[Sammelthread] Sockel 939 / 940 / 754 - Thread

[funkflix]

Zeig es uns!

 

[Ben Psyko]

@stunned_guy

Is this the earliest sample there is or did you see earlier weeks as well?

Interesting batch as well.

Its not a 3700+, but a 3400+ 0432, so way earlier.

 

[bschicht86]

Ich glaube, er fragte nach dem frühesten Datum für San Diego.

 

[Ben Psyko]

Ich wusste die Sache hat einen Haken 😄

Kann dann noch einen 4000+ , sollte ja SanDiego sein, aus 0505 zeigen.

 

[Sparksnl]

Yes I meant San Diego. I actually have 2003 A64’s…
That 0505 cpu is nice. I would love to see results of that cpu.

 

[funkflix]

Die CPU sieht ziemlich misshandelt aus, hoffe die lebt noch... @Ben Psyko

 

[Antinomy]

but a 3400+

Interesting sample. Despite being socket 939, it has HT 800MHz (4x multi) and it's an OEM part. Speaking of old, I have one of the oldest A64

P.S. does anyone have early samples that have a serial number starting with "F" or "RFB"? Photo would be very helpful.

 

[bschicht86]

Speaking of old, I have one of the oldest A64

When was it born?

 

[Sparksnl]

Interesting sample. Despite being socket 939, it has HT 800MHz (4x multi) and it's an OEM part. Speaking of old, I have one of the oldest A64

P.S. does anyone have early samples that have a serial number starting with "F" or "RFB"? Photo would be very helpful.

I have 2 starting with an F.
- FX-55 Clawhammer (AAA2C), F485562I40073. Week: 0437.
- FX-60 F49899C60803. Week 0611

The oldest cpu I have is an FX51, week 0342. Numer starts with a 9.
I have 2 ES cpu’s with a number starting with 100.
- FX-57, week 0516.
- A64 4400+, week 0517.

No photo’s for now.

 

[jumpel]

...
The oldest cpu I have is an FX51, week 0342. ...

Sure? The FX are this old? I'm surprised cause for me late 2003 is the golden age of Athlon XP / Socket 462 and not newer platforms like 754 oder 939!

 

[bschicht86]

My oldest CPU:
3200+ AW CBAEC 0426 Serial: 101...
3400+ AZ CBAEC 0430 Serial: 101...
3800+ AW CBAEC 0427 Serial: 101...
FC53 AS AAABC 0409 Serial F44...

Sure? The FX are this old?

Sockel 940

 

[Sparksnl]

I have 2 starting with an F.
- FX-55 Clawhammer (AAA2C), F485562I40073. Week: 0437.
- FX-60 F49899C60803. Week 0611

The oldest cpu I have is an FX51, week 0342. Numer starts with a 9.
I have 2 ES cpu’s with a number starting with 100.
- FX-57, week 0516.
- A64 4400+, week 0517.

No photo’s for now.

Athlon 64 4400+



AMD FX-51 (week 0342) / FX-53



AMD FX-55



AMD FX-57



FX-60

 

[Antinomy]

When was it born?

2002 From what I've seen 0248 is the oldest Athlon marking. Anything before had ES markings only.

- A64 4400+, week 0517.

Why do you think it's ES? Because it doesn't have X2 marking on it?
Thank you for the info! Looks like all 754/939/940 starting with F.. are pre-release samples. Maybe the ones with 100... too.

FC53 AS AAABC 0409 Serial F44...

Seems like pre-release too.

 

[Der Namenlose]

I don't think Sparks X2 4400+ is an ES, stunned_guy once mentioned that ES CPUs show duplicate letters for the first two characters, that would fit the FX-57 but not the 4400+.
Additionally, things like the "1.40V" voltage marking on the FX-57 to me would be the most notable indicator for a sample chip besides the stepping starting with AA.
The missing X2 designation was probably not uncommon during the early X2 weeks. My X2 4800+ for example is missing the X2 marking as well and I see chips like that every now and then.

Oh, and I think I can provide an example for a 939 CPU starting with F that's not a pre-release / early week version:

 

[Antinomy]

Oh, and I think I can provide an example for a 939 CPU starting with F that's not a pre-release / early week version:

Yes, seems that dual-cores might not follow this but for earlier models it might be true. For now, I'm just collecting data.

 

[jumpel]

... AMD FX-51 (week 0342) / FX-53 ...

Interesting to see that they were made with ceramic base like the K7 Thunderbird :]

 

[Sparksnl]

2002 From what I've seen 0248 is the oldest Athlon marking. Anything before had ES markings only.

Why do you think it's ES? Because it doesn't have X2 marking on it?
Thank you for the info! Looks like all 754/939/940 starting with F.. are pre-release samples. Maybe the ones with 100... too.


Seems like pre-release too.

Because of the AA and relatively early week.
All early samples did indeed have no X2 on it.
Oh and starting with 100 indeed.

 

[Antinomy]

Because of the AA and relatively early week.

Do you have a description of how to decipher stepping codes (what they mean)? And the part after the date too. I've seen only one source so far.

Also I've made a modBIOS for AsRock K8A780LM for benching purposes.

 

[Sparksnl]

No description so far here. Just guesses and what I think.

 

[Antinomy]

The only info I've found is this one: https://www.overclock.net/threads/o...-overclockable-stepping-averages-info.128767/
probably you've seen it.

 

[Sparksnl]

The only info I've found is this one: https://www.overclock.net/threads/o...-overclockable-stepping-averages-info.128767/
probably you've seen it.

I was looking at the old info of Stunned mainly.

 

[Masterchief79]

@Tzk Ich mach mal hier weiter mit den Ergebnissen.
Mittlerweile habe ich die Corsair Pros auf 310MHz durch SuperPi 8M bekommen. Allerdings mit sehr entschärften Timings, 2T Command Rate und 32M ist und bleibt instabil.

Ich teste jetzt nochmal die UltraX dagegen, und wenn ich dann noch Lust habe, eine andere CPU/IMC.

Aktuell sieht es so aus, als würde TCCD OC auf dem Epox schon ganz gut funktionieren, aber man muss im Vergleich zum DFI NF4 jedes Subtiming und jedes Setting einzeln einstellen. Mit Skews und Co habe ich auch noch nicht groß rumgetestet, irgendwo muss Schluss sein.
Oder die Sticks sind einfach nicht die besten, das kann auch sein. Aber die Lücke zu den Ergebnissen im NF4 Thread ist schon deutlich, deswegen tippe ich erstmal aufs Board.

Die UltraX Winbonds muss ich auch noch auf dem Epox testen, das dürfte aber deutlich schneller gehen und da habe ich Vergleichswerte durch die letzten 3-4 Pärchen. Editiere ich dann hier.

Edit: Hier der Winbond-Vergleich. Alle BH5, außer ganz links die Geil UltraX400 mit UTTs.

 

[Tzk]

Ich würde generell die Versuche auf 1T beschränken, außer es geht im Max. MHz Validierungen. Du wirst für hohen Takt alle Timings händisch einstellen müssen, daran führt kein Weg vorbei. Die meisten Timings haben Fenster in denen sie funktionieren, sprich was bei 200Mhz passt muss nicht bei 250 oder 270 funktionieren. Du musst dich also langsam vorarbeiten und Schritt für Schritt anpassen bis du den Zieltakt stabil hast.

Bei den BH gibt’s ein paar Spezialitäten, weil manche Riegel ab 3.4V kaum noch skalieren, andere Riegel mögen kleine TRC und TRFC Werte nicht. Wieder andere streiken bei kleiner TRAS und dann gibt’s natürlich Kombinationen daraus. Generell sind 3.2V nicht viel für BH, sehr gute Riegel machen 250Mhz stabil. Der Schnitt sollte aber zwischen 240 und 250 liegen. 512er Riegel laufen wegen der Menge an Chips tendenziell schlechter als 256er, du hast wegen Bank Interleaving aber mehr Performance.

Aus diesen Gründen habe ich meine BH Riegel auf Sockel A mit niedrigen Timings (trc 9, TRFC 12) und 3.3V gebinnt. Das langt um die guten Sticks zu finden, die niedrige Timings vertragen. Teils hatte ich sticks die damit bei 230Mhz nicht stabil waren, mit 13/15 aber locker 250 liefen.

 

[Sparksnl]

I always thought that the 3700+ s754 score was not that good.

AMD Athlon 64 3700+ @ 3482.32 MHz - CPU-Z VALIDATOR

[nnng8h] Validated Dump by (2025-03-09 17:51:43) - MB: DFI LP NF3 250GB - RAM: 512 MB

valid.x86.fr

And this score could have been 200MHz better if I did not have negative scaling from -100C and on.