Athlon64 Overclocking Optimizer

Gogar · 05-09-2005, 12:32 PM

Hi everybody i've been reading these forums for a while now but i've never actually posted here so first of all: greetings to everyone

I've recently made a script that allows you to fill in your maximum cpu, memory and htt speed, and it will try to find the best combinations of multipliers and dividers to run everything as close as possible to their maximum speed.

I've improved it a lot over the last few days since i read about how the memory clock is derived from the cpu frequency, it should now work correctly with half multipliers and memory dividers!

Update: I've added Rev E memory dividers to the script, the memory divider table, and the OC Browser. Hope everything is correct.

Athlon64 Overclocking Optimizer

Instructions:

1) Finding Maximum Values
First of all you should find out how far each component can overclock individually. You'll need to know your maximum HTT/FSB, CPU, RAM and optionally also the Hypertransport bus. People who don't know how to do this should have a read at this thread:

http://forums.anandtech.com/messagev...VIEWTMP=Linear

2) Fill Them In
Once you know all the maximum speeds that your hardware can do you simply fill them in in the form called "Known Maximum Values" on the page and hit " GO! ".
A table with HTT, multiplier and divider combinations should appear. The table is sorted according to how close the values are to your maximums, the best options should at the top.

3) Importance Weighting
The table Importance Weighting is where you can de define how important it is to you that the specific component runs close to it's maximum. It will make the trade-offs based on those settings.
Generally the importance weighting is fine on the default setting.

cpu 100 (cpu is king)
ram 10 (also take the memory into consideration)
ht 0 (not important)
htt 0 (not important)

When a weight is set to 0 it will still favour the highest possible multi or divider, so for example it won't suggest to run your HT bus at 2x while it could run at 3x.

In most situations you shouldn't have to change the weights.

4) Important Notes

To get a large number of results to choose from, it is important that you enter a high HTT value. Please do actually test what it's maximum is, instead of just entering the HTT that you're currently using.

Some memory dividers aren't available on all motherboards, if a memory divider is suggested in the table that isn't available on your board, you will have to look further down in the table for combinations that do work with your motherboard.
Another option could be to look for a hacked bios.

Use of my script or it's results is completely at your own risk.

mushk1n · 05-09-2005, 12:36 PM

looking good but what is that importance weighting and also i cant put my memory in a 9:10 divider :P

eshbach · 05-09-2005, 12:45 PM

not bad. compile it down to an exe and post it in the Coder's Corner.

Gogar · 05-09-2005, 12:47 PM

edit: info added to the top post

J-Mag · 05-09-2005, 01:29 PM

HAH! All your Base if you are screwing up (I typed in 2.55 for the mhz instead of 2550)
I like it makes it so I don't have to think (which hurts at times)

Although, your avatar kinda weirds me out... (is it something I recognize???)

cantankerous · 05-09-2005, 02:35 PM

nice one. Going to try it out now.

Major · 05-09-2005, 03:21 PM

Thats pretty cool

WeStSiDePLaYa · 05-09-2005, 03:31 PM

bookmarked, but would like exe.

nebuchanezzar · 05-09-2005, 03:59 PM

Nice..takes some scribbling outta the equation

Gogar · 05-09-2005, 05:27 PM

The 9:10(183), 7:10(141) and 3:5(125) divider should be available on the DFI NF4 boards.
I don't know which other boards support them but perhaps it's possible to choose those dividers with a hacked bios for your board.

They are actually 11:12, 17:24 and 5:8 though, but i guess DFI didn't know that because AMD tech docs specify them as "Reserved".

Scrufdog · 05-09-2005, 05:47 PM

looks good, i punched in my maxs and it told me what i am already running....good stuff

IvanAndreevich · 05-09-2005, 06:01 PM

Very good idea

I'll give it a go

Gogar · 05-10-2005, 12:41 PM

Hey i'm getting a few strange results sometimes. It seems that sometimes the integer divider gets rounded up (by the A64) even though it already is a perfect integer

like 11* 200 / 183.333 = 12 exactly yet someone reported that it was 13
this doesn't make sense... i'm trying to find out for which multi&divider combinations this occurs.. and trying to find a pattern..

Also that 4*200 / (1:1)200 = 160 this might also be a result of the effect rather than that the minimum divider is 5

This is the list so far of combinations where the A64 seems to pick the wrong divider integer:

4 * 200 / 200 one too high (possibly because the lowest divider is 5)

5.5 * 200 / 183.33 one too high
11 * 200 / 183.33 one too high
16.5 * 200 / 183.33 one too high

5 * 200 / 141.66 one too low
10 * 200 / 141.66 one too low
15 * 200 / 141.66 one too low
20 * 200 / 141.66 one too low
25 * 200 / 141.66 one too low

Added these offsets to the script.

Here's a table of the memory speeds that the script uses. (similar to Oskar WU's)

AMD for making errors.

quijonsith · 05-10-2005, 02:12 PM

Favoritized. Nice job.

saaya · 05-10-2005, 05:37 PM

hey gogar! welcome to XtremeSystems

very very nice tool! i thougt about asking somebody to make exactly a tool like that, you must have read my mind!

i stuck this thread

you should explain how to use it though in your first post

Mephisto · 05-11-2005, 06:54 AM

excellent!....iÂ´m going to test these steps....

ent · 05-11-2005, 08:16 AM

Very nice tool Gogar, when I enterned my limits on first place it showed my current settings. GJ!

wickedld9 · 05-11-2005, 08:27 AM

Very cool.

Now I don't have to wear out my calculator!

xavicl · 05-11-2005, 11:18 AM

very nice tool mate, very useful! but i dont like the 0.5 multis

Gogar · 05-11-2005, 11:46 AM

Quote:

Originally Posted by xavicl

very nice tool mate, very useful! but i dont like the 0.5 multis

Why don't you like them?
I know everywhere they advise against using half multipliers, but i think i've finally figured out exactly how it all sticks together. It should be perfectly fine to use the suggestions with half multipliers.

They add a lot more possible combinations, so the bigger the chances are that someone will be able to run both their memory and cpu really close to the max.

I don't see them as a bad thing anymore at all.

xavicl · 05-11-2005, 12:06 PM

running half multipliers causes memory frequencies to run lower than expected, thus lowering overall system bandwidth

Gogar · 05-11-2005, 02:18 PM

I know, but it takes that into account

When i enter 2700MHz as the maximum cpu speed and 250MHz for the ram, this is one of the entries in the table:
284 9.5 x 284 = 2698 9:10(183) x 284 = 245.27

You'll see that the memory speed it reports is correct. What you'd expect is 260,33MHz in this case.

dmo580 · 05-11-2005, 03:49 PM

So what if you don't have DFI and you can't do 9/10? =(

Gogar · 05-12-2005, 12:47 PM

Quote:

Originally Posted by dmo580

So what if you don't have DFI and you can't do 9/10? =(

I've completed some small changes that will result in a larger number of combinations in general. Mostly lesser ones, but it should provide more good combinations incase the divider used in the top row isn't available on your board.

Just look further down in the table for multi/divider combinations that you like.

I feel it's pretty much perfect now.
If it doesn't show combinations that run your cpu and your memory close to their max it just isn't possible. Don't blame me lol.

Ackbar · 05-12-2005, 12:52 PM

Interesting... so I noticed you use a Sum Square Error for your "objective function". Are you using an optimization algorithm or just searching the solution space for the maximum?

Edit: It seems like you're just displaying all solutions... are you weeding out any of them?