report: Athlon64 Overclocking HOWTO by mertsch
OK. . . at the end of last year I bought the rig you see in my sig At the time - pretty much up to now I did never OC my system since I saw no need in it everything did run smooth at max details except Riddick with soft shadows but I did not care about those now that I have much free time during vacation I thought ...
This topic was started by 24518f9903e3c3d5891f4cc2012e80afdb9f50e8,
OK ...
at the end of last year I bought the rig you see in my sig
At the time - pretty much up to now I did never OC my system since I saw no need in it everything did run smooth at max details except Riddick with soft shadows but I did not care about those
now that I have much free time during vacation
I thought about OC again and since most new games look Doom3 like its time for some extra juice
now why do I write this - there are numerous reports about OCing you CPU out there
because of the fact that there is no REAL guide how to do OC ... they post some results and some values bu no real guide how to try out if OC does work or not ...
any noob can set some settings but how do I verify that those do really work ?
OK lets go
first of all what is my system status
Athlon64 S939 Winchester with 1800MHz and 1,4V VCore (1000MHz @ 1,1V with QnC enabled)
Abit AV8 BIOS 24
2x CORSAIR XMS3200XL PRO PC3200 with 2-2-2-5 by default
After switching from my OCed AthlonXP I experienced some random crashes with my new PC. Any normal noob windows user wold now say "F***ing MS producing crappy Software - it crashes all the time"
But since I know that XP is an extremely stable OS I know that it has to work without crashes.
So before I OC anything I have to investigate the crashes first. By default the mainboard set FSB to 204 instead of 200 MHz FSB which I saw to be the main reason for the system instability since the memory timings are very aggressive. So I set the FSB to 201 and the crashes were gone. I still had some random problems but very very few so this seem to be the main reason.
I also set my RAM settings to 2-3-2-6 to improve stability (hopefully)
the command rate was at 2T, too
Now just before overclocking I saw that my memory voltage was at 2,75V which results in 2,79V real voltage and I remembered that most high performance modules are rated to work at 2,85V so I raised the mem voltage to 2,85V which results in 2,89V real.
I did not test the system stability after that I am sure it was stable now.
so the most important thing before you OC is:
VERIFY !YOUR! system to be 100% stable before OCing anything a fast system is nothing when it is not stable and when it is unstable before overclocking how will you later verify its stability ?
NOTE: many people and sites do say something like: your system is stable when you run Prime for a day or something like that
this is bullshit !
a system is stable when it does not crash in any way during normal operation
and normal operation means gaming, idle and multimedia ... prime is nice but you cannot verify anything with a single app.
When I had OCed my AthlonXP I was able to play hours every game I wanted and there was not a single crash but running eMule over night (idle for the PC) did show that it was not stable since eMule was no longer running in the morning due to restart(s)
now lets OC:
I once did already try to OC my system with a quick test
I set my FSB to 250MHz the HTT multi to 4x instead of 5x which result in the same 1000MHz HTT
and my RAM to DDR333 and my CPU to 7x which means 1750MHz so lower then by default.
but the result was a system totally unstable it did even crash during booting process
You see the method here ... always do OC step by step !!! never OC everything at once. But why didn't it work? maybe FSB 250 is too high...
After reading an article about the Venice core and its new memory controller I saw that the command rate does not really have something to do with stability but with compatibility and since my system is in the configuration to allow a command rate of 1T I set it to 1T from now on
So now I do try it again lets try a small FSB boost first 220MHz which results in:
1760MHz CPU
RAM set to DDR333
and 880MHz HTT
Oh btw all RAM settings are set to SPD settings @ DDR400 2-2-2-5
as you can see my system is now SLOWER then before! BUT I can verify is FSB220 is stable since everything else is underclocked.
and after 2 days of stress with gaming and running 3Dmark05 over night I saw not a single crash
I raised the FSB to 230
1840 CPU
RAM set to DDR333
and 920MHz HTT
stable
and I started to wonder ... hmmm what does that DDR333 mean .... I know that it is just a divisor and has nothing to do with DDR333 when you OC your system. So I created an excel sheet and started a bit calculating - how do I create DDR333 from DDR400 and what does this mean on 230FSB (DDR460) and I found that the setting DDR333 actually means divide the FSB by 6 and multiply the result by 5
400 / 6 = 66,66 66,66 * 5 = 333,33
now lets think again what does this mean on FSB250 (DDR500) ?
500 / 6 = 83,33 83,33 * 5 = 416,66
whooooo thats kinda surprising ... that is the reason why my first try with 250FSB did not work ... I overclocked my already unstable RAM
OK what would happen if I set the RAM to DDR266 ? (div 3 mul 2)
(500 / 3) * 2 = 333,33
ah OK now lets test that
FSB to 250
*7 = 1750 MHz CPU
(/3)*2 = DDR333 RAM (by setting it to DDR266 in BIOS)
*4 = 1000MHz HTT
OK everything is underclocked or at default except the FSB
NOTE: on an Athlon64 the FSB does no longer have a real impact on performance so the system is still SLOWER then before.
OK 3 days if intensive testing followed
hours of gaming passed by (CSS does really make fun - I hate CS)
idling was no problem ... and running 3Dmark05 about 100 times did work without a problem (NO! did not start it 100 times I ran it in a loop )
so this means 250MHz FSB is stable and I now have room for overclocking
NOTE: the Athlon64 multiplier is locked you can only set less which has been integrated to make Cool and Quite work
since HTT is at default and it does not make sense to OC this one (others have proved that already) 250FSB is a very good value
lets go in small steps and try CPU multi of 8 first
250*8 = 2000MHz CPU (a plus of 200MHz) since you (always) need to raise the Vcore a bit I set it to 1,45V (1,40V is default)
another 2 days passed by and not a single crash
OK now lets do some serious stuff
250*9 = 2250 MHZ CPU (a plus of 450MHz) @ 1,475V as precaution
again 3 days passed by without a single crash now that rocks
but one thing really sucks ... the RAM is now massivly underclocked the timings are at its best be the clock rate is too low - so what do we do ? right ! raise the FSB even more
set it to 260MHz FSB CPU back to *8=2080 HTT to *3=780
BUT the system was very unstable crashed within minutes
OK back to 250 with CPU*9 and HTT*4
lets have a look at the CPU again:
more VCore does mean MUCH more power loss !!! means higher bill and more heat
since I did raise it as a matter of precaution lets go back
1,45V @ 2250MHz - stable over 2 days
1,425V @ 2250MHz - again stable
Hell thats nice! a 450MHz plus with nearly no VCore increase
OK I can't stop thinking of the RAM at least a bit more FSB must be possible
lets try 256MHz FSB thats a good value which results in:
256*9=2304MHZ CPU @ 1,425V (even more - a plus of 500MHz)
(512/3)*2 = 341,33DDR (at least a bit more)
and 256*4=1024 HTT (OCed a bit but thats no problem)
and this was my last change for now the system now works over a week with those settings without a single crash.
EDIT: I almost forgot about CnQ
when Overclocking my motherboard automatically stops the CPU using QnC which is quite logic since no one can guarantee the stability.
but I still want to use CnQ since it is a very good thing
So I do use http://cpu.rightmark.org/products/rmclock.shtml RMCLOCK
to do CnQ per software and it works
256*5 = 1277 MHz @ 1,125V (note the app does not detect the +0,025V I have set setting it to 1,1V does result in 1,125V)
also in idle my PC does not crash ... maybe I try even lower Voltage for idle
I did not do much benchmarks
but 3DMark05 did increase (mainly in CPU benchmark for sure)
and the program I did write myself did calculate the faculty of 1.000.000 ( 1.000.000! ) in 3,2 hours instead of 4,3 before OCing which is a huge increase in calculating power as you see
also I did show the F.E.A.R. demo to a friend and it ran much better then when I had tested it without an overclocked PC
I was even able to turn on soft shadows without making the game impossible to play
but I still prefer playing without soft shadows - the effect is too small to waste so much performance on it
anyway the game runs fine with every detail ON except soft shadows which makes my 10 month old system ready for even the latest games
Bet on soldiers also did work well
This is it for now. Good luck with your overclocking and never forget:
- only overclock ONE THING at once
- verify every step you do with at least 2 days of excessive PC usage
- you can call a system stable when it does run at least 1 week without a crash (non stop on - not one hour per day )
I will do some more OC when I need more power
maybe I am able to adjust the RAM timings to make it work at DDR426
but maybe I go in the opposite direction and lower the timings even more which can have the same effect
at the end of last year I bought the rig you see in my sig
At the time - pretty much up to now I did never OC my system since I saw no need in it everything did run smooth at max details except Riddick with soft shadows but I did not care about those
now that I have much free time during vacation
I thought about OC again and since most new games look Doom3 like its time for some extra juice
now why do I write this - there are numerous reports about OCing you CPU out there
because of the fact that there is no REAL guide how to do OC ... they post some results and some values bu no real guide how to try out if OC does work or not ...
any noob can set some settings but how do I verify that those do really work ?
OK lets go
first of all what is my system status
Athlon64 S939 Winchester with 1800MHz and 1,4V VCore (1000MHz @ 1,1V with QnC enabled)
Abit AV8 BIOS 24
2x CORSAIR XMS3200XL PRO PC3200 with 2-2-2-5 by default
After switching from my OCed AthlonXP I experienced some random crashes with my new PC. Any normal noob windows user wold now say "F***ing MS producing crappy Software - it crashes all the time"
But since I know that XP is an extremely stable OS I know that it has to work without crashes.
So before I OC anything I have to investigate the crashes first. By default the mainboard set FSB to 204 instead of 200 MHz FSB which I saw to be the main reason for the system instability since the memory timings are very aggressive. So I set the FSB to 201 and the crashes were gone. I still had some random problems but very very few so this seem to be the main reason.
I also set my RAM settings to 2-3-2-6 to improve stability (hopefully)
the command rate was at 2T, too
Now just before overclocking I saw that my memory voltage was at 2,75V which results in 2,79V real voltage and I remembered that most high performance modules are rated to work at 2,85V so I raised the mem voltage to 2,85V which results in 2,89V real.
I did not test the system stability after that I am sure it was stable now.
so the most important thing before you OC is:
VERIFY !YOUR! system to be 100% stable before OCing anything a fast system is nothing when it is not stable and when it is unstable before overclocking how will you later verify its stability ?
NOTE: many people and sites do say something like: your system is stable when you run Prime for a day or something like that
this is bullshit !
a system is stable when it does not crash in any way during normal operation
and normal operation means gaming, idle and multimedia ... prime is nice but you cannot verify anything with a single app.
When I had OCed my AthlonXP I was able to play hours every game I wanted and there was not a single crash but running eMule over night (idle for the PC) did show that it was not stable since eMule was no longer running in the morning due to restart(s)
now lets OC:
I once did already try to OC my system with a quick test
I set my FSB to 250MHz the HTT multi to 4x instead of 5x which result in the same 1000MHz HTT
and my RAM to DDR333 and my CPU to 7x which means 1750MHz so lower then by default.
but the result was a system totally unstable it did even crash during booting process
You see the method here ... always do OC step by step !!! never OC everything at once. But why didn't it work? maybe FSB 250 is too high...
After reading an article about the Venice core and its new memory controller I saw that the command rate does not really have something to do with stability but with compatibility and since my system is in the configuration to allow a command rate of 1T I set it to 1T from now on
So now I do try it again lets try a small FSB boost first 220MHz which results in:
1760MHz CPU
RAM set to DDR333
and 880MHz HTT
Oh btw all RAM settings are set to SPD settings @ DDR400 2-2-2-5
as you can see my system is now SLOWER then before! BUT I can verify is FSB220 is stable since everything else is underclocked.
and after 2 days of stress with gaming and running 3Dmark05 over night I saw not a single crash
I raised the FSB to 230
1840 CPU
RAM set to DDR333
and 920MHz HTT
stable
and I started to wonder ... hmmm what does that DDR333 mean .... I know that it is just a divisor and has nothing to do with DDR333 when you OC your system. So I created an excel sheet and started a bit calculating - how do I create DDR333 from DDR400 and what does this mean on 230FSB (DDR460) and I found that the setting DDR333 actually means divide the FSB by 6 and multiply the result by 5
400 / 6 = 66,66 66,66 * 5 = 333,33
now lets think again what does this mean on FSB250 (DDR500) ?
500 / 6 = 83,33 83,33 * 5 = 416,66
whooooo thats kinda surprising ... that is the reason why my first try with 250FSB did not work ... I overclocked my already unstable RAM
OK what would happen if I set the RAM to DDR266 ? (div 3 mul 2)
(500 / 3) * 2 = 333,33
ah OK now lets test that
FSB to 250
*7 = 1750 MHz CPU
(/3)*2 = DDR333 RAM (by setting it to DDR266 in BIOS)
*4 = 1000MHz HTT
OK everything is underclocked or at default except the FSB
NOTE: on an Athlon64 the FSB does no longer have a real impact on performance so the system is still SLOWER then before.
OK 3 days if intensive testing followed
hours of gaming passed by (CSS does really make fun - I hate CS)
idling was no problem ... and running 3Dmark05 about 100 times did work without a problem (NO! did not start it 100 times I ran it in a loop )
so this means 250MHz FSB is stable and I now have room for overclocking
NOTE: the Athlon64 multiplier is locked you can only set less which has been integrated to make Cool and Quite work
since HTT is at default and it does not make sense to OC this one (others have proved that already) 250FSB is a very good value
lets go in small steps and try CPU multi of 8 first
250*8 = 2000MHz CPU (a plus of 200MHz) since you (always) need to raise the Vcore a bit I set it to 1,45V (1,40V is default)
another 2 days passed by and not a single crash
OK now lets do some serious stuff
250*9 = 2250 MHZ CPU (a plus of 450MHz) @ 1,475V as precaution
again 3 days passed by without a single crash now that rocks
but one thing really sucks ... the RAM is now massivly underclocked the timings are at its best be the clock rate is too low - so what do we do ? right ! raise the FSB even more
set it to 260MHz FSB CPU back to *8=2080 HTT to *3=780
BUT the system was very unstable crashed within minutes
OK back to 250 with CPU*9 and HTT*4
lets have a look at the CPU again:
more VCore does mean MUCH more power loss !!! means higher bill and more heat
since I did raise it as a matter of precaution lets go back
1,45V @ 2250MHz - stable over 2 days
1,425V @ 2250MHz - again stable
Hell thats nice! a 450MHz plus with nearly no VCore increase
OK I can't stop thinking of the RAM at least a bit more FSB must be possible
lets try 256MHz FSB thats a good value which results in:
256*9=2304MHZ CPU @ 1,425V (even more - a plus of 500MHz)
(512/3)*2 = 341,33DDR (at least a bit more)
and 256*4=1024 HTT (OCed a bit but thats no problem)
and this was my last change for now the system now works over a week with those settings without a single crash.
EDIT: I almost forgot about CnQ
when Overclocking my motherboard automatically stops the CPU using QnC which is quite logic since no one can guarantee the stability.
but I still want to use CnQ since it is a very good thing
So I do use http://cpu.rightmark.org/products/rmclock.shtml RMCLOCK
to do CnQ per software and it works
256*5 = 1277 MHz @ 1,125V (note the app does not detect the +0,025V I have set setting it to 1,1V does result in 1,125V)
also in idle my PC does not crash ... maybe I try even lower Voltage for idle
I did not do much benchmarks
but 3DMark05 did increase (mainly in CPU benchmark for sure)
and the program I did write myself did calculate the faculty of 1.000.000 ( 1.000.000! ) in 3,2 hours instead of 4,3 before OCing which is a huge increase in calculating power as you see
also I did show the F.E.A.R. demo to a friend and it ran much better then when I had tested it without an overclocked PC
I was even able to turn on soft shadows without making the game impossible to play
but I still prefer playing without soft shadows - the effect is too small to waste so much performance on it
anyway the game runs fine with every detail ON except soft shadows which makes my 10 month old system ready for even the latest games
Bet on soldiers also did work well
This is it for now. Good luck with your overclocking and never forget:
- only overclock ONE THING at once
- verify every step you do with at least 2 days of excessive PC usage
- you can call a system stable when it does run at least 1 week without a crash (non stop on - not one hour per day )
I will do some more OC when I need more power
maybe I am able to adjust the RAM timings to make it work at DDR426
but maybe I go in the opposite direction and lower the timings even more which can have the same effect
Participate on our website and join the conversation
This topic is archived. New comments cannot be posted and votes cannot be cast.
Responses to this topic
now this is strange I did try again to higher my FSB
I did try to put my HTT multi from 4 to 3 in order NOT to overclock it and started Memtest+ without any problem ... I raised FSB to 270 and lowered CPU to x8 instead of x9 and PC startet ... raised Fsb to 288 which results in 2304 MHz CPU which was working before, too
booting into memtest did work and memtest did run without a flaw
I thought everything is OK and tried to boot into windows which did not work - it did not even show the logo :-/
now I set back everything except HTT left on 3x and windows boots without a problem ... I will try to reinstall windows with the 288FSB maybe that helps
I will report back in some days
I did try to put my HTT multi from 4 to 3 in order NOT to overclock it and started Memtest+ without any problem ... I raised FSB to 270 and lowered CPU to x8 instead of x9 and PC startet ... raised Fsb to 288 which results in 2304 MHz CPU which was working before, too
booting into memtest did work and memtest did run without a flaw
I thought everything is OK and tried to boot into windows which did not work - it did not even show the logo :-/
now I set back everything except HTT left on 3x and windows boots without a problem ... I will try to reinstall windows with the 288FSB maybe that helps
I will report back in some days
humm....
i donĀ“t understand, why u want to install windows new with an other FSB? :hmm:
i donĀ“t understand, why u want to install windows new with an other FSB? :hmm:
Ok
I did a little more investigation ...
before I wanted to kill my windows for nothing I wanted to do a test first
since PC works @ FSB280 and memtest runs flawlessly
I thought what about windowsPE ?
its a very light version of windowsXP bootable from CD ... and guess ... yes it did start with FSB255 and FSB280
compared to normal XP which did show up the logo for a second and crashed
so my gues is that there is somekind of driver (maybe the AMD CPU driver) or system component which does not allow to make a "CPU change" ... this is only a guess ... I will try to reinstall windows soon and watch what happens then
since I am able to make a full disk image I can revert the action (if it fails) very easily
I did a little more investigation ...
before I wanted to kill my windows for nothing I wanted to do a test first
since PC works @ FSB280 and memtest runs flawlessly
I thought what about windowsPE ?
its a very light version of windowsXP bootable from CD ... and guess ... yes it did start with FSB255 and FSB280
compared to normal XP which did show up the logo for a second and crashed
so my gues is that there is somekind of driver (maybe the AMD CPU driver) or system component which does not allow to make a "CPU change" ... this is only a guess ... I will try to reinstall windows soon and watch what happens then
since I am able to make a full disk image I can revert the action (if it fails) very easily
I did try to OC my RAM again, but with no success
my system simple does not like to go over DDR400
I highly doubt its because of the RAM ... I temporarily had 2gig of real DDR500 RAM and I was not able to get it to work
the 256MHz FSB are a max, too
It works perfect over month but as soon as I add one more MHz my system will become unstable
I guess the VIA chipset is not the best overclocker out there
since I was not able to increase the RAM speed I reduced the RAM voltage to 2.60v which results in a real 2.58v
my system simple does not like to go over DDR400
I highly doubt its because of the RAM ... I temporarily had 2gig of real DDR500 RAM and I was not able to get it to work
the 256MHz FSB are a max, too
It works perfect over month but as soon as I add one more MHz my system will become unstable
I guess the VIA chipset is not the best overclocker out there
since I was not able to increase the RAM speed I reduced the RAM voltage to 2.60v which results in a real 2.58v
my max OC is
CPU: FSB 300Mhz x 8 x 1,4375V @2400Mhz
RAM: FSB 200Mhz x 2,7V @2.2.2.5
i could not run stable RAM with 256Mhz, even with 2,9V because of 4x512MB? But i could do some Everest Benchmarks....almost same as 200mhz
CPU: FSB 300Mhz x 8 x 1,4375V @2400Mhz
RAM: FSB 200Mhz x 2,7V @2.2.2.5
i could not run stable RAM with 256Mhz, even with 2,9V because of 4x512MB? But i could do some Everest Benchmarks....almost same as 200mhz