[Sammelthread] AMD K8/K9 - Sockel 939 | 940 | 754

[Sparksnl]

So I still got some results. Too bad my best 4 or 5 validation files were corrupt. So 3433MHz is best now (max validated was 3478MHz).
I made an in between screenprint:



Max boot was 3300MHz @1.67v. On air I needed 1.50v for a 2903MHz boot.

PI and PIFast:


Win 7 testing:


More important tests for me were max FSB on cold. I tested with -62C on my Single Stage head and -53C on the sensor next to the CPU.

Max FSB Boot with cpu watercooled was (LDT/FSB 1.20v / Chipset 1.62v):
LDT/FSB: 3.0x - <440MHz FSB
LDT/FSB: 2.5x - 460MHz FSB
LDT/FSB 2.0x - 460MHz FSB
LDT/FSB 1.5x - Not tested (but most of the time not that good)

Max FSB Boot with cpu at -62C was (LDT/FSB 1.20v / Chipset 1.62v):
LDT/FSB: 3.0x - 335MHz FSB
LDT/FSB: 2.5x - 325MHz FSB
LDT/FSB 2.0x - 340MHz FSB
LDT/FSB 1.5x - 345MHz FSB

So if I analyze these FSB results then to me it looks like this CPU can take a little more cold. 3600MHz would need a 300MHz FSB (multi 12x). That is a 45MHz gap (345 - 300).
And I guess I would not mind going a little further because even more cold could make it more stable at slower speeds. It would just top out earlier for max validated speed in CPUz.
Example: 3400MHz would need a 283MHz FSB. If I could bench all benches on that speed I guess I would be happy loosing a bit top speed. Then I would have a margin of 62MHz (345 - 283).
That way I could possibly give it 20C to 30C extra on cold (so -80C to -90C). In theory...

 

[Tzk]

Interesting that some LDT multiplier scale different to others. For example 2.5x scales way worse than 3 and 2... 3 even scales best ("looses the least") FSB when going cold. I wasn't expecting that.

 

[Sparksnl]

I was thinking that DFI or AMD might have programmed or added the lower ldt/fsb multi’s for cold. But that is guessing and maybe.

Edit: Now I think of it, 460/465MHz is possibly max FSB for my motherboard.

 

[Raiden Zero]

Example: 3400MHz would need a 283MHz FSB. If I could bench all benches on that speed I guess I would be happy loosing a bit top speed. Then I would have a margin of 62MHz (345 - 283).
That way I could possibly give it 20C to 30C extra on cold (so -80C to -90C). In theory...

Nice results. That’s exactly why I keep binning my CPUs for max. BCLK under single stage. Just in case I somehow find the time to actually bench some of my CPUs with LN2 anytime soon. This way it should be possible to bench at higher speeds even if max Valis is lower. That being said I’ve been trying to get some LN2 sessions going for one year now but that requires me to have like 2-3 days off just for benching.

Edit:
Did anybody go sub 20 for 1M on any K8 platform so far? According to HWbot 20.109 should be the WR for now. So this might be a nice goal as my 6400+ can run WPrime 1024M at 4.02G. 4.1+ is certainly possible given the right temperature and board.
Sub 30s PiFast would be nice as well.

 

[Sparksnl]

Well it depends on how much ln2 you have, but 2 sessions should be possible with 30L or so.

By the way: I noticed max fsb on cold with Athlon XP as well a few years ago.

 

[McZonk]

Sounds like there might be the need for a retro LN2-Session with some ambitious goals for a few records?

 

[Sparksnl]

That would be nice. But if we want to have some records I would like it that Stunned would go out of (sub zero) retirement for a weekend or so. 😉
Somehow I think he still has some hardware which would make a chance for that goal.

 

[Sparksnl]

I managed to buy a 3700+. It is a CABGE 0516 WPAW. A very early sample for a 3700+ which is from the 0516 batch.
My only other BG is expected to be strong for sub zero (very max high fsb on air).

Any expectations for such an early sample?

 

[stunned_guy]

Ich hatte mit den ganz frühen Exemplaren keinen Erfolg, auch unter Extremkühlung nicht. Der ursprüngliche Weltrekord bei den 3700+ war aber mit einem 0515 oder 0516 von "starostice" gemacht worden. Kann also nicht so schlecht sein.
Bei dir oben fehlt ein Buchstabe beim Stepping: soll sicherlich "CABGE" heißen.
Kälte verträgt CABGE gut, auch hoher Referenztakt ist drin.

 

[Sparksnl]

Yes indeed it is CABGE. I will see what it does then.

 

[Raiden Zero]

Da ich momentan echt die Sch**** am Schuh kleben hab, mein DFI NF4 bei der letzten Bench-Session gestorben ist (recap hilft nicht, es bleibt bei - - - - debug led und beep code 1xlang, 2xkurz) und keins zum akzeptablen Preis bei Ebay drin ist:

Hat jemand ne Ahnung, ob das Asus A8N32 den RAM über 3.3V oder 5V versorgt und hat zudem die nötigen Mods zur Hand?

Hab auf die Schnelle nichts gefunden.

 

[Sparksnl]

Too bad the dfi nf4 died.
I actually modded my a8n32sli. But just slightly, I removed the memory mosfet so I could use FireRAM. I can’t remember how much voltage it has, but since I did remove the mosfet it probably is 5V. Because: FireRAM can deal ok with 3.3v but won’t with 5v coming from the board as well. Besides that most nf4 boards use 5v as far as I know.

Edit: You can measure here:



Edit 2: As you can see this mod is final for this board... (oops)

My Dr Eperts (I have 2) will be getting a recap soon. Caps are in so the preparing phase is now. After this they will get more cold…

 

[Raiden Zero]

Thanks. In that case I might get myself a A8N32 just to see how it performs while I wait for a reasonable DFI for sale.

Still have to give the elmor EVC2 a try so this might be a good candidate to use it. Easiest way to voltmod those old boards by far.

Plus I’ve got a lot of CPUs I’ve extensively tested on my DFI so comparing those scores to what I can achieve on the Asus night be interesting.

 

[Sparksnl]

I always thought that the Asus is the best board out there. Just missing bios optimization and max voltages.
If it is tweaked it probably outperforms everything out there.

 

[Sparksnl]

I was just trying to replace some caps on my dr expert. While I am at it I thought I would also replace the 2.5v cpu caps.

For those thinking of it: absolutely not recommended!
Even with 480C after 15min I still only have 1 leg out of the first cap and the other broke of and still won’t move. This thing takes heath as easy as a modern board.

 

[digitalbath]

Had a simmilar situation (other caps though). Lead based solder and a bigger tip helped me in my situation. Good luck!

 

[Sparksnl]

I gave up, one board less to use. It is impossible.

 

[Raiden Zero]

Just send it over here. At least I can’t break what is already broken.

 

[Sparksnl]

I think I will try and heath up (with a brenner @160C) the board/parts surrounding the caps and then try and desolder it.

 

[Tzk]

Yes, try to preheat the board. How much power does your soldering station have? I also had massive issues with my 50W Weller soldering station, but my ZD-915 desoldering gun (80W) worked great. Usually it's not the temperature of the iron which is the issue, but it's the power. It can't transfer heat fast enough to get the solder into liquid state if power is insufficient.

A heatgun or even preheating it in your kitchen oven might work. Just do not overheat or burn the board. And if this doesn't work for you, please stop before killing the board. Someone in here (including me) might be able to help you out with that recap.

 

[Sparksnl]

Well yes I stopped before going to far. At least I think so.
I just looked at the soldering station, it is 48W. So that might not be enough. I will look around for a different one then.
I will probably go to Auto660 if I can not complete it myself. Lucky for me I took the reserve board first.

Edit: thinking of buying an Ersa rds80

 

[WMDK]

Well yes I stopped before going to far. At least I think so.

That is why I always say it's all fun and easy like all those YouTubers claim ignorantly until you come across a board / gpu loaded with copper and you're lacking proper equipment and practice. Already had multiple boards & gpus sent to me for fixing after the owners saw only excessive heat as a last resort and failed nevertheless. You wouldn't believe the amount of abuse some of these had to take in the attempt to get the caps off. Some were burned several layers deep, eliminating everything from the pads to the traces. Some even found mechanical force as a valid tool and tried to pull out the soldered legs after they had ripped off the can because it just didn't want to come out at all. Sometimes things were fixable with a good amount of reconstruction, sometimes all hope was lost especially when heavy tools like drills come into play and the art of correct operation is in line with the rest of the skillset But most of the time it's only the tip of the iceberg anyway, because those who do such things do not shy away from other cruelties starting by the cap selection...



As funny as it may sound, it often breaks my retro heart seeing hardware treated like that and sometimes being lost forever.

So it's always a good idea to rather stop as you did than to risk something with a rare piece of hardware

Also, I found the advise of preheating is something to be used cautiously as well or at least to describe the process in much greater detail than just throwing in the word "preheat". It may sound harsh but sometimes people seem to lack even the fundamental understanding of what is going on and see no problem in pointing a 2.000 watts heat gun at the pcb for minutes, causing all the previously easily soldered in alu-electrolytes in adjacent areas to the VRM section to gas out and bulge even before the first power-up I've seen everything from things like that to dozens of accidentally fallen off smd parts to permanent pcb burn marks with bubbles forming.

No offense though, I'm not pointing in your direction with this at all, just wanted to point out generally that in my opinion far too many people kill hardware for nothing by just using the wrong tools or simply lacking the skills because every YouTuber tells them it's all as easy as can be and they fail to realize that there was a point where they'd rather stopped than to push on by all means.

So my advice is, even if you get a better soldering station with more power, !!please!! practice on other heat-eating-nightmare-boards before turning to the DFI – it will pay off big time.
AND always remember to have the right soldering tips for what you're trying to do. There's no point in having 80 watts of power when you only have the standard very thin SMD pointy-tip on your iron which has basically no surface area to transfer all that heat in a reasonable way.

Looking forward to see that beauty of yours back to life with all new caps equipped

 

[Sparksnl]

Thanks for the reply.
Those caps are soldered on a way I never saw before. Wow.
I actually did have some soldering lessons back then on school. That is why some things came back fast. Normally replacing caps is no issue. SMD is a little harder for me.
I used the heather 1 time before now. That went well. Just use it carefully and do not heat that long or all parts.

I used a Velleman solderstation up to this point. Obviously it was not enough.

I now ordered an Ersa I-Con 1 and some tips to start with. Let’s see which tip I like before ordering more. I always find it hard to choose from a picture.

Station


Tips


Edit: I was thinking of a better station the first time, this time I did order better. Not taking anything less anymore.

 

[Sparksnl]

I found this Newcastle (CG) 3500+. I am not expecting that much of it, but I don’t see them too much so now I have one.

 

[Sparksnl]

I found myself an Athlon 64 3500+ of the BW batch as well (LBBWE 0619FAE). It can boot 3GHz @ 1.55v. But is has a nice max FSB on air as well.

490MHz FSB boot and validated.


495MHz FSB boot, no validation.

I did not try to make a max validation by setting FSB in Windows XP. So it might validate between 490-495MHz
I am expecting a really good Cold Bug for this one. -60C on my Single Stage should be easy. On LN2 it might even beat my golden air 3500+ (3630MHz on air).

Edit: Thinking of it. When raising CPU voltage (more heat) it might even do 500MHz FSB.

Edit: It ran 2.5x 495 (1237MHz HTT)

 

[Raiden Zero]

Which batch was your golden 3500+? Wondering because even compared to the best Venice CPUs I had (3G+ 24/7) that’s like +300MHz better.

The 0619 one looks very promising. Should be on par with the best Venice CPUs I had recently but those were only capable of running 3.2X on cold due to low max BCLK.

 

[Sparksnl]

That golden is a CAB1E 0621EPMW.
3630MHz is max validated. All my subs for 3500+ were made with that golden cpu.

Edit: I actually have another Ok air 3500+. It is CAB2E and can boot 3.2 on air.

 

[Raiden Zero]

Thanks. One more batch to look out for.
I already figured. Best Venice I had was capable of 3.15G on air iirc. But that was running 1.6V or so.

Still haven’t found one that would beat my 3000+ 0519WPMW. That one passed 3.45G on my modded Prometeia back then.

 

[Sparksnl]

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

 

[Strunkenbold]

My golden 3200+ Venice is also CAB1E but 0612EPMW. 3.2 Ghz 1,5V on Air.