[Sammelthread] Sockel 939 / 940 / 754 - Thread

[Sparksnl]

For reaching high fsb I use a 5x multi and set ldt/fsb at 2.5 or 2.0x. Besides that the cpu indeed needs some more voltage at a high FSB. But 300MHz should be possible for most cpu’s. Which batch is it?

 

[Raiden Zero]

Actually it’s not the CPU. Tested multiple CPUs and looks like the board won’t do anything above 300. 300 is very shaky and 305 won’t run 1M.
I already tested different settings. Not sure how I was able to run 330+ once but now I can’t anymore. Certainly some kind of bios issue.

 

[Sparksnl]

Does raising chipset voltage help??

 

[Raiden Zero]

Nope. But CPU HT Bus Drive Strength does.
Managed to run 330 with „Strong“ setting. That’s at least something and enough to test TCCD 1:1.
Running 300 1:1 is still not possible tho.

Will switch to my Asrock 939A785 later on anyways as I see the CFX3200 more as a collector’s item.

 

[Sparksnl]

Well then it is about the same as with my Striker II Extreme. Not good enough with fsb but rare.

 

[digitalbath]

So, ich habe bei meinem DFI NF4 SLI-D die BIOS Einstellung "DRAM Response Time" mit verschiedenen BIOS Versionen getestet.
Ich habe diese BIOS Einstellungen im BIOS verstellt und dann anschließend in Windows die PCI register verstellt. Es werden dabei immer nur die Einstellungen für "Read Preamble" und "Max Async Latency" verstellt. Deswegen ist diese Einstellung im BIOS auch zwischen den beiden Einstellungen. Ich denke mal, es dient sozusagen als eine Art "preset".

Bei den BIOS Versionen habe ich Wu's 704er Versionen benutzt. Das kam bei den Einstellungen heraus:

DRAM Response Time
==================

BIOS: N4D702-1 "all kind of RAM"
==================
Normal  ---> Read Preamble = 5,0ns ---> Max Async Latency 6ns
Fast    ---> Read Preamble = 4,5ns ---> Max Async Latency 5ns
Slow    ---> Read Preamble = 5,5ns ---> Max Async Latency 7ns
Fastest ---> Read Preamble = 4,5ns ---> Max Async Latency 5ns


BIOS: N4D702-2 "Winbond RAM Table"
==================
Normal  ---> Read Preamble = 5,0ns ---> Max Async Latency 6ns
Fast    ---> Read Preamble = 4,5ns ---> Max Async Latency 5ns
Slow    ---> Read Preamble = 5,5ns ---> Max Async Latency 7ns
Fastest ---> Read Preamble = 4,0ns ---> Max Async Latency 4ns


BIOS: N4D702-3 "TCCD RAM table"
==================
Normal  ---> Read Preamble = 5,5ns ---> Max Async Latency 6ns
Fast    ---> Read Preamble = 5,0ns ---> Max Async Latency 5ns
Slow    ---> Read Preamble = 6,0ns ---> Max Async Latency 7ns
Fastest ---> Read Preamble = 4,5ns ---> Max Async Latency 5ns

Die Einstellung "Fastest" war meist eine harte Nuss. Ich musste bis auf bei der -3er Version meine BH-5 Riegel wählen und diese auf 3V prügeln, um damit ins Windows zu kommen.

Falls das hier bekannt ist, kann der Post ignoriert werden.

edit:
Die Einstellung slow müsste in etwa den default Werten für 200MHz entsprechen:

 

[Tzk]

Hast du getestet was passiert wenn man die beiden Werte händisch setzt und trotzdem die Response Time? Also überschreibt das manuelle setzen das Preset? Da muss ich tatsächlich mal schauen, ob es diese Einstellung am DFI NF3 auch gibt. Dort habe ich massive Probleme, sobald ich Werte manuell setze (kein POST) oder sobald ich mich 280Mhz mit TCCD nähere.

 

[Raiden Zero]

Sollte nicht überschrieben werden. Aufm Ultra-D is „Fast“ preset und trotzdem bootet das System mit den BIOS settings 5.5/6ns

 

[Raiden Zero]

CAB1E for 512kb cpu’s are rare. I have never seen one before or after.

Talking about rare…
I ordered a couple of 3500+ and got:
3x CAB1E 0629DPMW

Also 3x 3000+ LBBWE 0540WPMW and 3x 3200+ (LBBWE 0536APIW, 0550FPPW, 0535EPGW)
Will test those CAB1E real quick.

First one does 3.2G 1.5V watercooled. I feel like I really need to get a new DFI NF4

Edit: 2nd CAB1E can do 3.27G 1M 1.47V, one more to go. 1.54V it manages 3.3G+ but I feel like I’m BCLK limited in my current board. Might swap to a different one later.

Third one does 3.28G. Next step is to delid one and see how it scales. My best CPUs gained up to 100MHz.

 

[Sparksnl]

Those cab1e are nice

Dr Expert nr 2 cap restoration in progress. My new Weller does well.

Having good tools is everything.

Edit: I did the worst ones first

 

[Raiden Zero]

Looking good.
Unfortunately I don’t think they will perform well with phase change cooling. But at least I’m slowly getting more BCLK on CFX3200. Managed 345 with one of the 3500+. But that’s like completely random it seems. Sometimes the board won’t boot 310 then again it does 330 just fine.

 

[Sparksnl]

That is strange.
Yes those CAB1E are not known for there good sub zero clocking. Max -10C I would say.

 

[Raiden Zero]

Just tried to boot 300 at -42c. No chance.
Might try my chiller next month or so. I think 3.5+ is certainly possible when staying within the correct temperature corridor. Right now I’m kinda RAM limited as running 9:10 isn’t possible (275+) but at 5:6 memclock is too low to get any good times in 1M.
Best was 25.6s running 3.35GHz.

3800+ CAB1E might also be fun if there any.

 

[Sparksnl]

You culd try and put an extra alu plate between cpu and cooler.

 

[Raiden Zero]

Tried that on other CPUs. Never really seemed to make much of a difference.
Still need to test for max BCLK (cold bug) but i also suspect something like -10c to be the sweetspot for those CPUs.

 

[Sparksnl]

Board 2 working with replaced nb and cpu caps.



Now waiting for the other caps to be replaced.

 

[Raiden Zero]

Looking good. Any tests so far regarding better OC?

Did some quick tests with my CAB1E 3500+ today. Managed to get it to run at 10C and -5C core. -5 is too cold, won’t run > 285, while 10c is too warm or at least there should be some more headroom. Managed to run 1M at 3450 this way. -30C core max. was 268.
Seems like cold bugs don’t really scale linear with cold. Considering that going from 10c to -5c (15c delta) made the cpu lose 35MHz baseclock while the next jump from -5 to -30c (25c delta) only resulted in 17Mhz less baseclock.

Also got some more Opterons. 275s this time. B1, BB and BW. Haven’t had any BW Opterons so far so this might be interesting. B1 and BB most likely won’t be any good for cold.

 

[Sparksnl]

So my 3500 B1 has probably the same CB temp then.
No I just quick tested it. I do not have much time in evenings this week and tomorrow morning is out as well. I was glad to be able to quick test it running CB 11.5. That was it.
And… I probably will be replacing cpu and nb caps on my nr1 board first.

 

[Raiden Zero]

Most likely, yes. I’ve had some B1 which did worse.
My Opteron 275 B1 won’t even run 250 but that CPU is delidded which resulted in 10C lower temps. So this might be the reason even tho core temp reports only -18C which isn’t too low tbh.

Delidding resulted in up to 20c lower temps btw.

For 940 dualcore OC it’s still the same: Good CPUs (mostly B1/B2) will OC pretty much as good as shitty ones (BW) running phase change.
B1 does 312x watercooled, BW 313x with single stage.

 

[Sparksnl]

I just tested my Nr.1 Dr. Expert after recapping CPU and NB part. I looks like it's CPU part is just a little more stable for my 3500+ when doing boot tests (max 1.47v was tested).

Edit: After testing some more: NB OC results seem to do way better on my 3500+ (I hit 410MHz at ldt/fsb 2.5x multi before and now I reach 445MHz).

 

[Raiden Zero]

Nachdem ich mit S939 nicht sonderlich weiter gekommen bin hab ich nochmal auf das K8N-DL umgebaut und meinen Opteron 254 ein wenig getreten:

AMD Opteron 254 (2P) @ 3846.22 MHz - CPU-Z VALIDATOR

[114v9p] Validated Dump by raiden (2023-04-24 22:31:52) - MB: Asus K8N-DL - RAM: 620 MB

valid.x86.fr

Viel mehr wird nicht gehen, da bei knapp unter 290BCLK der cold bug wartet.

Und damit sich @TAGG - sollte das mit den AT-6 ECC Riegeln funktionieren - noch ein wenig strecken muss:



Erste S940 CPU unter 23s 1M



Hab sicher 2h erstmal am Windows arbeiten müssen, weil die Zeiten so gar nicht in dem Bereich waren, was gehen sollte. Aber am Ende passt 1M schon ganz gut. PiFast hab ich nur einmal zum Testen laufen lassen. Da dürfte noch ein wenig mehr gehen.

 

[Sparksnl]

Nice clocks! Which batch is this?
Edit: giving it more cpu voltage or choosing a different ldt/fsb multi might give some more MHz for FSB. From what I see 2.0x was chosen. While running subzero 1.5x might do well.

 

[Raiden Zero]

CAB1E 0601RPMW
Got two of them. Sadly the other one runs in some kind of wall at around 3.4G. Not cold bug related but core clock.

I tried all settings. Didn’t really make a difference. For 1M it’s actually the maximum CPU and RAM will do. Worst DIMM does 286MHz 1M and CPU won’t go 10MHz higher. So this is as good as it gets without running Winbond.

 

[TAGG]

Und damit sich @TAGG - sollte das mit den AT-6 ECC Riegeln funktionieren - noch ein wenig strecken muss:



Erste S940 CPU unter 23s 1M

Nice Sind dass meine DE-5 oder 4 sticks SR CC-6/CF-5?

 

[Sparksnl]

At which temps was this run?
I might run my 154 as well now mu board is recapped. It was promising on air.

Edit: Looks like I have one 254 from 0601 as well…

 

[Raiden Zero]

Nice Sind dass meine DE-5 oder 4 sticks SR CC-6/CF-5?

Das sind die CF-5, die ich von dir hab. Also 4x512MB SR. Vier Module gehen aber quasi genauso, wie zwei. Von daher is das recht stressfrei.

@Sparksnl -44c on evaporator. Software reports -49
CPU is delidded so temps are really good but takes quite some remounting everytime I install the CPU. Sometimes it won’t even hit 3650 1M when mounting is bad.

As long you don’t run into CB trouble delidding is usually worth it. Worst I’ve seen was 10C decrease in temps. Best was 25c.

If only those 256 Opterons weren’t that expensive. I bet there are some 4GHz valid CPUs out there.

 

[Sparksnl]

@Sparksnl -44c on evaporator. Software reports -49
CPU is delidded so temps are really good but takes quite some remounting everytime I install the CPU. Sometimes it won’t even hit 3650 1M when mounting is bad.

As long you don’t run into CB trouble delidding is usually worth it. Worst I’ve seen was 10C decrease in temps. Best was 25c.

If only those 256 Opterons weren’t that expensive. I bet there are some 4GHz valid CPUs out there.

Thanks,

I just ordered 2x Opteron 256… I sure hope I find a motherboard someday.

 

[Der Namenlose]

I just ordered 2x Opteron 256… I sure hope I find a motherboard someday.

So I guess you ordered the remaining 2 from cbo-do - curious to see if your batches are mismatched as well.

 

[Sparksnl]

I just thought of it, it are 3 256’s.
I will post the batches, we will see then.

 

[Raiden Zero]

Nachdem ich im Marktplatz-Thread ja schon den Verdacht geäußert hatte, dass CAB1E Venice vermutlich eigentlich Opterons mit teil-deaktiviertem Cache sind:



Von links nach rechts:
Opteron 280 - CAB1E Venice - NBBWE Venice

Hab jetzt gerade keinen geköpften 1MB Singlecore zur Hand, die sehen aber genauso wie der CAB1E aus.

Wobei das grundsätzlich ja klar war, da CAB1E immer 1MB Cache bedeutet.

Edit: Hab gerade mal geschaut. 3800+ gibts auch als CAB1E. Das könnte richtig interessant sein, wenn’s bis ca. 310BCLK bei passender Temp geht.

Und ich hab auch noch nen interessanten 3700+ im Zulauf. 0630CPMW