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

Easy as 1, 2, 3, Overclocking the A64, Opteron, X2

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

infinitevalence

Member
Joined
Nov 14, 2002
Location
Nashville
***bios descriptions are from the DFI Ultra-D and SLI-DR, yours may differ some but the principal will be the same***

!!!!! This is not a definitive guide to overclocking, it is MY quick and dirty way to get a fast overclock. This guide will not maximize your performance, it will only enhance it. If you want to find the maximum performance read some of the stickies that help you find the highest overclock for each of the 3 steps described below. As usual I and the owners of the forums take no responsibility if you damage your stuff. You break it you bought it. Use your brain when overclocking keep an eye on temperatures and don't use unreasonable voltages, you can damage your equipment if you don't. This guide is a living document I will update it as corrections are made and new techniques are adapted. Please do not post flames, keep it to constructive information or questions regarding this guide. !!!!!

1. Test the limits of your HyperTransport bus (HTT).

To do this set your CPU to its lowest multiplier. Generally 6x is good enough for this test.

Now set your memory to its lowest divider, 133 is general recommend.

Begin your testing by raising the FSB with the FSB/HTT multiplier set to 5x. Push your HTT up in 5MHz increments until it becomes unstable, generally you back off 5MHz from the last unstable speed. As Quailane pointed out its also a good idea to test with the 3x multi to ensure that your motherboard is not FSB limited, using the same steps test at both 5x for max HTT and 3x for max FSB. Testing should then be done, boot to windows and run a few loops of prime95 or superpi 32m, if the system does not create any errors then you know its OK to move on to step 2. If it does fail out then reboot and back off the FSB another 5mhz, and retest.

If your motherboard is supported by clockgen you can speed up this process using the following method.

Set your CPU to a low multiplier (6x), memory to a low divider (133) and your HTT to 5x. Boot into windows and start up clockgen. While running either prime95, superpi 32m or some other stress testing program increment the FSB by 5MHz. Let the stress testing program run for around 5 minutes or so and then increase the FSB by another 5MHz. Repeat this process with your HTT multiplier set to 3x and record both numbers.

Once your done note your max HTT number, a pad of paper is your friend so write it down if you think you might forget. You should have two numbers after doing this testing one is the maximum HTT value i.e. 1234MHz and the other will be your maximum FSB value i.e. 318MHz. HTT is the product of the HyperTransport multiplier and the Front Side Bus so if you get stuck at 220MHz Front Side Bus with a 5x HTT multiplier don’t freak out it just means that your motherboard cannot do faster than 220x5, when you lower it to 3x you may be able to get much higher.


2. Test the limits of your memory.


This time around we are going to rest the FSB to 200, set the FSB/HTT multiplier to 3x and leave the CPU at 6x. Go into the memory section of the bios and set the memory speed to 200mhz (1:1) this means that for every MHz you increase the FSB you increase the speed of the memory. Again leave the CPU at 6x so it does not become the bottleneck.

While in the memory section of the bios you should set it to your preferred timings, I'm not going to go into detail on how to overclock your memory. I generally set most things to auto for this part of the testing other than the main four timings (Tcl, Trcd, Tras, Trp).

Again we start testing by incrementing the FSB by 5MHz at a time. You can safely start off at the rated speed of your memory and rated timings if you want (ie. PC3500 at 215fsb). Use Memtest86+ for testing, I generally use 5 passes of test #5 and 5 passes of test #8 for a quick test, this does not mean its stable but it does mean its probably safe to continue on. If you start getting errors in either test you can either lower your timings or increase your voltage. TCCD based memory does better when you loosen timings, BH5/UTT does better when you increase voltage.

Always keep an eye on the bandwidth in the upper left hand corner of the screen this tells you if you improving or hurting your performance with the timings you have set. If for example you can hit 240fsb with 2-2-2-6 or 250fsb with 2.5-3-3-8 you may find that 240fsb gives you more bandwidth and is there for more desirable, so pay attention to this value and WRITE IT DOWN along with your timings that way you can find the best performance.

When you think you have found the limits of your memory let Memtest86+ run through at least 10 full passes of all tests. Again this does not mean its stable only that its most likely good enough to move on to step 3.

There is no windows based memory test that is as good or accurate as Memtest86+ so don’t even bother trying to cut any corners on this one.

3. How high can your CPU go. UPDATE for X2

This time around you want to set you CPU to its stock multiplier. Drop your memory back down to 133, and keep your FSB/HTT multiplier set to 3x. Reset your FSB to 200MHz and then your ready to test.

Using 5MHz increments start to increase your FSB. Test using prime95 or SuperPi 32m, if it passes then repeat until it fails out, you can then either back off or give it some more voltage. When you feel you have achieved your maximum overclock on the CPU test for at least one hour with Prime 95 Torture Test set to large FFT. No this is not considered stable but your almost there :)

X2 Update
If your using this guide to overclock your dual core AMD your going to do the same thing until you get to this point. Here things change a little. Your going to have to test each core independently. I like to use prime95, you will have to make two directories ( Prime95_1 and Prime95_2 ) for it to run, you will also need to set processor affinity, this option is under the Advanced menu.

Using 5MHz increments start to increase your FSB. If prime95 passes then repeat until it fails out, you can then either back off 5MHz or give it some more voltage. When you feel you have achieved your maximum overclock on the CPU test for at least one hour with Prime 95 Torture Test set to large FFT. No this is not considered stable but your almost there

Like single core stress testing you should run this for 12-24hrs on BOTH cores with no errors. Now your done.

A little FYI keep your eyes on heat, when I was running my 4400+ I was seeing temperatures around 60c on the PWMIC as well as 60c on the CPU when running at 2.6ghz. Two core's means more power so if you have a cheap board keep your eye on the heat so you don't burn out your motherboard.

If your motherboard is supported by clockgen you can speed up this process using the following method.


Set your CPU multiplier to its stock setting (the highest one), your memory to a low setting (133), and your HTT multiplier to 3x. Now boot into windows and start up clockgen and prime95 (both instances if your on an X2 or dual core Opteron). With Prime95 running start to increase your FSB in 5MHz increments, allow it to run for a few iterations and when it switches to the next FFT length you can increase your FSB again. Keep increasing your FSB until Prime95 fails and then either give your CPU more voltage or back off your overclock. Don’t forget to write down your highest speed reached were going to need it in just a second.

Putting it all together….
So you just spent several hours testing and rebooting your system and you have a scratch pad with a lot of numbers on it that don’t seem to mean much. Well guess what they do and were going to show you have to put together a basic but close to optimal overclock out of that information.

What we have tested is the limits of each of your components, each has a limit and in order to keep your system stable and usable we need to keep everything under these limits but as close to them as we can get them. So for the sake of making this easy I’m going to create an example.

Assume the following, your using a DFI Ultra-D with a 3800+ X2 and some OCZ Platinum EL memory.

Your maximum overclocks are as follows:
HTT= 235x5, 1175MHz
FSB= 321MHz
Mem= 265MHz
CPU= 272x10, 2720MHz

AMD A64 Calc Use this little tool to help you figure out what setings to use.

In this example we would like to run the CPU as fast as we can so close to 2700MHz (270x10) but because our memory cannot run at 270 FSB we cannot use the 10x multiplier. We could keep the CPU at 270x10 and simply drop the memory down to 166 but this would only give us a speed of 224MHz ( [166/200]*270 ) well under the 265MHz limit of the memory. So we drop down to the 9x multiplier and the memory down to 166. Now because we have lowered both we can start to increase the FSB but we must also keep the HTT limit in mind to get 2700MHz on the CPU we will need a FSB of 300MHz because 300x5 is more than 1175 we will need to run at 3x giving us an HTT of 900MHz (a slight underclock but this is the least important performance part). With the FSB at 300MHz the CPU will be running at 2700MHz and the memory will be running at 249MHz, much closer to 265MHz, and the HTT will be running at 900MHz. For our purposes this is as close to optimal as were going to get. If you wanted better performance it would be best to find memory that could run at the full speed of the CPU’s FSB i.e. 272MHz. That would allow you to run the CPU at 2720MHz the memory at 272MHz and the HTT at 1088MHz giving overall better performance.

The last thing you’re going to want to do when you have your numbers in place is set everything to its optimal setting and then run Prime95 for a few hours and 3dmark05 for a few hours. This will make sure that all the components work together at the respective speeds you have set them for.

Good luck and good overclocking


3DMark03
3DMark05
Memtest86+
SuperPi
Prime95
S&M
 
Last edited:
I've been looking for a post like this forever! Nice good way to start your overclocking...then to use the rest of the stickys and members to fine tune! Good Job! *Bump*
 
Thanks for the vote of confidence folks, it means allot to me when I put effort into something like this so that others can use it. I guess its just nice to see that effort justified <cue music>
 
awesome guide, easy and to the point. no long pages of paragraphs to look through.

im gonna use this once i get my A64 system back with me.


simple, yet effective.....STICKY THIS.
 
^^Agreed. Everyone needs a simple guide that everyone can follow. The beginners hate reading through all that crap up there in the stickies, this is a lot easier to use and follow.
 
Thanks for the post man. Hopefully it gets stickied so I can find it again when I get my new system. (going for broke this time.. AMD X2 and a Vapochill.. mmmm OC Heaven)
 
About a week or two ago when i was digging into the A64 world to see what to buy, most other guides are completely indepth and a ton to take in at once, this one simplifies it!~ *Bump for Stuckness*
 
Aww man, now you're making me want to trade in my AXP for a 64. Nice though, as A64 OC'ing confused the hell out of me...until now.
 
One question. Why would you test your max HTT speed if this is a quick overclocking guide. It is not like the speed matters at all in terms of performance, so why not just do 3x 300 in your case if you are just trying to quickly overclock it?
 
The reason for testing the max HTT is so you know that its not a limiting factor when it comes to your overclock. If you have a system like a Neo2 that does not like doing much over 1000mhz then you need to know that or when your testing your cpu/mem you could get results that look like the max on either component when in reality its you HTT that maxed out.
 
I know it can get unstable if it goes too high, but why not just bump down the hyper transport multiplier to begin with. There shouldn't be any negative effect.
 
Back