[Sammelthread] Sockel 939 / 940 / 754 - Thread

[Sparksnl]

My CAB2E 0608BPMW 3500+ does:
2900 1,35v
3000 1,44v
3100 1,49v
3200 1,67v

Let’s call it a silver cpu…

 

[Sparksnl]

I was talking with Sam today (Tapakah), I asked him about max FSB for s939 and his findings for sub zero. I told him I test max FSB on air.
He answered with this answer:

---
Yep, I know that max FSB is inversely proportional to temp on 939 but the scaling depends on the CPU Some chips can do 400+ ambient and stop below 300 already at -40. Some chips barely do 350-360 ambient, yet still do 320+ at -100. Some chips even have different cold bugs with different multis or RAM dividers (like this one which wouldn't do above 3.7 with x13)

It is, however, a good way to find LN2 candidates when benching on single stage. If a chip maxes out at 340x10 on single stage, yet can do 400+ FSB at around the same temps, then it will do 3.6-3.7 on LN2.
---

I think there is still lot's of info we still miss which people don't share about this subject.

 

[Raiden Zero]

I already figured that there is a lot of info others don’t really like to share or at least they did not want to do so when the platform was somewhat relevant and now most don’t ask or care anymore.

That’s why I like benching oldschool stuff. Most people don’t really mind sharing their knowledge anymore.

Edit:
Just finished first tests on my Asus A8N32. As you said it might be even better than DFI for CPU OC. At least I was able to bench my X2 4800+ up to 3.26G while it maxed out at 3.25G on my DFI. 10MHz is not much tho. So I might have to test more CPUs to be sure.
Still a really good board. Have to give Mem OC a try later.

 

[Tzk]

but the scaling depends on the CPU

Now the interesting part is if it depends on the exact chip or the stepping. If it is stepping dependent, then that will make your life a lot easier.

 

[Sparksnl]

I am guessing it depends on what is written where I marked BW here LBBWE.
Stunned should be able to tell more about good batches I think.

 

[Raiden Zero]

Thing about BW chips is that they usually clock really bad overall or have a low (MHz wise) cold bug. Those I had which were capable of high core clocks would only hit around 300bclk on water while those with high BCLK would suck core clock wise. So at least judging from my observations one would have to be lucky to find a CPU capable of both, high BCLK and core clock.

BTW: After one day of extensive testing on 939 without any DFI available I already miss it. All those other boards are really good for CPU overclocking or at least not worse than my DFI. But once you try to get good 2D performance and need decent Mem clocks there’s just no way around a DFI NF4.
While I was able to run DDR600+ with my TCCDs on S754 that’s not the case for S939 using A8N32 or Abit KN8 SLI. With the same exact settings (which need a lot more effort to even set) my TCCD max out at 280MHz.

Guess there’s a reason everybody sticks to DFI even nowadays.

 

[TAGG]

(like this one which wouldn't do above 3.7 with x13)

That explains a lot of things... this chip is really weird, time to run it at x12 next time i try

 

[Sparksnl]

Thing about BW chips is that they usually clock really bad overall or have a low (MHz wise) cold bug. Those I had which were capable of high core clocks would only hit around 300bclk on water while those with high BCLK would suck core clock wise. So at least judging from my observations one would have to be lucky to find a CPU capable of both, high BCLK and core clock.

For BW I have 2 3GHz (air) 3800+ cpu’s and 1 3GHz (air) 3500+. All have a good fsb above zero. Let’s see if they still have a high fsb sub zero. If they do have that, they are ln2 candidates.

 

[Raiden Zero]

CAB1E for 512kb cpu’s are rare. I have never seen one before or after. 3.45 is nice for a 3000+.

Just been searching the various platforms for some CAB1E when I remembered that CAB1E is mainly Opteron S940 chips. Might explain why your 3500+ clocked so much better than most of the other chips.
E.g. my best CAB1E Opteron 254 runs 3680MHz 1M at 1.6V and still might be CB limited.
So I guess it’s worth just buying any CAB1E 939 CPU out there.

 

[Sparksnl]

Yes they are rare and good. That is why I bought some 150’s as well.

 

[Raiden Zero]

Yes they are rare and good. That is why I bought some 150’s as well.

Rare indeed. No luck so far. Even using google there are only a few results.
A little further research and I came up with this:
AAB1E, CAB1E, CCB1E, ACB9E, CCB9E, LCB9E seem to be the steppings used for high end Opteron 940 CPUs. Given that every CPU I had on S940 was able to run 3.1G on water I think it’s safe to assume that looking for those (B1, B2, B9, BB) on 939 is the way to go. Especially for Venice as they might just be partially deactivated High-End Opterons which of course outperform any Venice by far.

 

[Sparksnl]

Rare indeed. No luck so far. Even using google there are only a few results.
A little further research and I came up with this:
AAB1E, CAB1E, CCB1E, ACB9E, CCB9E, LCB9E seem to be the steppings used for high end Opteron 940 CPUs. Given that every CPU I had on S940 was able to run 3.1G on water I think it’s safe to assume that looking for those (B1, B2, B9, BB) on 939 is the way to go. Especially for Venice as they might just be partially deactivated High-End Opterons which of course outperform any Venice by far.

Well that depends if you want to go sub zero…

 

[Raiden Zero]

Well that depends if you want to go sub zero…

Well that might be true. Interestingly KAB1E on 939 cold bugged at around 240MHz while CAB1E managed to hit 265MHz. BB on 940 did 285-290. Still those results of course vary in between CPUs.
So -10C on a very good stepping might still be superior to a shitty stepping with LN2 on high-Multi Venice CPUs.

 

[Sparksnl]

So my 2nd dr Expert lost some of its good looks. But at least now it works again.



Putting the caps through the holes was near impossible. At least for now.
Thanks @auto660

 

[Sparksnl]

Dr Expert nr1 full recap except original solids. Please note the soundcard recap…

I replaced those small 16V 100uF Nichicon for Rubycon.

Edit: I mis ordered those 16v 560uF so they have to do with 16v 220uF.

 

[Raiden Zero]

Nice. Will do the same with my CFX3200 once I finished playing around with it.

Did you test it already and if so is there any change in CPU/Mem overclocking?

 

[Sparksnl]

No, no time yet. This took longer then expected. Somehow they managed to ring at my door 4 times, call me at least 4 times and my cats were active also. So will be testing later.
I did post my nr2 board this morning. I recapped that one as well. At least… half of it. Since it is damaged anyway.

 

[Raiden Zero]

At least for memory OC I could imagine some positive effects on max clocks.
—————

Mal ne Frage an alle DFI Besitzer (idealerweise mit Debug Karten):
Wie verhalten sich eure Boards, wenn keine Grafikarte eingesteckt ist?
Mein vermeintlich defektes DFI bootet nämlich gem. Debug auf dem Board normal hoch und auch die Lüftergeschwindigkeit geht runter. Auf der Debug-Karte bleibts aber bei — —

Hab derzeit folgende Vermutungen:

1. Chipsatz hinüber, daher bootet das Board zwar gem. Board-LEDs, bleibt aber tot.

2. Trotz Recap is da eventuell ne Lötstelle unsauber gearbeitet oder irgendwas anderes im Argen.

Andere Ideen? Als Beep-Code kommt übrigens Lang, Kurz, Kurz. Sollte gem. Handbuch n VGA Problem sein.

 

[Sparksnl]

I actually have a motherboard (not dfi) where the debug card is not working like in your board. But motherboard works fine. I also have a motherboard (not dfi) which will not post with the debug card in the last pci slot.

Besides that: Could that part be defective which puts those debug signals on the pci slot?

 

[Raiden Zero]

I actually have a motherboard (not dfi) where the debug card is not working like in your board. But motherboard works fine. I also have a motherboard (not dfi) which will not post with the debug card in the last pci slot.

I have some boards which behave the same but on my DFI it never was a problem.

Besides that: Could that part be defective which puts those debug signals on the pci slot?

That’s what I suspect as well. Would explain why I’m getting no video signal and debug card doesn’t give me any codes as well.

1 long, 2 short:
„Indicates a video error has occurred and the BIOS cannot initialize the video screen to display any additional information.“

Might have to resolder all caps just to make sure.

Strange thing is tho: Only one LED „on“ is supposed to indicate that CPU, Mem and GPU were checked successfully and the board is loading bios.
I’ll just hot-flash the BIOS even tho I already have a second bios chip. Because that’s my third theory: bricked BIOS

 

[Sparksnl]

Did you try a pci and a pcie gfx card?

 

[Raiden Zero]

Did you try a pci and a pcie gfx card?

Only PCIe so far but I don’t think using PCI will change anything. Usually if no GPU is installed my debug card should still output all the codes instead of — —. That’s usually a sign of defective hardware (CPU, Chipset, Caps) or a bricked BIOS. Still it’s strange that the board seems to boot.

 

[Sparksnl]

I was just wondering. You never know

 

[Sparksnl]

New in. Strange batch BA. Let’s see what it can do:

 

[Sparksnl]

I am working on a sheet where I can collect my info for Coldbug info about batches. I was thinking of something like this. Would this be clear enough and understandable?

 

[Raiden Zero]

Seems to have everything one needs from my perspective.
I might be able to add some more detailed information to that once I get around to run my chiller. Might be some months as I first have to figure out how to run this thing below 20c (Alphacool Eiszeit) but it should be possible.
This way I could test some CPUs regarding their behavior under cold. E.g. max BCLK going from 25c to 0c water temperature in 5c steps or sth.

Still I think I will step away from testing CPUs for max BCLK under ambient after all. Too much effort and I haven’t really found any pattern that fits all batches/steppings.
Ill just test them for max clock 1M under ambient and if it’s any good go straight for subzero and test 1M+BCLK. That’s except for stuff like LCB9E which from my experience cold bugs at below 230 anyways.

 

[Sparksnl]

Well we all test on the way we like best. I an just trying to make some paterns so it becomes more clear on the long therm.
I was thinking of making a public Excel sheet dor this.

I am thinking of adding at least 2 columns: Coldbug (post) and Cold bug (boot).

 

[Raiden Zero]

Of course. Actually this way we can at least figure out different stuff instead of everybody testing for the same.
What I meant by this is that I haven’t been able to find any at least somewhat fixed BCLK offset going from ambient to -40c.
There is of course information on which steppings will suffer from cold bug in general but some BW CPUs will lose a lot of max BCLK when cold, some only a few MHz and one even showed improved oc-ability.

 

[Sparksnl]

Well then we can at least document that it differs. So we know this.

Edit: Oh and I retested more then 60 cpu’s so far for max fsb on air…

 

[Raiden Zero]

Auch interessant:
Habe gerade auf dem CFX3200 über eine Stunde lang versucht 300MHz+ mit meinen TCCD laufen zu lassen.
Am Ende hat sich herausgestellt, dass die CPU einfach keinen hohen BCLK machen wollte. Hab dann einfach mal auf gut Glück mit der Spannung gespielt. Und siehe da: mit über 1.5V will die CPU keine 300MHz machen, bei 1.45V waren dagegen 330MHz bei gleichen settings machbar. Hab nicht mal neu gebootet, sondern einfach per A64Info die VID um 0.05V gesenkt.

Sowas hatte ich bisher auch noch nicht. Merkwürdig.

Edit: 1:1 bekomm ich den Takt trotzdem nicht hin. Definitiv irgendeine Einstellung im bios. Für Max OC seh ich das CFX3200-DR definitiv nicht. Zu zickig, teilweise bootloops ohne Grund bei zuvor absolut stabilen settings, teilweise Unkalkulierbarkeit was Verhalten des Boards allgemein angeht. Da hätte DFI definitiv mehr Arbeit rein stecken müssen.