• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

OC'ing an AM2: An update to Easy 1, 2, 3, Overclocking ...

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

QuietIce

Disabled
Joined
May 7, 2006
Location
Anywhere but there
As infinitevalence noted in his thread, this is not the definitive guide to over-clocking as much as a personal, down & dirty method to get to the tweaking stage. I personally want to thank infinitevalence for the hours of time I by-passed using his basic methods and hope he takes no offense at this obvious high-jack of his short-cuts. They've served me well over the years and I honestly appreciate it, as should everybody who learned this system the way I did!

This thread is all about new OC'ers to the AM2 platform. I'll try to keep this and the following five posts updated as I'm able.

If you have suggestions, comments, or additions on how to make this thread better or easier for beginners I'm all ears! :)

Legal Disclaimer: Note that changing values in BIOS can cause damage to your computer hardware! Overclocking is not manufacturer friendly and most manufacturer warranties do not cover components that have been over-clocked! Neither the author nor anybody associated with OCForums.com is responsible for hardware damage!

(You know I had to do that.)


Another OC Thread?!? :eek:
Anybody looking at how to get started overclocking (OC'ing) has probably seen infinitevalence's thread on the A64. I used this very thread to get through my first OC and it taught me a LOT about A64 systems in the process. But that was almost two years ago when the socket 939 platform was king. Over these two years the AM2 platform has become dominant in the AMD world with AM2+ already released. A64 K8's are the CPU of the AM2 platform as they are for the s939 platform, but the sub-systems around the CPU have changed a little and many things in the original thread are confusing to newer over-clockers looking for help. This thread is an attempt to update the theories and methods posted by infinitevalence, not replace them. I've also been more verbose in some areas to point out little things some may find interesting. If this doesn't interest you that's fine, I made up a "Cheat Sheet" in post#26. ;)

The biggest change is the DDR RAM of s939 being replaced with DDR2 RAM used on AM2 systems, a much cheaper solution today. There have also been experiments with the HT Link leading us to the conclusion that overclocking the Link doesn't really accomplish much on the performance side and certainly not enough to risk losing valuable data! Motherboard naming conventions have also changed and while all of them cannot be covered, hopefully there will be enough information here to lead potential OC'ers in the right direction. With that in mind let's start!

Naming Conventions, BIOS Labels, and Common Values
NOTE: CPU-Z labels are taken from CPU-Z v1.44.1

The System Clock
The crux of the s939, AM2, and AM2+ platforms, the system clock has had many names over the years and is often referred to as the FSB. This is and has been a misnomer since the A64 came out! The closest thing to a "FSB" the A64 has is the HT Link, but more on that later. The system clock (or just "the clock") is a 200 MHz clock which, with help from multipliers and ratios, governs the speed of the CPU, RAM/memory, and HT Link. To adjust the clock setting you often have to find some label usually containing "overclocking", "OC", or "configuration" and set it's value to "Manual" or "User", which will highlight or show other options. The clock is usually adjusted using the "+"/"-" keys or entering a new number from 200-400. CPU-Z labels this the "Bus Speed" (but at least they finally got rid of the "FSB" label!).

CPU Multiplier, Speed, and vCore
The CPU Multiplier (aka CPU multi), when multiplied by the system clock, gives you the CPU speed. CPU speed is the most important part of over-clocking and has the greatest impact on performance bar none - it's the Holy Grail of OC'ing. Most AMD CPUs have a multiplier that is "upward locked", meaning you can't raise the multiplier above the factory setting. The FX and Black Edition series are exceptions to this rule but all others are locked. However, the multiplier can be clocked downward. Most BIOS' use some form of "CPU Multiplier" as a label for this parameter so it should be easy to find. Optional values will start with the default CPU multiplier and go down by 0.5x multipliers.

vCore is electronic short-hand for the CPU Voltage. vCore is labeled vCore, CPU Voltage, Core Voltage, VID and others. It is often found with the other CPU overclock settings but can sometimes be found on it's own page with other system voltages. AFAIK all AM2 CPUs have a default vCore of 1.30-1.35v per AMD specs but options usually range from less than 1.0v up to 1.55v or farther. Some boards have odd voltages with a 0.05v spacing, others are spaced at 0.025v intervals, and others get tighter still to 0.125v. Other boards don't list voltages at all and instead use a "%" method where the extra percent is added to the stock vCore.

NOTE: As with the clock setting you sometimes need to set another value in BIOS to find the CPU configuration options. CPU-Z labels CPU speed as "Core Speed" and vCore as "Core Voltage".

Memory Timings and vDIMM
The Memory Timings are not just for advanced OC'ers, everybody should use them to stabilize their system. Memory timings are often grayed out unless a "MemTiming", "Timing Mode", or "Timing and Configuration" option of "Manual" or "User" is set. The main timings we are concerned with are labeled CAS (or CL), tRCD, tRP, and tRAS. These are the basic timings listed in most RAM specifications and can often be found on the RAM stick in that order. Sometimes the BIOS does not have them in that order, so be careful! CAS, tRCD, and tRP values usually range from 3-10 and tRAS values have a range of 6-24 or more. There are often other timings here which you can explore after checking out the OCF Memory section. For this thread we'll leave these advanced timings on Auto and not mention them again.

vDIMM is also electronic short-hand and is the RAM Voltage or DRAM Voltage. vDIMM is usually located near the vCore in BIOS but sometimes it's with the other RAM settings. vDIMM can be labeled many things including DRAM Voltage, DDR2 Core Voltage, DVID and more. vDIMM is 1.8v default for "normal" DDR2 but performance RAM is often higher than 1.8v. Check the RAM stick or manufacturer's specification sheet for the default vDIMM of your RAM. vDIMM value options can range from 1.7v up to 2.6v or more. vDIMM is not shown in CPU-Z.

RAM Ratio
This has probably caused the most confusion jumping from s939 with DDR to AM2 with DDR2 RAM. It has many names and many numbers associated with it, though they all do the same thing in the same way. The first thing you often have to look for is the memory configuration page. Some labels for this are Memory Mode, Memclock Mode, DRAM Timing, DRAM Config, CPU Config, etc. What you're looking for usually has a "Manual" or "Limit" option that you need to set to show more detailed RAM options. Once you find these you look for DRAM ratio, RAM Divider, Memclock Value, Memory Limit, or any combination of these words. The options are one of three sets of numbers usually containing: (200, 266, 333, 400), OR (400, 533, 667, 800), OR (1:1, 4:3, 5:3, 2:1). Sometimes there are other numbers in addition to those listed. CPU-Z labels this as "FSB : DRAM" and has a value showing CPU/x.

HyperTransport Link
This is another overly-labeled component of the AM2 platform having seen names like HTT, HT Bus, I/O Bus, HT Link, etc. HT Link is the common usage standing for HyperTransport Link and is used by CPU-Z as well. Note this is not the same as HyperThreading on Intel platforms! The HT Link is very close to the "FSB" of Intel platforms, handling all of the system I/O to PCIe, PCI, and USB buses. The CPU to RAM access does not go through the HT Link so changing it's speed does not effect RAM performance. The HT Link multiplier has many labels, K8 <> NB, CPU <> NB, FSB, and HTT are but a few. Common HT Link options are (200, 400, 600, 800, 1000) OR (2X, 3X, 4X, 5X).


That's the basic over-clocking variables for an AM2 rig. Most of these are very simple but sometimes hard to find. If your motherboard manual and the pointers above aren't enough to find what you're looking for just post in the AMD Motherboard Socket AM2 section and I'm sure someone will help. There are other things to adjust and play with but let's stick with those listed for now. Take some time to find all these values and variables in the BIOS and ask questions if you're unsure.


Crawling before Walking
Don't take the title personally. Everybody running an OC'ed rig has been a n00b and learned somewhere, either from hours of experimentation or from others - usually on forums like (ad plug :D) OCForums.com!
Just open your mind a little, be prepared for some pointing & clicking, learn how to use the arrow keys in BIOS, and this will be easier than you think. :cool:

Programs
Over-clocking is about finding the limits of the hardware. The best method known for this is the scientific method and it's just like a high school science lab. First there are several "benchmark" programs (BMs) out there including PCMark and SuperPi for CPUs, 3DMark for video, and Everest for RAM. By running some or all of these BMs before overclocking we can document and study the differences between system default (stock) and overclocked settings, which will be valuable data later when "tweaking" the system for best performance. There are also a few other programs that are nice to have for OC'ing and determining system stability:
CPU-Z, as already mentioned, to keep track of actual values as BIOS settings are changed
MemTest86 is used to determine RAM stability and is included in some BIOS'
Core Temp is a great little program used to keep track of the CPU die temps. AM2 CPUs have some variance in the die sensor and usually are not the same between two cores of a dual-core CPU. The sensors can also be off as "real" temperature readings but are very good for calculating differences in temps on the same core.
Prime95 or Orthos is used to "load" the CPU (make it work as hard as possible). If Prime95 is used with a dual-core CPU make sure you have version 25.1 or newer to support dual-core CPUs OR make two folders and run them both at the same time setting affinity for the different cores. Prime95/Orthos (P95) creates high CPU temperatures and, if pushed far enough, will cause a shut-down of the system if temps get too high. Make sure you have the CPU temperature shut-down in the BIOS set ON to avoid any CPU damage! And pick a low number like 70°C (~160°F) - most AM2 CPUs shouldn't normally run above 65°C.
ClockGen or a board manufacturer's program to set system hardware variables in Windows/Linux. While this is not required for OC'ing it can sure save some reboots and time while doing it!

Having decided on BM programs go ahead and run them while the system is still stock. Note the CPU-Z values as well. Write down the settings and record the BM scores under them, you'll want them later to see how much you've accomplished!

Manual Entry and BIOS
After running the stock BMs and recording the scores reboot the computer and enter BIOS. Find the above variables, set them all to manual entry, and enter the system default (stock) values for each. This is done because the BIOS has a very nasty habit of adjusting values on reboot if they are left on Auto or Default. Just like a lab experiment the idea is to change one variable at a time and Auto values can mess that up. The default values for common system components are:
System Clock: 200 MHz
CPU Multiplier: CPU rating divided by 200 (i.e., a 2400 MHz CPU will be 2400/200 = 12)
vCore: 1.35v
RAM ratio: 400 OR 800 OR 2:1 (depending on the BIOS list values)
vDIMM: 1.8v (but may be as high as 2.1v)*
RAM timings: This is the stock timings of the RAM as noted in it's specifications (4-4-4-12, 5-5-5-15, etc.)*
*RAM timings and vDIMM are often printed on the side of the RAM stick and can always be found on the manufacturer's website
HT Link: 1000 MHz OR 5X (depending on the BIOS list values)

While in the BIOS also turn off CoolnQuiet (CnQ) and any CPU or system fan regulators. CnQ is found in a number of places in BIOS and allows the BIOS to change the CPU multiplier and vCore depending on the CPU load. It might be good for future use after over-clocking but not during. CPU and system fan regulators aren't a bad thing either, but it's better to keep the components as cool as possible right now so turn them off and turn them on afterwords if desired.

Final Stock Systems Check
Reboot the system just to make sure everything is running as it should be after the manual entry. Most motherboards have a drop in vCore between BIOS and idling (some worse than others) so run CPU-Z and note the vCore (Core Voltage) shown. Leave CPU-Z up and running, start Core Temp, then record the CPU idle temp(s). Leave CPU-Z and Core Temp open and start P95 - the CPU temp(s) should increase almost immediately. Wait a couple of seconds for the temps to stabilize a little and write down the load temps under the idle temps just recorded. (If the CPU is water cooled temps may take several minutes to stabilize.) Also, record the load vCore reading by CPU-Z. Compare the idle vCore to the load vCore - the difference between them is known as vDroop and is quite common on all systems (even Intel). A vDroop of 0.03v or less is fine, 0.03-0.05v is average. Over 0.05v is not so good but if it's steady it's workable. The real problems come from a load and/or idle vCore that jumps around a lot - greater than 0.05v up and down as you watch it with CPU-Z is NOT good and may effect your OC. Motherboard power chips (MOSFETs) and under-rated power supplies are often the culprits for this behavior.


The First Steps
The factory system and settings are still at stock values but now there's a little control over them. BIOS can't change the important values on it's own anymore and data is recorded about the stock configuration. It's time to move on in the OC experiment and start checking out individual hardware components one by one. While this is easy enough it is time consuming and other component values need changing to make sure they can't interfere with the component being tested. Since the clock is the focal point of the system experimenting starts there ...
 
Last edited:
The System Clock

The System Clock
The system clock is part of what governs the speed of the HT Link, CPU, and (indirectly) RAM:
HT Link = clock x (HT Link multiplier)
CPU speed = clock x (CPU multiplier)
RAM speed = CPU speed / RAM ratio

If the clock is increased these other components will also speed up. While this is critical for overclocking, it also imposes a problem for testing. Our goal here is to find the highest stable setting (the "top end") of the system clock, but we need to keep the other components at or below their stock speeds during the test. If other components are also overclocked then how do we know the system clock has failed and not something else? To avoid this complication make these changes in BIOS and reboot:

System Clock: 210 MHz
CPU Multiplier: 6X
vCore: no change
RAM ratio: 200 OR 400 OR 1:2
vDIMM: no change
RAM timings: no change
HT Link: 3X

Notice the system clock has been overclocked to 210 MHz? This is the first step in our testing. Run P95 or SuperPi for a couple of minutes and, if it runs OK, reboot and enter BIOS. (If you have and can use ClockGen or some other program to change the clock speed in Windows then use it instead of the reboot to change the clock speed.)

Increase the clock speed by 10 MHz to 220 and boot to Windows. Run P95 or SuperPi for a couple of minutes and, if it runs OK, reboot and enter BIOS.

Repeat this same procedure over and over, each time increasing the clock speed by 10 MHz. (You can use a smaller interval like 5 MHz if you wish.) If you're using ClockGen I recommend rebooting at 250 MHz to change the clock speed in BIOS, then continue using ClockGen.

If you make it to 300 MHz reboot to BIOS and change the CPU multiplier to it's lowest setting. Calculate what clock speed is needed at the lowest multiplier setting to reach the stock speed of the CPU. (For example, if the lowest setting is 6X and the stock speed of the CPU is 2.0 GHz (2000 MHz) then 2000 MHz /6 = 333 MHz.) If the clock setting is pushed higher than this calculated value then the CPU is also overclocked! Not good. When the clock speed test reaches the CPU stock speed and is still stable, stop the test!

At some point in the testing P95 or SuperPi will give you an error OR the computer will fail to boot, lock-up, "blue screen", whatever. This means the current test has failed. Record the last stable clock speed and reboot to BIOS.

If your BIOS has a setting for North Bridge (NB) chipset voltage increase it by one step, boot to Windows, and continue testing. When the system becomes unstable again record the highest stable clock speed and reboot to BIOS.



This experiment has determined the top end (highest stable speed) of the system clock and will be important as we continue testing. It's time to move on to the next test, overclocking the CPU.
 
Last edited:
The CPU

CPU Speed
CPU speed is what most of us think of when talking about OC'ing and rightly so. CPU speed is the single, most important piece of performance gain. CPU speed is derived from the clock speed times the CPU multiplier. Again, we need to find the top end of the component but this time we have some extra data, we know how fast the clock will run and still be stable. However, we still have to make sure other non-tested components won't interfere. Generally K8 CPUs will run up to 1.45 vCore without a problem but this is your choice - AMD recommends 1.30-1.35v. This test can be run at 1.35 vCore but you may not see much of an OC with it. Set these values in BIOS and reboot to start CPU testing:

System Clock: 210 MHz
CPU Multiplier: Highest possible (or "stock" for FX/BE chips)
vCore: 1.40v*
RAM ratio: 200 OR 400 OR 1:2
vDIMM: no change
RAM timings: no change
HT Link: 3X
NB Voltage: normal (but increase if the clock speed is high enough later in the test)

Hopefully the system booted fine and Windows is running with the CPU overclocked by 5%! Run P95 (don't use SuperPi for this test) for a couple of minutes and, if it runs OK, reboot and enter BIOS. (Sounds familiar doesn't it?) Since we're testing the CPU and every 1 MHz increase in the clock increases CPU speed by the CPU multiplier we'll use smaller increments for this test.

Increase the clock speed by 5 MHz and boot to Windows. Run P95 for a couple of minutes and, if it runs OK, reboot and enter BIOS.

Repeat this same procedure over and over, each time increasing the clock speed by 5 MHz. At some point in the testing P95 will give you an error OR the computer will fail to boot, lock-up, "blue screen", whatever. This means the current test has failed. Record the last stable clock speed, note "1.40v" next to it, and reboot to BIOS. Set the clock back 5 MHz to it's last stable speed, boot to Windows, and start P95.

CPU Temperature, Speed, and Voltage
Before going farther let's take a look at the CPU temps. When the CPU speed increases the temps increase a little too. In addition to speed vCore also increases CPU temps and the next step will increase the vCore another 0.05v. AM2 chips are usually very good about keeping low temps at 1.40 vCore but it's still good to check since a lot of other computer components and the environment influence this. Let P95 run a few minutes to stabilize the temps then run Core Temp. If the CPU temps are 55°C or less that's good - go to the next step!
>> If CPU temps are above 55°C this test is over. To proceed farther will require better cooling.

Stepping It Up
Increase the vCore to 1.45v and set the clock to the last stable clock speed plus 5 MHz (where it erred before). Run P95 for a couple of minutes then run Core Temp while P95 is still running. If it runs OK and CPU temps are below 55°C, reboot and enter BIOS.

Increase the clock speed by 5 MHz and boot to Windows. Run P95 for a couple of minutes and check temps. If it runs OK and CPU temps are below 55°C, reboot and enter BIOS.

Continue this cycle as before until the test fails again. Record the last stable clock speed, note "1.45v" next to it, and reboot to BIOS. Set the clock back by 5 MHz (to the last stable clock speed) and boot to Windows. Run P95 and leave it running. Calculate the top end of the CPU at 1.40v and 1.45v by multiplying the last stable clock speed for each vCore by the CPU multiplier and note "1.40v" or "1.45v" next to it.

It's time to check the CPU temps again. Run Core Temp and record the temps. If you're running 55°C or less then everything is fine, you've found the top end of the CPU. Record the clock speed and temps.
>> If temps are over 55°C then reboot, lower the clock by 5 MHz, run P95 for a few minutes and check temps again. Keep repeating this until the CPU temp is 55°C or less. Record the clock speed and temps. Compare the clock speed at 1.45v to the last stable clock speed at 1.40v. While rare, sometimes the 1.40 vCore setting is higher - again, this depends a lot on cooling. Later on, use the best setting found for the CPU speed test.

Final Notes
This is as far as this test goes. Some CPUs will hit a "heat barrier" before hitting the top end. If this has happened in your test then find some way to increase CPU cooling or decrease the heat around the CPU. Video card exhaust, case flow, and ambient temps (room temperature) all play a role in increasing CPU temps.

There are many people (including me) who will run the vCore higher than this but anything above 1.45v has to take many other things into account such as stability of the PSU, cooling capacity of the HS, case cooling, etc. If you've already hit a heat barrier don't even attempt it! For CPUs that are still running cool, feel free to experiment above the 1.45v limit but I suggest many more hours of quality time with your rig and BIOS before attempting it. Also, as vCore is raised above this line CPU life expectancy lowers - just keep that in mind. Again, this is your decision and responsibility.
 
Last edited:
RAM

RAM Speed and vDIMM
RAM, though very important for performance, is a distant second to CPU speed. Nonetheless, every MHz of RAM is an increase in performance so let's make the most of it. If you have four sticks of RAM remove two of them leaving the best matched set in the board. Set these values in BIOS and reboot to start testing RAM:

System Clock: 205 MHz
CPU Multiplier: stock setting
vCore: 1.40v
RAM ratio: 400 OR 800 OR 2:1 (set RAM rated less than DDR2-800 to it's stock speed)
vDIMM: stock + 0.1v (not to exceed 2.2v)
RAM timings: stock
HT Link: 3X
NB Voltage: normal

Just like the system clock and CPU, this test simply steps up the clock in 5 MHz increments then checks for stability. If you have a CPU multiplier that's an odd number the RAM will be running a bit under 400 (as shown in CPU-Z). For this test the CPU limits are known so when the system gets to those CPU top end settings you need to reset the BIOS for the next step found in the CPU test. Run P95 for a couple of minutes and, if it runs OK, reboot and enter BIOS.

Increase the clock 5 MHz and reboot to Windows. Run P95 for a couple of minutes and, if it runs OK, reboot and enter BIOS.

Keep repeating this until the test fails OR you reach the CPU top end for 1.40 vCore. If the CPU reaches top end, set the vCore up to 1.45v (if this passed before) and continue increasing the clock 5 MHz at a time.

Repeat this same procedure over and over, each time increasing the clock speed by 5 MHz and adjusting CPU settings as needed to keep it stable. At some point in the testing P95 will give you an error OR the computer will fail to boot, lock-up, "blue screen", whatever. This means the current test has failed.

Record the clock speed and vDIMM then reboot to BIOS and increase the vDIMM +0.1v (not to exceed 2.2v). Reboot to Windows and run P95 for a couple of minutes. If this passes reboot and enter BIOS.

Increase the clock 5 MHz and reboot to Windows. Run P95 for a couple of minutes and, if it runs OK, reboot and enter BIOS. Repeat this until the test fails again then record the clock speed and vDIMM.

At this point a lot of RAM will take more vDIMM but not all will. Because this is a general OC thread part one of testing RAM stops here. Feel free to experiment farther on your own and at your own risk! A lot of pushing the RAM farther has to do with cooling and quality of RAM. Better RAM will take a lot of abuse, cheap RAM will fry.

Before going on to the second part of RAM testing I suggest you search OCForums for RAM similar to yours. If others are already OC'ing it then you'll have a better idea what timings will work and how far the RAM will go with those other timings. Also search the Net, many reviewers OC their RAM as part of the review and you may get some good pointers there as well. You're looking for stable timings, vDIMM limits, and speeds for those settings.



RAM Timings
RAM is a collection of memory chips all put together on a mini-board along with controller and communications chips. RAM timings (correctly called latencies) are the cycle delays required for certain RAM operations to "clear" so the next operation can start. If the timings aren't set properly then data in the RAM module starts over-running itself and locks up. If more time is allowed for operations then data won't overrun itself but the RAM runs slower internally (more lag time). RAM timings are "tight" when the timings are set as low as possible without the RAM freezing. The first test pushed the stock timings to this upper limit so the timings were as tight as possible for that speed, resulting in the highest bandwidth. Bandwidth is the same for RAM as for networks, the greater the bandwidth the more data can be sent back and forth across the connection. By loosening the timings, allowing more cycles per operation, we can increase the speed of the operations - in this case the RAM speed.

Finding the right "step" for looser timings is sometimes difficult since there's no hard and fast rule for it. Generally, the next step from stock is +1, +1, +1, +3 to the timings but sometimes the last number is +6 instead. I like to try different timings near the top end of the stock settings so let's start there, but you'll have to experiment with these numbers on your own since all RAM is different. Set these values in BIOS and (hopefully) boot to windows:

System Clock: stock RAM timing top end less 5 MHz
CPU Multiplier: stock setting
vCore: 1.40v
RAM ratio: 400 OR 800 OR 2:1 (set RAM rated less than DDR2-800 to it's stock speed)
vDIMM: stock + 0.2v (not to exceed 2.2v)
RAM timings: stock+1, stock+1, stock+1, stock+3
HT Link: 3X
NB Voltage: normal


If your machine wouldn't boot get back into BIOS and try stock+6 on tRAS (the last timing number). Once you're running windows we follow the same procedure as above when checking the top speed for stock timings. Run P95 for a couple of minutes and, if it runs OK, reboot and enter BIOS.

Increase the clock 5 MHz and reboot to Windows. Run P95 for a couple of minutes and, if it runs OK, reboot and enter BIOS.

Keep repeating this until the test fails OR you reach the CPU top end for 1.40 vCore. If the CPU reaches top end, set the vCore up to 1.45v (if this passed before) and continue increasing the clock 5 MHz at a time. At some point the test will fail indicating you've reached the top end for these timings. If you reached the CPU top end first then the test is over. Either way, record the all CPU settings, RAM timings, and vDIMM.

If you started with low latency RAM (CAS 3) then you can loosen the timings a second time to CAS 5 and repeat the test. You'll have to use your own research and experiments to find the correct RAM timings.



To go any farther toward a good overclock requires a side-trip into RAM ratios, RAM speed, and how they relate to the CPU settings. If you reached the top end of the CPU before finding the top end of RAM then there's further testing to be done. If you found the top end of RAM there's still some tweaking which will require knowledge of this relationship.
 
Last edited:
RAM Ratios

The CPU and RAM Speed
First off, RAM speed is directly related to the CPU and it's speed settings. There is no direct relation between the system clock and RAM speed. For those of you coming from the pre-A64/K8 AMD or any Intel system this is a major change and the reason K8's have such great RAM bandwidth and low latency. The RAM bus is connected directly to the CPU. Remember, A64/K8 systems have no north bridge - just a RAM bus and the HT Link.


RAM Ratios
The RAM ratio represents the speed at which the RAM clock ticks off cycles in relation to the CPU speed. With DDR it was a simple divide function to find the RAM clock result. With DDR2 and an odd-numbered CPU multiplier things got a lot more complicated. If you have a CPU with a stock multiplier that's an odd number (9, 11, 13, etc.) you may have already noticed that the RAM doesn't run at 400/800 when the system clock is at 200. No, your board's not broken! This is normal behavior and when you work through the processes below you'll see why.

Determining RAM speed
For this discussion RAM "ratio" will be used for the BIOS setting instead of the often used RAM "divider". The RAM divider is really the number we get near the end of the process to determine RAM speed. The ratio values will also be used, "2:1" instead of "400" OR "800" for the "stock" RAM ratio. Since many BIOS' use 400/800 numbers here is a quick list of the common ratios and their associated values (many boards don't have all these):

1:1 200 400 = 1.0
4:3 266 533 = 1.3333333333333
5:3 333 667 = 1.6666666666667
2:1 400 800 = 2.0
8:3 533 1066 = 2.6666666666666
10:3 667 1333 = 3.3333333333333

NOTE (K10 only): If you are over-clocking a K10 CPU (Phenom and newer) the only thing that matters will be the RAM ratio and clock speed. Use the ratio shown and multiply it by your current clock speed to get the RAM speed in MHz (which is half the DDR2 rating).

The process for determining RAM speed is pretty simple - just take it one step at a time:

1. Take the CPU multiplier and divide it by the RAM ratio.
The CPU multiplier is the CPU multiplier being used now, which is not always the same as the stock multiplier.
The RAM ratio is the ratio listed, not the BIOS value of 400/800.

CPU multiplier = 11; RAM ratio = 2:1 (= 2.0); so, 11 / 2 = 5.5
CPU multiplier = 12; RAM ratio = 5:3 (= 1.6666666666667); so, 12 / 1.6666666666667 = 7.2

2. Take the result from #1 and round up to the nearest whole number.

5.5 -->[/I] 6
7.2 -->[/I] 8

This result is the real RAM divider and why RAM ratio was used earlier. You should find this same number in CPU-Z next to the "FSB : DRAM" label on the memory tab.
<<Side note: The little "trick" of the second part of this step is called the "ceiling" function, which is just a fancy way of saying "round up to the next whole number". The reason this is done is pretty simple, the internal clocks need a whole number to use for the "divider" so a simple fraction is created (1/6 or 1/8). The reason some BIOS' use "limit" for the RAM ratio label is because this is the fastest we want the RAM to run. If we round down instead of up the RAM would exceed the limit instead of staying below it.>>

3. Now it's a simple process to calculate RAM speed. Take the CPU speed and divide by the RAM divider.

CPU speed = 2200; 2200 / 6 = 366.67 (NOT 400!)
CPU speed = 2880; 2880 / 8 = 360

This is the RAM speed or in DDR2 "ratings" DDR2-733 and DDR2-720.

Let's take one more example and work it straight through:
CPU speed = 2730 (260 x 10.5)
RAM ratio = 5:3 (= 1.6666666666667)

2730 / ceiling ((10.5 / 1.6666666666667) = 390 (DDR2-780)

It should be obvious, as shown in the first example, why odd CPU multipliers do not yield DDR2-800 at the stock 200 MHz clock setting. This also applies to any half-multiplier settings as well, as shown in example three.

As you can see, RAM speed really is dependent on the CPU multiplier, CPU speed, and RAM ratio - not the system clock. Notice the system clock is nowhere in these equations, though it is indirectly represented by the CPU speed, which is "system clock x CPU multiplier". This concept will be critical in the next step.


RAM Speed Test Pt. III ...?
If your CPU reached top end in the previous RAM speed tests before the RAM gave out there is a way to keep increasing RAM speed to find the top end without going beyond the CPU top end. You'll either have to calculate or do some trial and error here, however. Set these values in BIOS and boot to windows:

System Clock: last clock speed used for RAM testing less 30 MHz
CPU Multiplier: stock setting - 1
vCore: 1.40v
RAM ratio: 400 OR 800 OR 2:1
vDIMM: stock + 0.2v (not to exceed 2.2v)
RAM timings: Last settings being tested
HT Link: 3X
NB Voltage: normal

Run CPU-Z and note what the RAM speed is now as the settings above will vary from system to system. Calculate the difference between the last RAM speed tested and the current RAM speed. Reboot to BIOS and adjust the clock by the difference in RAM speeds (last - now) and boot to windows. Run CPU-Z to confirm the RAM speed. If it's correct run P95 for a couple of minutes and, if it runs OK, reboot and enter BIOS.

Increase the clock by 5 MHz and continue testing as before. Check stability with P95 then raise the clock 5 MHz over and over until the test fails. Now you've found the top end speed of your RAM at these timings. Note the speed, timings, and vDIMM.



Now that you understand a little of RAM dividers and how RAM speed is calculated let's use all this information you've been gathering about your system and try for a good overclock.
 
Last edited:
Overclock and Small Tweaks

The Stable Overclock, Putting It All Together
For starters look at the numbers you currently have for CPU and RAM top end. Remember, the CPU speed is the most critical number of all. The idea is to get the highest RAM speed we can without having to down-clock the CPU from it's top end. For those having gone through RAM Speed Test Pt. III you're almost there. The system clock is the common denominator to look at so take the CPU top end speed and make a list of the possible clock x CPU multipliers that will work. Let's take an example with a CPU top end of 3159 MHz on a 5000+ chip (top multiplier of 13):
243 x13
263 x12
287 x11
316 x10


Let's say the top RAM speed is 520 (DDR-1040) with loose timings. To get a rough idea of what to look for take the CPU speed / RAM speed = RAM divider desired, so 3159/520 = 6.075 so the "6" divider is a little fast but not much. Let's check that:

243 x13 :: 2:1 = (13/2) = 7 divider (a little low at 3159/7 = 451 MHz)
263 x12 :: 2:1 = (12/2) = 6 divider (a hair too high at 3156/6 = 526 MHz)
287 x11 :: 2:1 = (11/2) = 6 divider
.............. 5:3 = (11x3/5) = 7 divider
316 x10 :: 2:1 = (10/2) = 5 divider (way too fast at 3160/5 = 632 MHz)
.............. 5:3 = (10x3/5) = 6 divider

So we have a choice, go with a 7 divider (243x13 2:1) and take a 75 MHZ hit on the RAM speed OR stay with the 6 divider but lower the CPU down to 3120 MHz (260x12 2:1) and take a 39 MHz hit on the CPU speed. In this situation, with the small CPU speed change, I would benchmark both settings to see which works better for your common applications. Anything lower on the CPU would generally be worse but 39 MHz of CPU in exchange for 75 MHz of RAM (DDR2 +150) could balance pretty well. The only way to know is to test it out using benchmarks or timed tests on your favorite (and most often used) programs.


This example is a general scenario for determining the basic OC for your system. As you can see, it wasn't very hard to come up with a logical starting point since we've already collected all those important numbers. But this is just a starting point, tweaking the advanced RAM settings might buy you a few more clock ticks, which would bring your CPU a little closer to it's top speed. I won't go into RAM tweaks, that takes up whole threads in itself, but we can explore a couple of things not yet covered.


HyperTransport Link and More
The HT Link is the least important OC piece, which is why it's the last thing to play with. All the HDD data runs through this link but even a pair of SATA II's won't over-tax it at 800 MHz and if it's running too fast you can get data corruption on the HDDs. In the example above there are two possible outcomes for the clock, 243 and 260. Since the HT Link should be kept near 1000 MHz the 243 clock speed would work fine with an HT Link multiplier of 4X, yielding a speed of (243x4 =) 972 MHz. 260x4 is a little fast at 1040 MHz but most mid- to high-range boards will run that easily. Older, cheap boards may have problems with 1040 which may not be apparent until some important file is corrupted. :( Generally, anything over 1100 MHz is too high - even for an excellent board and quite honestly you'll never see a difference from 840 to 1120 MHz HT Link unless you're running a file server or burning an awful lot of CD/DVD's all at once! If you're running a server I just wouldn't risk the extra speed for the possible loss of data. For burners, let's say I'd have to see some really big performance difference to exceed this speed limit!

RAM Tweaks
As mentioned earlier, advanced RAM settings can be tweaked to get a couple of extra MHz out of the clock. If your board has a skew setting you might play with that - I managed 2 MHz more from my s939 system when I found the right clock skew. Other RAM tweaks can be found in the memory section.

Cooling
Cooling is another issue to look at now that you have a (semi) final OC. If the RAM, MOSFETs, or NB is hot then some active cooling (especially on RAM and MOSFETs) can give you another few clock MHz sometimes. A better CPU cooler will generally give you a better OC but not much more than 100 MHz or so, depending on the existing HS (heat sink) and the new cooling method. Stock air to good WC (water cooling) often yields ~150 MHz. Case cooling can be critical as well, especially if you're running on air. The CPU temp is directly proportional to the air being used by the CPU HS so if your case is 10° hotter inside than outside your CPU will run ~10° hotter until you supply it with some cooler, fresh air.

Stability
After you've got what appears to be a good OC you really should check it for stability. I like to be sure my machine will run problem free for months on end once I've set it up! ;) To ensure that performance, running stress tests and benchmarks to really push the hardware is the last critical step. How long and what programs to run depends on how you'll be using your machine.

>> If it's used mostly for games or editing video then test the machine using P95 for 6-8 hours and 3DMark for another 6-8 hours. Then test running both at once for an hour or so.
>> If it's your main file server then P95 for 24 hours along with some HDD tests would be in order to be sure your components stay cool for 24 hours - hot HDDs can cause data corruption. Checking video isn't so important for a file server but you still might want to run 3DMark for an hour or so (at the same time!) just to make sure your case cooling doesn't pass the heat load over to the RAM, CPU, or HDDs.
>> If you Fold or Crunch then run P95 for a couple of hours then stop P95 and start Folding/Crunching. Nothing is a better test of machine stability than Folding and Crunching! Again, run 3DMark a couple of hours to make sure it won't interfere with your other components. If you GPU Fold set that to running - it's a sufficient test in my book.
>> For other applications the 24 hour P95 test is the best test out there. Throwing in 3DMark for an hour or so is still a good idea to test case cooling when everything is running flat out. For a poor case set-up video heat can swamp the SB easily, the NB sometimes, and maybe even the CPU, MOSFETs, or HDDs.


RECAP & Cheat Sheet
In my opinion the most efficient method for getting a good OC is testing each component of the machine one by one. Having done that, a few quick and easy calculations will point to the correct settings for CPU ad RAM to make the most of the hardware. But this method does something more than just provide a path for simple OC'ing - it also gets one familiar with the settings in BIOS that determine component speeds as well as pointing out the interdependencies between those components.

This is a system, a whole machine - the system clock and multipliers/ratios affect everything else in the system in some way. By "playing" with all these settings a person quickly learns what does and does not work. If you've had to reset CMOS several times, well that's part of the game. If you didn't have to reset CMOS at least once then you're very lucky or not pushing hard enough! ;)

Whatever it is that brought you to OC'ing, whether it's pushing an older system a little harder to stretch it's lifespan or just plain curiosity I hope this has helped. Please feel free to start threads in any section of the Forum if you have questions. We've ALL been new to this at some point or other and I think you'll find a lot of members more than willing to help you work through issues and explain things to you, even if they seem minor. But please don't abuse the friendliness shown by others, Search is still MY best friend and maybe it can be your friend too! :D


Good luck in your efforts - I'll be looking for your system posts!:):thup:
And here's a little something to print out ... Cheat Sheet
 
Last edited:
1st in great thread potential!!

In the future I believe the addition of Quad overclocking would be great!
 
1st in great thread potential!!

In the future I believe the addition of Quad overclocking would be great!
OC'ing a quad on the AM2 is pretty straight-forward and most of what will be in here should cover it with some very minor notes added.

OC'ing a quad on the AM2+ platform would require a new thread including DDR3, ganged/unganged modes, multiple CPU multipliers, and much, much more ...! :eek:
 
Nice work. This would have helped me out about 4 months ago. LOL. Keep up the good work.
 
Looking good and looks like the beginning of a sticky. Nice work so far. Can't wait to see what you do with the rest!

Anything in there that can make my Phenom get off its rear and decide to be a good overclocker?? ;)
 
Good one QI, will be watching. Note, as for those ratios in RAM, they are useless without a FSB but you will want to leave them in for reference. On a 14X 5000BE thats 2800 where most will be with ease once they have a 5000BE up and running no sweat. 2800 / 400 = 7 doubled to 800, which to me says the ram is running 1 : 7. This is only relevant with the K8s as they still have the NB running at CPU speed where the K10s run the NB on a different multi.
 
Last edited:
Good one QI, will be watching. Note, as for those ratios in RAM, they are useless without a FSB but you will want to leave them in for reference. On a 14X 5000BE thats 2800 where most will be with ease once they have a 5000BE up and running no sweat. 2800 / 400 = 7 doubled to 800, which to me says the ram is running 1 : 7. This is only relevant with the K8s as they still have the NB running at CPU speed where the K10s run the NB on a different multi.
I'll get into the irregularities of RAM when I do that post. I know it's really odd for K8 DDR2 - I think that's the main reason there are so many questions about RAM on AM2.

Just keep in mind, the numbers in those lists are the BIOS options, not what's really happening ... ;)
 
Hello,

First I would like to commend you on a superb newbie guide! Even I can understand it, and that is a good sign! :D

However, I have a question before I proceed. In my BIOS I have no option that I can see for the HT Link (or anything similar). Obviously my card isn't the typical overclocker card (I should've read before I bought it, what's done is done I guess). The BIOS is simply called the 'CMOS Setup Utility' from American Megatrends, v.02.59. Am I missing something, does this BIOS actually have this option? There aren't that many places to look in there. Is it possible to upgrade to a different BIOS with this card?

Apparently the '1A', which really leaves me blank as the only version number I could spot was '02.59'. The readme also includes a warning not to flash the BIOS if the system is working and I don't know what I'm doing, both of which would apply, really.
 
Last edited:
I don't think I've ever heard of a BIOS that doesn't have an adjustable HT Link. What board did you say that was ...?

MSI K9N Platinum, sorry. I was positive I put my specs in the signature yesterday. :confused: edit: There they are. Thought they'd go into older posts too.

In any case, it might be there and simply be labelled with a silly name, or maybe I'm just bad at searching, but I couldn't find anything with a reasonable name or adjustable number in there that might be the HT Link.
 
Looks like there's a page under advanced settings for HyperTransport MCP55 Configuration ...

Oh, yeah, that's right - I overlooked it because it looked so strange.

In your guide you say that there are two types of HTL settings, 600/800/1000/etc MHz and 2x/3x/4x/5x - This setting only gives me two options that don't seem anything like these two at all: '8' and '16'. It says 8 and then an arrow down, another 8 and then an arrow up, apparently in/out values or something. It also has an apparent 'auto' setting but the choices aren't 'auto/manual', they are 'enabled/unenabled' which makes me a bit wary to change it..

Do you have any suggestions?
 
Oh, yeah, that's right - I overlooked it because it looked so strange.

In your guide you say that there are two types of HTL settings, 600/800/1000/etc MHz and 2x/3x/4x/5x - This setting only gives me two options that don't seem anything like these two at all: '8' and '16'. It says 8 and then an arrow down, another 8 and then an arrow up, apparently in/out values or something. It also has an apparent 'auto' setting but the choices aren't 'auto/manual', they are 'enabled/unenabled' which makes me a bit wary to change it..

Do you have any suggestions?
That setting would be the bus width (8 byte = 64-bit, 16 byte = 128-bit) but I believe if you look farther you'll see a "1000 MHz" value on that page as well.

Like many manual settings in BIOS you may have to turn an "Auto" setting to "Manual" to get more options ... ;)
 
Back