount does not help, the problem may lie elsewhere, either with temperature or memory settings. With larger multiplier-only overclocks, other voltages will sometimes have an effect on stability, but don't go ahead and change them without first researching! It can get confusing since the BIOS of different motherboards use different names for voltage and bus settings, be sure to research your specific board before you mess with it. Among others, "CPUNB" and "NB" can mean entirely different things on different boards, please do some googling before you start over-volting something that doesn't need it(for example, in specific cases, one of the "NB" voltages might refer to on-board GPU instead of what you want).
In order to "go all out", a stable memory overclock is usually achieved first (with a low CPU multiplier) before attempting to reach the desired CPU speed. Doing it in this order is the only way to maximize both speeds. Get some CPU benchmark programs to make sure the memory overclock is actually giving you significant improvement. Use something that runs a lot of math using a lot of memory, SuperPI is popular and free.
AMD CPUs reportedly respond better to memory overclocking than Intel CPUs, but again this requires more BIOS options and more risk than a simple multiplier overclock(read: it's harder). Specifically, low memory latencies have much more of an impact here than on Intel systems. The most obvious factor effecting overall memory latency is the "timings" that your RAM is set to. Better RAM supports better timings out of the box - CAS Latency(CL) is usually the first (and most important) number in memory timings - lower is better for timing #'s. To stay stable, it is generally recommended to choose quality RAM of the speed you want with a low CL and keep memory overclocking to a minimum. The overall latency is not always linear with timings and can be measured with certain programs (PCmark and AIDA64 include this test). Phenom II's love low latency above all else and yet are happy at all but the lowest supported RAM speeds but memory overclocking is said to be more important with the FX-series chips. Due to flaws in the "Bulldozer" micro-architecture's current implementation, FX systems are accepted by many to be held back or "bottlenecked" by all but the fastest, lowest-latency memory(FWIW, many of the initial reviews that made everyone hate FX were done with 1333, while the more recent tests are often 1600 or 1866).
On all current AMD systems (with or without overclocked memory) it is common to increase the multiplier for the "northbridge frequency" and raise the associated voltage if necessary - this speed often effects memory latency as much as the timings and is seen as an essential part of a full AMD overclock. It is important to note that this NB memory bus is different from the Hyper-Transport link. Increasing HT-Link speed has almost no effect besides lowing stability, it is best to keep it near stock speed so you don't need extra voltage and heat.
Going over the recommended voltage for RAM modules is not usually necessary and not a good idea in general since it often voids the RAM's warranty. In AM-socket systems, the DDR2 and DDR3 memory controllers are inside the processor chip, and the relevant memory controller will have its own voltage setting(s) in BIOS. These voltages are much more important to stable memory than the DDR voltage itself. Being on-chip, altering these voltages increases load on the CPU heatsink. HT voltage can sometimes add stability, but this is mostly with large overclocks and/or specific motherboards(did I mention researching your board?).
Although reliant on modified BIOS's in the past, with modern motherboards the memory "command rate"(CR) can often be set to either "1T" or "2T" with a BIOS setting - adjusting how often the CPU accesses the memory(some people will still call this "CPC-on"/"CPC-off" respectively, meaning "Command Per Clock"). 2T command rate increases stability, but 1T is faster(but not as much faster as it should be!) For many years AMD has designed their stuff to tolerate 2T command rate in lieu of better error-correction, 2T is quite common for overclocked AMD systems and not as big of a performance hit as it sounds like. Running CR2 is fine if you have a lot of RAM or pushing your RAM faster than supported, or if you are running an unsupported memory configuration. For Phenom II, filling all 4 slots of DDR2-1066, or running any DDR3 over 1333 is technically unsupported. FX-series currently supports dual-channel up to 1866. Most of these "maximums" can be violated with quality ram and luck. DDR3-1600 is quite common on Phenom II, although many have had to set it to 1333 and run a 240 reference(keep in mind that most buses like HT and NB frequency assume a standard 200mhz ref clock and therefore must be turned down accordingly when significantly increasing the reference clock).