PS3 #16 - Mini update
I'm waiting on a thermal camera to arrive. I'm hoping it will tell me which XDR ram chip is not heating up (powering on). Then I'll try replacing it. If they all heat up, then the issue could be the CPU BGA itself. ACE Console repair just had that happen and it caused a similar error...
Anyway his thermal camera saw that it wasn't heating up. He first tried replacing the ram but it didnt work. Then reball. Unfortunately it failed with 3034 and he had to called it unrepairable.
Regardless it's a path forward for me to explore when the cam arrives.
In the meantime this gave me the oppritunity to investigate a theory I have been working on. There are two 65nm RSX models (CXD2982 & CXD2991). The earlier was released in August 2008 in J/K models and only 2 months later was replaced with the latter. Sony "said" they were replacing the 40GB model with the 80GB. But the only the J was 40GB. The K was 80GB. So that sounds like a cover story. What really changed was the MB went from DIA-002 --> VER-001. And the 65nm RSX went from 2982 --> 2991. Why?
The MB was radically redesigned and they were cutting costs to frantically make a profit, so it could just be coincidence. But late 2008 was around the time Nvidia was releasing "fixed" chips. But they were still struggling to understand the defect in thir designs. They were hiring thermal design engineers at the time, signaling they were stumped.
Around that time the fixed chips they released had white underfill material (High Tg). So I wanted to see the underfill material of the these two 65nm chips. To see if there was a color difference to the 90nm, and truely fixed 40nm. I had a 2991, but not a 2982. They are actually kinda rare, since they were only on J and K models. But this MB squeept sent me had one..
I didn't like the idea of delidding it, but I litterally could not fin any pictures of a CXD2982 online anywhere! So I did it and found this.
@squeept you may be interested in this part. I have questions!
First, clearly that paste is new! Meaning
@squeept delided and used thermal epoxy to reattach the IHS (like a good boy). THIS WAS PERFECT!
Whatever he used it was the right stuff! It was super stiff and I had to apply heat to get it off. 100C for maybe 30s and it loosened then came off really nice. Absolutely the right epoxy. So my first question is, was that the epoxy I reccomended? Or something else. Because now we know what to use.
Second, you can see the underfill isn't white. But it's also not the same color as the 90nm, indicating it's not the same defective underfill as the 90nm..
The middle one is a CXD2991. There is perhaps a slight color difference between the two 65nm underfills, but I wouldn't say it's significant enough to conclude they have different ones. It's probably just lighting. I did take these in direct sunlight at the same time, to try and mitigate the effect of exposure and color correction processing. So they are comparable to one another, unlike most photos I've seen on the net. There is a milky whiter color to the 40nm however. The 90nm is greyer. And the 65nm is more brownish.
This tells me the underfill was changed for the 65nm, as expected. Presumably to the proper High Tg needed for High-Lead bumps. That's good, but isn't the full story.
The G9600 was released in feb-2008 with High Tg underfill, High-Lead bumps, and no polyamide stress layer! The PI layer is a coating placed on the die that absorbs and transfers stress from the bumps. It helps protect the dielectris coating underneath from contamination and delamination due to stress. When you use High Tg underfill, yes it protects the bumps from cracking, but it also transfers more stress to the PI and Dielectric layers, making delamination possable. So it's standard to use a certain thickness PI layer to prevent that. And it has been proven with electron micrographs that the G9600 didn't have it! Nvidia traded bump cracking for delamination. The result is the same, the chip dies.
The RSX is not the G96. IDK if it does or does not have the PI layer it should. But I know that chipsets released around the same time, designed by Nvidia's clearly struggling engineers. It's not far fetched to think they may have left the PI layer off the 65nm RSX too. Or perhaps just the 2982. Maybe that's why the 2991 was released only 2 months later and why SONY released the VER-001. To get them fixed fixed chips.
Like I said. It's a theory. And this MB gave me an oppritunity to have a look at a chip I could not get an image of. And it turned out to be useful because it confirms whatever thermal epoxy
@squeept used is the perfect stuff to reseal the RSX with.
That's critically important for 90nm at least, Because it's underfill has a Tg of 70C and the RSX routinely runs hotter. The underfill softwns and the bumps crack prematurely. WebMAN can keep the temps below this point if you used dynamic at the default 68C. But good paste is needed to keep fan noise at a minimum. And if you have to delid to achieve this, then tge IHS is no longer stiffening the package and it makes BGA faikures more likly, especially in the corners where the syresses are greatest. Gluing it back on is important for this reasn. And now we know what to use. Or squeept does anyway.
