ABit AX78 Overclocking/Stability

[KTE]

You can see my latest here and here

Board is going solid! I've not added any voltage yet though..

This is where Abit utility/BIOS reads sensors from:

;===========================================================
;	W83627EHF Index and Internal Pin Define
;===========================================================
;	Pin name			Index	Bank		
;	---------			----	----
;	"VCORE"			0x20	0
;	"VIN0"			0x21	0
;	"VAVCC"			0x22	0
;	"V3VCC"			0x23	0
;	"VIN1"			0x24	0
;	"VIN2"			0x25	0
;   	"VIN3"			0x26	0
;	"VSB"			        0x50	5
;	"VBAT"			0x51	5
;	"VIN4"			0x52	5
;	"CPUTIN"			0x50	1
;	"AUXTIN"		        0x50	2
;	"SYSTIN"		        0x27	0
;	"SYSFANIN"		      0x28	0
;	"CPUFANIN0"		0x29	0
;	"AUXFANIN0"		0x2A	0
;	"CPUFANIN1"		0x3F	0
;	"AUXFANIN1"		0x53	5
===========================================================

[justapost]

;===========================================================
;	W83627EHF Index and Internal Pin Define
;===========================================================
;	Pin name			Index	Bank		
;	---------			----	----
;	"VCORE"			0x20	0
;	"VIN0"			0x21	0
;	"VAVCC"			0x22	0
;	"V3VCC"			0x23	0
;	"VIN1"			0x24	0
;	"VIN2"			0x25	0
;   	"VIN3"			0x26	0
;	"VSB"			        0x50	5
;	"VBAT"			0x51	5
;	"VIN4"			0x52	5
;	"CPUTIN"			0x50	1
;	"AUXTIN"		        0x50	2
;	"SYSTIN"		        0x27	0
;	"SYSFANIN"		      0x28	0
;	"CPUFANIN0"		0x29	0
;	"AUXFANIN0"		0x2A	0
;	"CPUFANIN1"		0x3F	0
;	"AUXFANIN1"		0x53	5
===========================================================

Thank you for the effort KTE.

So ABit AX78 uses the W83627EHF sensor chip. This table seems to be the label mapping for this chip.
I'd need a list of detected chips from the SMBus interface. I explained it as far as i could over at the Unified DFI thread.

I wonder if the freezing might also occure if one of those better clockers run with a too high voltage for them.

[hokiealumnus]

Thank you for the effort KTE.

So ABit AX78 uses the W83627EHF sensor chip. This table seems to be the label mapping for this chip.
I'd need a list of detected chips from the SMBus interface. I explained it as far as i could over at the Unified DFI thread.

I wonder if the freezing might also occure if one of those better clockers run with a too high voltage for them.

+1. Great info to know, both of you. Thanks!

[KTE]

Latest stable so far: 3000-2900-2700-2800 (MHz) stock volts
12W AC higher full load power than stock 9850



Still testing...

I'd need a list of detected chips from the SMBus interface. I explained it as far as i could over at the Unified DFI thread.

Can't get any info, installed Etch and started Lm-sensor detection and system froze
Does not happen with Ubuntu, its a Debian thing it seems.

This is the only official info I know, not much there though;

I wonder if the freezing might also occure if one of those better clockers run with a too high voltage for them.

Hmm.. I've already done this test quite a while ago. With mine what caused freezing at stock volts also caused it at 1.55v - if it was idle instable, its always idle instable no matter what you do

For me, its just some weaker cores, I mean, you can see my above results and compare that I was stuck at 2700MHz before max

[justapost]

Can't get any info, installed Etch and started Lm-sensor detection and system froze
Does not happen with Ubuntu, its a Debian thing it seems.

Umm, that's why I run and asked for "i2cdetect 0" first.
That does only detect the addresses where chips live and does not test the chips. I'd not run sensors-detect with a phenom mounted. I tested all other boards i have here with an X2 cpu. The system only froze on the Sapphire board,not on the M3A(770) the GBT780G and the M2A-VM (690G).
Running "i2cdump 0" with the i2c-piix4 and the i2c-dev module loaded did never freeze a system but it makes the smbus/i2c chip unusable till i restart the system (cold start). It does not interfer with the it87 module which is used for the sensor-chips on all of my boards.
I wonder if "i2cdump 0" also shows a chip at 0x2e, whom is responsible for the freezing.
What version of lm-sensors comes with etch? I use lenny so i had the latest stable 3.0.1 version here.

This is the only official info I know, not much there though;

Thanks, that is from the SB600 docs i guess?

Hmm.. I've already done this test quite a while ago. With mine what caused freezing at stock volts also caused it at 1.55v - if it was idle instable, its always idle instable no matter what you do

Yeah but it can be that with 1.55V the bad clocker at stock whould have been stable and the better clocker are causing the freezes now because they do not like those hight voltages.

For me, its just some weaker cores, I mean, you can see my above results and compare that I was stuck at 2700MHz before max

For the p-states amd says the multis and dividers must be equal for all cores.
We know that this is no requirement but I'm not so sure if it will not cause problems 24/7 longterm.
My last 9850BE also had a bad third core and the fourth also was no good clocker. Cores one and two ran 3GHz at ~1.325V.

[KTE]

New update - core0 3000 stable stock volts, but core1 will fail, max is 2900 at stock HT - so bump in voltage, now added 0.05v, at 1.344v idle/1.338v load at 3000/3000/2700/2800 testing for another 24 hours or so. No fail yet, just ran 2 benches and EVEREST stability test 30 mins. +1 hr gone now...

Umm, that's why I run and asked for "i2cdetect 0" first.
That does only detect the addresses where chips live and does not test the chips. I'd not run sensors-detect with a phenom mounted. I tested all other boards i have here with an X2 cpu. The system only froze on the Sapphire board,not on the M3A(770) the GBT780G and the M2A-VM (690G).

Yeah, I know what it does but like I said it never worked. Installed latest lm-sensors manually (2.10.1) and then ran i2cdetect 0, never gave any feedback, failed and then second time froze my system. Rebooted and then ran it again, no output. Then ran i2cdetect -l and it again froze my system. No idea if its broken but none of these commands nor sensors-detect does anything odd in Ubuntu, I had tried it once before a few days back.

The guide on lm-sensors mentions clearly that its not safe to run:

WARNING

This program can confuse your I2C bus, cause data loss and worse!

So I'm a bit weary of trying it again, I don't want a screwed up system again like so many times before.

Running "i2cdump 0" with the i2c-piix4 and the i2c-dev module loaded did never freeze a system but it makes the smbus/i2c chip unusable till i restart the system (cold start). It does not interfer with the it87 module which is used for the sensor-chips on all of my boards.
I wonder if "i2cdump 0" also shows a chip at 0x2e, whom is responsible for the freezing.

Will try in Ubuntu and check - just give me some hours to test stability first.

Thanks, that is from the SB600 docs i guess?

Yes

Yeah but it can be that with 1.55V the bad clocker at stock whould have been stable and the better clocker are causing the freezes now because they do not like those hight voltages.

I didn't understand what you said there

For the p-states amd says the multis and dividers must be equal for all cores.
We know that this is no requirement but I'm not so sure if it will not cause problems 24/7 longterm.

I'm perfectly fine with it since AMD themslves demo'd and allowed us to clock cores individually in their own public software: AOD.
If it was problematic, they wouldn't do that. Anything causing them headaches, they always remove, even if its something rare.

My last 9850BE also had a bad third core and the fourth also was no good clocker. Cores one and two ran 3GHz at ~1.325V.

Looks like mine and yours are similar. Mine is on 3000 1.344v/1.338v idle/load stable now for about an hour. Load, it can do 3055 on all cores at that voltage stable, but idling I'm still testing

[justapost]

Yeah, I know what it does but like I said it never worked. Installed latest lm-sensors manually (2.10.1) and then ran i2cdetect 0, never gave any feedback, failed and then second time froze my system. Rebooted and then ran it again, no output. Then ran i2cdetect -l and it again froze my system. No idea if its broken but none of these commands nor sensors-detect does anything odd in Ubuntu, I had tried it once before a few days back.

Ah ok, thought you ran sensors-detect and not i2cdetect.
Hmm 2.10.1 is an old version. SB700 support was added at 2.10.3 and the latests is 3.0.1. Starting with 2.10.4 there is also a module for your sensor chip available (w83627ehf).
You need SB700 support in i2c-piix4 to dump the i2c/smbus devices.

Will try in Ubuntu and check - just give me some hours to test stability first.

Thanks, I got my hands on the spec of the clock chip they used on the M3A and the Sapphire board, the GBT uses an very similar chip. I also have the specs of the clock chip used on the M2A-VM. So i'll try to write a small kernel module to read out ref HT and pci express clocks. It might be extended to change those frequencies in the future. Depends on how much time it will take.
Yes

I didn't understand what you said there

Say core0 need 1.3V to be stable at 3GHz but it is not stable at say 1.45V.
Core2 is stable up to 2.7GHz with 1.3V and is also stable at 3GHz with 1.45V.
In this situation with 1.45V applied it might be core0 causing the freeze.
It's a vague theory and i don't like those.

I'm perfectly fine with it since AMD themslves demo'd and allowed us to clock cores individually in their own public software: AOD.
If it was problematic, they wouldn't do that. Anything causing them headaches, they always remove, even if its something rare.

Good point.

Looks like mine and yours are similar. Mine is on 3000 1.344v/1.338v idle/load stable now for about an hour. Load, it can do 3055 on all cores at that voltage stable, but idling I'm still testing

The question is, is the goal (3GHz phenom) reached now.

[KTE]

No, not stable
3000/3000/2700/2800 1.344v ID failed just after 2 hours
3000/2900/2700/2800 passed stock volts though.

I think I've reached a limit whereafter voltage won't make a difference. I only want to add upto 1.45v real max, no more but I don't think I can get more out of core1/2/3. Core0 may do 3.1G, I know it'll bench/load test it but it may not be idle stable after 1-2 hours.

Hmm 2.10.1 is an old version. SB700 support was added at 2.10.3 and the latests is 3.0.1. Starting with 2.10.4 there is also a module for your sensor chip available (w83627ehf).
You need SB700 support in i2c-piix4 to dump the i2c/smbus devices.

Ah ok, thanks. The version I downloaded said it was the latest but the version number was exactly what I posted. Nevermind

Thanks, I got my hands on the spec of the clock chip they used on the M3A and the Sapphire board, the GBT uses an very similar chip. I also have the specs of the clock chip used on the M2A-VM. So i'll try to write a small kernel module to read out ref HT and pci express clocks. It might be extended to change those frequencies in the future. Depends on how much time it will take.

Great. Is there a pin on the I/O chip of those chips reading the HT Ref?
I also have the GBT 780G again BTW.

Say core0 need 1.3V to be stable at 3GHz but it is not stable at say 1.45V.
Core2 is stable up to 2.7GHz with 1.3V and is also stable at 3GHz with 1.45V.
In this situation with 1.45V applied it might be core0 causing the freeze.
It's a vague theory and i don't like those.

Agreed, it could be. That's a good point, I'll have to think that over - how to eliminate. Maybe testing by disabling one core at a time would be ok?

I'm not sure but, if you disable a core in Windows but reboot setting say 3000 across all cores in BIOS -- does the 4th core get set to 3000 aswell or do they keep it at stock on boot?

I got into Ubuntu, ran it and got all I/O details fine. Was posting them to you but the system froze (3000/3000/3000/3000 1.355v). So have to go back into now and will post.