820-3787 CPU FETs, what are the number of the pieces to replace


New member
820-3787 replace the CPU FETs, can you guys tell me which are the ones that i have to replace

i have this 820-3787 that has random shutoffs, been reading and duke had posted that was the issue,


Man, I'd say don't waste your time. This "fix" works minimally at best, and the issue recurs after a very short time. IF you insist on this repair, do it ONLY on 3662/00138 boards. The DG boards just do not like this repair, and don't seem to "take" the repair around 90% of the time. The IG boards seem to be much more successful. These chips are pretty expensive, so you'll just be throwing $10-15.00 out the door each time you repair a board when it doesn't work (and you'll likely never need that specific chip other than on this specific repair). Just my opinion; maybe others have had better luck. I was working with Duke in testing this repair a few years back when we found evidence of Apple replacing them (to try to "fix" the issue). Had high hopes, but they just don't seem repairable after the issue appears. Avoid heat to the CPU when you do this, and make sure you check for STG on S5_HS after replacing each one. These have huge ground pads/pin arrangements and can be difficult to replace. Check your soldering very well, don't overheat, and maybe you'll get lucky.

I've repaired around 40 of these; I think maybe 1 or 2 DG boards never came back, but around 15 did. Of the IG boards, I'd say a solid 60% were actually "fixed" long-term (>6 months), and the rest came back. Stopped even screwing with these models, as it got to be a colossal waste of time.
Last edited:


Super Moderator
Staff member
I have had much better luck with these. I have done many, and none have come back so far. I've caught a few in testing that have still done it after the replacement, but so far none have come back.

How I test these.

At least 1hr of prime 95 large FFT and unigine valley running at the same time.

Let it sit on the desktop at idle for 30 min. Every 30 min, open something random, then close it. Repeat this several times in 30 min increments.

Have someone do some light web browsing on it as well. I've found the duds will crash while you still have it.


TCRS, are you using FDM6708s or 6808s? Replacing only the power supplies or controller as well? The way we test is to let the unit sleep overnight (6+ hours). If CPU is damaged/bad, it usually either won't wake from sleep (most common--LCD stays black, etc), or will wake/crash instantly. I've tried editing the hibernate function via terminal to have it go straight into hibernate (so we could speed up testing, etc), but it didn't work the same. The overnight sleep seems to be about 98% effective in catching the duds. Confident in the soldering, no issues there, but a significant percent of the boards end up having the same issue within a few months. We use new 6708s. For what it's worth, I've heard the same (repair doesn't seem to take or last in many cases) from a few other repair shops. Curious if you're lucky (lol) or if there's something we're missing.

If you're interested, I have around ~50 3662/3787/00138/00163/00426 boards with reported issues that all sound like vcore dropout. If you'd like to purchase them, shoot me a message. None tampered with or attempted-repair.
Last edited: