Phenom 9500 w/ MSI K9A2 Platinum

[KTE]

I have two settings in the bios labeld prozessor and north bridge voltage. Those settings match exactly with the CPU/NB VID settings (AMD PM, PCI registers, AOD).

MSR only stores VIDs, so yep, you are talking about CPU/NB VIDs. Now those are different to voltages, but they affect voltage. Each VID decides the lowest and highest voltage available and if you don't manipulate the voltage, it will be set to default as the lowest voltage available for that particular VID. I'll show you a quick VID->default voltage mapping here, it's still incomplete though.



If you place 00 and 10 for both CPU and NB, and say both give you the same MHz, then you won't be able to do better with any voltage option, because they do exactly that, control the voltages being fed. These are set with the Phenom CPUs not the BIOS, thus 000 = 1.536V and 010 = 1.416V.

ATM I test 212x11 and my Vcore readings are 1.264V load/1.248V idle. They are closer to the bios settings than on your mobo, noted em a few times but they are always in the rounding rounding area (+-0.016V I guess) so I always talk about what I set in the bios.

Ah OK. Well make sure you let others know too because it's confusing. Since VID is not the voltage. You're probably around CPU VID 22 there I'm guessing.

It should take around 4 hourst to find those two prim95-30min stable points. Using the max values for cpu and nb as a starting point for a long term stability test should save alot of time. It's a pitas to find em step by step.

Yes it is.

U know u know something I don't know.
http://www.abload.de/image.php?img=cpuvrf.jpg
But I have pictures. Much less units in this area.
Can you see on this image which resistors are responsible for cpu and nb voltage?

This is the back of the board, CPU VRM area? I can't even tell where on the board that is TBH.
There is no way to tell like this if I haven't seen and played with the board/specs to be frank. You have to change the VID, know the voltages it sets and then measure around the MOSFETs/inductors to find the right legs to measure which correspond to the voltage you're after. Then you have to start adding/decreasing resistance (as needed) to the resistors controlling the voltage to those components, and see how it affects the component voltage overall.

BTW: The specs of this L6740L chip are very interesting read, but I guess you know most of this stuff. Since how long do you overclock so far?

I haven't read its specifications so I don't really know, will do now but it should be 1 phase for CPU NB power control and 4 phase for CPU cores for Phenom MBs.

If you're asking how long I've been in computer oc, then not too long, probably since Pentium 66MHz, that's the first one me and a friend tried to overclock and destroyed it somewhere around '94-'95.
Had to work 9 months to cover the expenses, they were that bloody expensive and in rare supply where I was living!

[justapost]

MSR only stores VIDs, so yep, you are talking about CPU/NB VIDs. Now those are different to voltages, but they affect voltage. Each VID decides the lowest and highest voltage available and if you don't manipulate the voltage, it will be set to default as the lowest voltage available for that particular VID. I'll show you a quick VID->default voltage mapping here, it's still incomplete though.

Thx, I'l see what values I noted.

If you place 00 and 10 for both CPU and NB, and say both give you the same MHz, then you won't be able to do better with any voltage option, because they do exactly that, control the voltages being fed. These are set with the Phenom CPUs not the BIOS, thus 000 = 1.536V and 010 = 1.416V.

You said the VID is limited depending on your vcore setting. Have you checked MSR register C001_0071? Bits 48:42 are MinVid and bits 41:35 are MaxVid. I assume both are zero on the unlocked be. SviVID codes are listed on page 29 in the AMD BKDG and are identical to your table.

Ah OK. Well make sure you let others know too because it's confusing. Since VID is not the voltage. You're probably around CPU VID 22 there I'm guessing.

Sry, thought I mentioned it in the past. Gonna check the msd registers tomorrow, but amd pm seems to show cpu and nb vid's translated via the table.

This is the back of the board, CPU VRM area? I can't even tell where on the board that is TBH.
There is no way to tell like this if I haven't seen and played with the board/specs to be frank. You have to change the VID, know the voltages it sets and then measure around the MOSFETs/inductors to find the right legs to measure which correspond to the voltage you're after. Then you have to start adding/decreasing resistance (as needed) to the resistors controlling the voltage to those components, and see how it affects the component voltage overall.

Guess it's easier to find the resistances connected to the FB and FB_NB pin of that pwm chip. NP

I haven't read its specifications so I don't really know, will do now but it should be 1 phase for CPU NB power control and 4 phase for CPU cores for Phenom MBs.

If you're asking how long I've been in computer oc, then not too long, probably since Pentium 66MHz, that's the first one me and a friend tried to overclock and destroyed it somewhere around '94-'95.
Had to work 9 months to cover the expenses, they were that bloody expensive and in rare supply where I was living!

He He my Pentium 60MHz survived, but I payed round 4000€ for that system so it was to valuable even thinking about overclocking it. If overclocking by accident counts I might have started earlier however.

[Daveburt714]

Thanks for that chart KTE... Thats really helpful!!

Last time I tried I couldn't set NB Vid in bios though...

I wish I could set NB Fid to 8.5x, but dividers (DID) don't
seem to work either... I tried NB FID=13 / NB DID =2, which resulted
in a no post situation.. Do you know if 1/2 multis are possible on the NB?

Thanks.. Dave

[KTE]

Thx, I'l see what values I noted.

You said the VID is limited depending on your vcore setting.

Other way round, exactly.

If you want access to lower volts, decrease VID.
If you want access to higher volts, increase VID.

I believe the CPU and NB VIDs are mapped to the same range of voltages.

Have you checked MSR register C001_0071? Bits 48:42 are MinVid and bits 41:35 are MaxVid. I assume both are zero on the unlocked be.

Hmm..

SviVID codes are listed on page 29 in the AMD BKDG and are identical to your table.

Yeah, they're the same. We get 000-127 options too.

But they use binary=decimal VID values instead, which is of little use to us.



Have you seen this below BTW? This is why we were having boot-up problems with low NB DID then.



I wish they had unlocked HT downwards too! From >150 would've been good, especially handy when troubleshooting and limiting memory when testing through FID and DID combinations.

Guess it's easier to find the resistances connected to the FB and FB_NB pin of that pwm chip. NP

You sure?

Last time I tried I couldn't set NB Vid in bios though...

So you're using one of the new BIOSes I gather?

That's why I reverted to 1.13 BETA.

On this'an I'll test if DID works for NB or CPU.

I wish I could set NB Fid to 8.5x, but dividers (DID) don't
seem to work either... I tried NB FID=13 / NB DID =2, which resulted
in a no post situation.. Do you know if 1/2 multis are possible on the NB?

They are supposed to work you know, even 0.25 dividers are supposed to work and they are fairly simple options with PLLs, but for some reason both the CPU DID and NB DID options changed from 1 result in no POST IME.

[Daveburt714]

Thanks KTE, I thought the dividers should work, but I haven't had much luck with them. I'm not done yet though...

I tried a few of the other Bios versions but reverted back to 1.13 myself.

Sorry to interupt you guys conversation, it's fun to watch the masters at work!! Even though I don't understand 1/2 of it, I'm learning...

[Daveburt714]

I just had the strangest hard crash... I had to reset CMOS, but it still kept my presets in Bios!!

Pretty sure I violated the "Core COF/NB COF <=2" rule you quoted above though.... Excellent find BTW!

HT Ref 200, CPU Multi 11.5x, NB Multi 5x... Yep, pretty sure thats meets the criteria!!!

I hate to sound stupid, I get the general Idea, but what does "COF" stand for?

I wonder if there is a something similar to this with HTT Link speed? Just from playing with it, there always seemed to be some kind of corrilation between NB Multi and HTT Multi... They always seemed to run better at the same or with the HTT 1x lower.

[KTE]

If you get no POST, hold Insert and keep tapping Delete for a few reboots until you see the red error message and white writing below it, then hit F1 and you should be able to get into the BIOS.

I found something to note here =>

Underclocked my system, 2.3G/1.8G/800M CPU 1.175 VID 1.160V and IMC 1.050VID - all else the same at stock as before.

Power draw:
Idle = 90W AC (right now)
Load = 157W AC

Same as above but with CnQ enabled=>

Power draw:
Idle = 97W AC
Load = 159W AC



CnQ throws NB VID/voltage higher although it lowers CPU VID/voltage. Thus, manually setting low VID/volts gets me lower power than not. The power today is less because I have no HT/RD790 volts on today than earlier.

Thanks KTE, I thought the dividers should work, but I haven't had much luck with them. I'm not done yet though...

I tried a few of the other Bios versions but reverted back to 1.13 myself.

Have you tried:

NB FID = 18
NB VID = 25
NB DID = 2 (is there such an option?)

Sorry to interupt you guys conversation, it's fun to watch the masters at work!! Even though I don't understand 1/2 of it, I'm learning...

Well I know one thing for sure.. I'm not a master at all, just a normal guy.

COF: Current Operating Frequency

NB speed has to remain minimum 0.5x CPU speed and maximum 32x CPU speed.

Not read the guide enough to see for any more details.

jonspd: Which exact settings are you after? Which BIOS are you on?

[Daveburt714]

Have you tried:

NB FID = 18
NB VID = 25
NB DID = 2 (is there such an option?)

My chip is a standard 9600, if I'm figuring that correctly that would result in a NB multi of 11x... AFAIK this chip won't take a NB over 9x, but I'll admit I haven't tried yet.

When I do set NB VID in P-States it will take it, but AOD shows no change (still @ 1.1v). The SS below is with NB VID @ 32 which should be 1.15 if it was working correctly, If I change the NB VDDC in Bios AOD does show the change..



Those settings have proven to be ROCK SOLID on my system, but I know the NB will run @ 2200... Unfortunatly, it won't run @ 2250. That's why I was looking for 8.5 or 8.75 multi on the NB. I'll play with it somemore tommorow night but so far I haven't been able to get any combination with a DID to work...

Thanks for the tip on Insert/Delete combination!

And I don't care how modest you are! If such a thing as Phenom Masters exist, You and Just are them........

[KTE]

My chip is a standard 9600, if I'm figuring that correctly that would result in a NB multi of 11x... AFAIK this chip won't take a NB over 9x, but I'll admit I haven't tried yet.

Sorry about that, don't try those values (specifically DID). You'll have to clear CMOS or get no POST.

Plus 9x NB and 11.5x CPU on Phenom 9600 doesn't work, you're right.

When I do set NB VID in P-States it will take it, but AOD shows no change (still @ 1.1v). The SS below is with NB VID @ 32 which should be 1.15 if it was working correctly, If I change the NB VDDC in Bios AOD does show the change..

Nice settings there. For NB VID, you're looking at the wrong tool.

That'll change the RD790 voltage insteead, which is labeled NB and doesn't need to be touched.

Check in AMD Power Monitor: http://www.amd.com/us-en/assets/cont...ties/setup.exe

It'll show you the current operating VIDs for CPU and NB, as below:



The large bars in the center below core labels show core load and the smaller bars on the left show if CnQ is functioning. The frequencies are always 200 HT ref. x Multiplier, so if it shows 2400MHz, that could be anything above 2.4G but it has to be using the 12x multiplier.

Those settings have proven to be ROCK SOLID on my system, but I know the NB will run @ 2200... Unfortunatly, it won't run @ 2250. That's why I was looking for 8.5 or 8.75 multi on the NB. I'll play with it somemore tommorow night but so far I haven't been able to get any combination with a DID to work...

Thanks for the tip on Insert/Delete combination!

And I don't care how modest you are! If such a thing as Phenom Masters exist, You and Just are them........

Well, in that case, everyone at XS whose ever opened a thread is a Master...

[justapost]

Other way round, exactly.

If you want access to lower volts, decrease VID.
If you want access to higher volts, increase VID.

I believe the CPU and NB VIDs are mapped to the same range of voltages.

Can be it's a feature of your mobo's bios to prevent harmfull settings.

Hmm..

Yeah, they're the same. We get 000-127 options too.

But they use binary=decimal VID values instead, which is of little use to us.

Guess they use binary values because it's easier for developers, whom have to work with the registers.

Have you seen this below BTW? This is why we were having boot-up problems with low NB DID then.

Yepp, I saw it some time ago as i skimed thru that document looking for those dependencies. That time I thought I'll never be affected by those limits.

I wish they had unlocked HT downwards too! From >150 would've been good, especially handy when troubleshooting and limiting memory when testing through FID and DID combinations.

Yepp wished for lower ref HT also.

You sure?

I identified the chips and the relevant pins, just gonna find the resistances connected to those pins. The only issue is that area is infinitesimal in size. Can't follow the conducting paths with the eyes and the camera I have is also not high res enough.
Guess I can not measure resistances with the old osciloscope I have.

Do you have in mind which AMD document shows the pin assignment of the phenom cpu's?

BTW: My mobo has unganged settings in the bios (can be unhidden by hitting F4) so I can run everest benches in unanged mode for comparison here, but they will be lower than the ganged mode results. Most of your expamples in this thread use cpu frequencies i can not reach with my bios limiting vcore so i'd be happy if you could run everest at stock just for mobo clock per clock and timing by timing comparison.


So you're using one of the new BIOSes I gather?

That's why I reverted to 1.13 BETA.

On this'an I'll test if DID works for NB or CPU.

They are supposed to work you know, even 0.25 dividers are supposed to work and they are fairly simple options with PLLs, but for some reason both the CPU DID and NB DID options changed from 1 result in no POST IME. [/QUOTE]

[justapost]

Other way round, exactly.

If you want access to lower volts, decrease VID.
If you want access to higher volts, increase VID.

I believe the CPU and NB VIDs are mapped to the same range of voltages.

Can be it's a feature of your mobo's bios to prevent harmfull settings.

Hmm..

Yeah, they're the same. We get 000-127 options too.

But they use binary=decimal VID values instead, which is of little use to us.

Guess they use binary values because it's easier for developers, whom have to work with the registers.

Have you seen this below BTW? This is why we were having boot-up problems with low NB DID then.

Yepp, I saw it some time ago as i skimed thru that document looking for those dependencies. That time I thought I'll never be affected by those limits.

I wish they had unlocked HT downwards too! From >150 would've been good, especially handy when troubleshooting and limiting memory when testing through FID and DID combinations.

Yepp wished for lower ref HT also.

You sure?

I identified the chips and the relevant pins, just gonna find the resistances connected to those pins. The only issue is that area is infinitesimal in size. Can't follow the conducting paths with the eyes and the camera I have is also not high res enough.
Guess I can not measure resistances with the old osciloscope I have.

Do you have in mind which AMD document shows the pin assignment of the phenom cpu's?

BTW: My mobo has unganged settings in the bios (can be unhidden by hitting F4) so I can run everest benches in unanged mode for comparison here, but they will be lower than the ganged mode results. Most of your expamples in this thread use cpu frequencies i can not reach with my bios limiting vcore so i'd be happy if you could run everest at stock just for mobo clock per clock and timing by timing comparison.


Master does not sound bad, but i'm heading for guru state of mind.