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

[Raiden Zero]

I've written an article about all those blanks in the Athlon 64 we had so far. Enjoy it if you like the AMD K8 architecture.

Hammer time

AMD: “It’s Hammer time” Athlon 64 codename “Hammer” became an architectural design revolution. From a single core architecture bonded by a front side bus to a northbridge which implemented a memory controller and southbridge for peripherals the paradigm has shifted to a multi-core design with an ...

docs.google.com

Great read. There is one more unicorn for Venice E6: former 940 Opteron Troy core CPUs with half of its cache being disabled.
At least that’s it what I suspect them to be.

raiden`s SuperPi - 1M Score 25 sec 735 ms with Athlon 64 3500+ (Venice) @ 3324.00

hwbot.org

This one for example is CAB1E 0629BPMW.
Core is definitely 1MB single core, I already posted a picture of it‘s delidded die somewhere over here. I actually suspect all CAB1E (754/939 San Diego) are actually 940 Opteron dies.

 

[WinstonChurchill]

Gigabyte KN8NS nForce3 250 (5€)
AMD64 3000+ Newcastle 130nm S754 (2€)

Max CAS2.5 2.7v 290FSB

 

[Sparksnl]

Great read. There is one more unicorn for Venice E6: former 940 Opteron Troy core CPUs with half of its cache being disabled.
At least that’s it what I suspect them to be.

raiden`s SuperPi - 1M Score 25 sec 735 ms with Athlon 64 3500+ (Venice) @ 3324.00

hwbot.org

This one for example is CAB1E 0629BPMW.
Core is definitely 1MB single core, I already posted a picture of it‘s delidded die somewhere over here. I actually suspect all CAB1E (754/939 San Diego) are actually 940 Opteron dies.

I just wonder about 1 thing. Troy is E4 where Venice is E6.
How would that be explained?

Edit:
While packaging today I found an Asus A8N SLI Deluxe. I am not sure if it is fully working (usb ports), I need to test that.
I might sell it or give it away later.

 

[Antinomy]

Great read. There is one more unicorn for Venice E6: former 940 Opteron Troy core CPUs with half of its cache being disabled.

Thank you. Yes, the CAB1E/CAB2E are an interesting case for overclocking. But since AMD decided to make then seem as normal Venice CPUs, it's almost impossible to detect them by software. I think we can identify them by K8 manufacturing ID (CPU-Z txt report shows this value) but I didn't have enough data to include this in the article. So far, all normal Venice s939 seem to have K8 ID 4.2 whereas San Diego and "Venice" CAB1E/CAB2E have 4.1.
Although, I have some info related to special steppings which I will add soon.

I just wonder about 1 thing. Troy is E4 where Venice is E6.

As your understand, AMD can write anything to CPUID registers (and not only them). Maybe the software difference between late E4 dies and E6 Venice isn't crucial so they disguized E4 core to look like E6 not to make too many OPNs (like 3500+ San Diego-512 which is typically the same as Venice 3500+ CAB1E/CAB2E). Also don't forget the Newark and Lancaster CPUs which are E5 and use CAB1E/CAB2E silicon too. So the difference might be not so strict as with single-core Toledos.
Some vendors had to make BIOS updates to fix 3700+/4000+ CF (single core Toledo) support but didn't see anything like this for "Venice" CAB1E/CAB2E.

 

[Sparksnl]

Did I share this?

wPrime - 32mOverclocking Records @ HWBOT

hwbot.org

WPrime 32M record for s 939.

 

[Sparksnl]

So I found myself an 3800+ Venice (E6, NBBWE 0638BAD). So far this is the best (BAD ass) OC 3800+ I have, at least on air. For a BW that is.
It can boot XP at 3001MHz and 1.45v (1.424v). That is a lot better then my best 2 chips (which max at 3787MHz and 3768MHz)...

 

[funkflix]

Good luck with it!

 

[Sparksnl]

I still don’t have much time because of moving houses. But this one I will bench.

 

[Sparksnl]

Oh and by the way, when looking for a good ln2 3800+. The NBBWE 0638 (X117) batch seems golden for all I have seen. All 3 I have are real good and the one from @I.nfraR.ed is good as well (2nd on HWbot).

 

[Sparksnl]

0638 BW Batch 3096MHz boot at 1.55v (no PI 1M).

I have never seen a BW boot this high on water. Max FSB on air was 485MHz, so more then enough play room on cold.

 

[bschicht86]

So I found myself an 3800+ Venice (E6, NBBWE 0638BAD)

I found a LBBWE 0638CAA X117...

Can that be the same or are there to much differences?

 

[Raiden Zero]

I found a LBBWE 0638CAA X117...

Can that be the same or are there to much differences?

Sollte nach meiner Erfahrung ähnlich laufen. Produktionswochen bei gleichem Kern sind eigentlich immer ein recht guter Indikator.

 

[Sparksnl]

I found a LBBWE 0638CAA X117...

Can that be the same or are there to much differences?

I never found a 0638 before but all 3 which I have start with X117.

 

[Sparksnl]

I noticed that the DFI NF4 dr. Expert comes with a Reatek ALC850 card by default. The soundccard of my 3rd board is a bit dirty and damaged. Now I knew I had a spare one. I did not remember having the ALC 882 card as a spare as well and since it is a 2005 module this is interesting.
Any idea if this would work out of the box? Since we can swap them and it fits fine…

 

[Sparksnl]

More bad news... It looks like that my go to Retro Dr. Expert is having a defective PCIe x16 slot. And that is not wanted for an SLI setup...
Not sure why it worked the first time I posted it before recappig, but the damage is visible.

 

[bschicht86]

Have you a picture? Are pins bend? If this, you can desolder the slot and grab some pins from a socket that come from the same company like your slot.

 

[Masterchief79]

Not sure why it worked the first time I posted it before recappig, but the damage is visible.

Post a picture of the damage pls

 

[Sparksnl]

i noticed some black pins. Or at least darker. I knew the yellow pcie was painted orange on some parts, but this I missed.
Boot code is 45/54. I will test the other slot soon,
I can make a photo and post it, I am a bit busy this weekend.
There might be a mosfet or cap that is burned.

If the other slot works fine I can make it my ln2 bench board as well. The current ln2 board is in a 100% working and optical state.

 

[Sparksnl]

So I checked, the 2nd PCIe x16 slot will boot the board. I need to check what the issue is with the 1st PCIx16 but that will probably have to wait because I cut my finger (deep). I am (almost) 2 handed, but still fixing things needs 2 hands.

 

[Tzk]

It's the same on my Ultra-D. First PCIe is dead and second does only work when i move the SLI jumpers. I suspect the usual solder issue or the chipset is dying. My board works perfectly fine with a PCI gpu. As i mainly use it for ram testing, i couldn't care less...

 

[Sparksnl]

There is more to it here, see it's pictures. But indeed, it could be my bench board.

 

[Raiden Zero]

Iirc I had that issue on my first DFI. Was able to fix it replacing some cap(s). Not sure which one though as it was a long time ago and I even think I did a complete recap anyways.

 

[bschicht86]

Its look like a sticky substance. Can you clean this part and show, that the one pin springs back cleanly?

 

[Sparksnl]

I already tried cleaning it twice. I might take my better cleaning stuff this time.
And yes I was planning to do a full recap, that might help. I already did the cpu and nb caps.
Besides the residu it’s pins are a bit black.

 

[Raiden Zero]

I‘m really confident in it just being a bad cap. At least it seems like an easier fix compared to what other dead DFI require.

 

[Sparksnl]

I sure hope so.

 

[Sparksnl]

I tested max FSB on the semi defective board. The test was a limited test because my best CPU was not good enough for this board.
Max I could test was 530MHz where my CPU is known to do 532 with effort (I did not try to push it to 532). So for max FSB testing a worthy board.