AMD Ryzen Threadripper MCM De-lidded and De-packaged

[Dave65]

I'm pretty sure the CPU was already dead. In his original TH de-lidding video he said that the CPU did not survive the de-lidding process. So I'm guessing he used that dead CPU.

And he wasn't intending to kill the CPU when he did the original de-lidding, it was unfortunate that it died.

No, if you watch his video he went out and purchased this cpu with his own money, it was NOT the engineering sample from before..

 

[damric]

Didn't AMD say that they DIDN'T use non-functional dies, and the non-working dies were just spacers without any transistors? Or was this more BS by sites that AMD should have never supported? ROFL

I wonder what would happen if you stuck a TR CPU in an EPYC board? Would there possibly be a way to enable those "EPYC Failures"??

If it works, I'm sure ASROCK will capitalise on this with some interesting boards. Is Epyc even available yet?

 

[Slizzo]

That's probably because not all TR CPU will have four dies, reminds me of the time when Phenoms could be unlocked, so most likely YMWV.
Who knows AMD might do an Intel here & unlock these, for an undisclosed amount.

But, all TR CPUs DO have 4 dies, from the 8 core (lowest core count able to be produced) up to the 16 core, all have 4 dies, 2 active, 2 now apparently known as failed (but still "dummy").

 

[OSdevr]

But, all TR CPUs DO have 4 dies, from the 8 core (lowest core count able to be produced) up to the 16 core, all have 4 dies, 2 active, 2 now apparently known as failed (but still "dummy").

I think he's suggesting that some TRs have blanks and Der8auer by chance got one that didn't. Plausible but not easily provable one way or the other.

 

[Slizzo]

I think he's suggesting that some TRs have blanks and Der8auer by chance got one that didn't. Plausible but not easily provable one way or the other.

After reading Der8auer's comments, I don't think that there ever were blanks. Wouldn't make sense when it comes to manufacturing.

That's a lot of silicon to waste to just make blanks. Silicon ain't all that cheap either.

 

[OSdevr]

After reading Der8auer's comments, I don't think that there ever were blanks. Wouldn't make sense when it comes to manufacturing.

That's a lot of silicon to waste to just make blanks. Silicon ain't all that cheap either.

I agree. I just wonder what if anything this means about Ryzen yields. Is it below 50% or are many of those dummies good and just unusable because of the package? Chip makers have laser disabled good parts of a die to make more of a popular lower tiered product before, this could be the same idea applied to a MCM.

EDIT: I just realized something obvious. Ryzen uses 1 good die and EPYC uses 4 good dies. Depending on demand for the various products Ryzen yields could still be very high and they would still have plenty of failed dies for TR!

 

[newtekie1]

No, if you watch his video he went out and purchased this cpu with his own money, it was NOT the engineering sample from before..

Wow, yeah, then I guess SCIENCE!

 

[OSdevr]

I must admit I was not only completely wrong in some past posts but also completely surprised by this. I guess this method of packaging is very cheap for AMD to waste two entire dies (I assume that at least some of the dies aren't working out of the factory so cannot be EPYC processors).

That's really no-one's fault, really. They were purposefully vague.

What's more interesting is the implication that has for Ryzen's yields.

That's probably because not all TR CPU will have four dies, reminds me of the time when Phenoms could be unlocked, so most likely YMWV.
Who knows AMD might do an Intel here & unlock these, for an undisclosed amount.

Sorry, but those 2 extra dies are probably dead in all TRs, but it only means that Ryzen yields are less than 100%.

The Zeppelin die is used in Ryzen, TR and EPYC, but only TR and possibly some EPYCs need dummies. Further TR is an HEDT product, Ryzen sells much better. Say AMD sells 2 Ryzens and 1 EPYC for every TR they sell. Say all 4 dies in the EPYC are good. That's a total of 10 dies. But they don't have to start using good Zeppelins as TR dummies unless their yield is higher than 80%, not 50%!

We've never seen something like this before. This sort of thing will only happen with MCMs in which all the dies are identical. It doesn't occur in traditional GPUs or CPUs because you can't recycle a bad part of a chip into another package.

 

[eidairaman1]

Are we sure about that? I'd be willing to try...

It be awesome if we could do a ribbon cable mod like how 771 xeons could work in 775 motherboards and be unlocked, or how the AXP days were where you could make an AXP into an AXP-MMP lol (mobile multi processor)

 

[trparky]

Wow, yeah, then I guess SCIENCE!

Yeah but to do that to a chip, to purposely kill it? Man, I couldn't do it; I just couldn't.

 

[MrGenius]

I highly doubt they're putting two extra "good" dies on there. They must be defective. Even if they were "good" it's highly unlikely you'd ever be able to "unlock" them. Not without delidding it and soldering on the missing 30ish capacitors first.

 

[newtekie1]

Yeah but to do that to a chip, to purposely kill it? Man, I couldn't do it; I just couldn't.

Yeah, but this guy lives to delid. Hell, he'll probably make the money back and then come selling Threadripper delidding tools based on what he learned from this CPU.

I highly doubt they're putting two extra "good" dies on there. They must be defective. Even if they were "good" it's highly unlikely you'd ever be able to "unlock" them. Not without delidding it and soldering on the missing 30ish capacitors first.

I'd bet they are defective. I'm not too worried about unlocking them, I'm more excited for the possibility of a 32 core Threadripper in the future.

 

[Chaitanya]

Yeah, but this guy lives to delid. Hell, he'll probably make the money back and then come selling Threadripper delidding tools based on what he learned from this CPU.



I'd bet they are defective. I'm not too worried about unlocking them, I'm more excited for the possibility of a 32 core Threadripper in the future.

Actually there is 0 incentive for selling deliding tools for AMD Ryzen based CPUs which all have soldered IHS. There is more chance of people actually damaging the CPU during the process.

 

[StrayKAT]

The one positive you can take from Intel cheaping out....

 

[Imsochobo]

Didn't AMD say that they DIDN'T use non-functional dies, and the non-working dies were just spacers without any transistors? Or was this more BS by sites that AMD should have never supported? ROFL

I wonder what would happen if you stuck a TR CPU in an EPYC board? Would there possibly be a way to enable those "EPYC Failures"??

They've only said they weren't functional.
they have sorted trash issue, no paying money to trash broken chips

 

[cadaveca]

They've only said they weren't functional.
they have sorted trash issue, no paying money to trash broken chips

Heh. That's one way to look at it. Looking online at stock numbers, it doesn't really amount to that large of a number of dies anyway, so while I might seem to be critical, in the grand scheme of things, it is not really that big of a deal if they are 100% functional, anyway. Locally the 1900X is $789.99, so whatever the cost is to AMD, I am sure it's fine. Like that's turning 4 "defective" dies into $200 each. I know what they should really cost, and that's not just paying money for broken chips, that's making bank!!

I mean, then the 1920X is just over $1100 at normal prices, making that single fully-functional chip worth $500, and then rightly so the 1950X is $1400, sealing the deal. A 1300X is what, $160? Two of those chips, in a single socket, for an added $370 bucks? You're paying for those "dead dies" anyway. Also note that a 1300X, while clocked a bit lower, is only a 65W chip, but the 1900X is a 180W one...

 

[Vya Domus]

I highly doubt they're putting two extra "good" dies on there. They must be defective.

Remember that the dies used in here have pretty much the same yields as Ryzen 7 , it doesn't cost them any extra to make dies for TR/Epyc than it does for Ryzen. Epyc is were they are going to be move a lot of volume in the following months , the HEDT market is a lot smaller in comparison so it shouldn't surprise you at all that maybe these CPU were actually fully operational.

 

[newtekie1]

Actually there is 0 incentive for selling deliding tools for AMD Ryzen based CPUs which all have soldered IHS. There is more chance of people actually damaging the CPU during the process.

Thing is, he is probably going to be the only one selling them. And for the record breakers out there, there is incentive. He'll sell a few of these to the extreme overclockers for $500 a piece and make his money back and then some.

 

[R-T-B]

"Sir, why did you destroy a one thousand dollar CPU?"

"FOR SCIENCE!"

Personally, I like the rainbow colors the silicon makes.

 

[goodeedidid]

Stupid to destroy the CPU. You learned what? Nothing.

 

[dj-electric]

Stupid to destroy the CPU. You learned what? Nothing.

They basically gave a non-functioning one with every review kit

 

[R-T-B]

Stupid to destroy the CPU. You learned what? Nothing.

We learned the dies are indeed at least attempts at a zepplin die, no "silicon blanks."

 

[GoldenX]

And that is totally possible to have a 24 core TR, even a 32 one.

 

[DonCaffe]

Nice brener work....

 

[Imsochobo]

I think he's suggesting that some TRs have blanks and Der8auer by chance got one that didn't. Plausible but not easily provable one way or the other.

blanks?
why not throw in broken chips and get rid of trash.
You wouldn't believe how much trash costs... I expect a few cents per cpu, packaging shared with epyc = maybe a dollar.
they are really maximizing their margins without letting us consumers suffer (looking at intels TIM here!)