The Ryzen "ThreadRipper"... 16 cores of awesome

Page 45 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

scannall

Golden Member
Jan 1, 2012
1,960
1,678
136
I'm curious what Dell's Area 51 Threadripper systems are going to look like. Knowing them, they'll probably come standard with a pathetic 8 GB of RAM, a low end video card like a 3 GB GeForce 1060, and a 1TB 5400RPM hard drive. If you're going to put crap like that in the box, you might as well just get a Ryzen 5 instead.

By the time you "upgrade" to a proper amount of memory, a good video card, and an SSD in it it will probably cost $5,000.
Well, you can look for yourself now. ;-) Alienware Threadripper
 
  • Like
Reactions: lightmanek

Space Tyrant

Member
Feb 14, 2017
149
115
116
Threadripper Delid.
https://www.youtube.com/watch?v=ZoVK6rJR5VE

There are 4 dies, not 2!
The mystery, obviously, is why?

The only possibility that makes sense to me is that there are a small number of bad dies. With Threadripper being a low volume platform, they simply feed those dies into the unused die slots of threadrippers -- and avoid having two different physical assembly processes.
 
  • Like
Reactions: IEC

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
The mystery, obviously, is why?

The only possibility that makes sense to me is that there are a small number of bad dies. With Threadripper being a low volume platform, they simply feed those dies into the unused die slots of threadrippers -- and avoid having two different physical assembly processes.

So the unused dies are spacers? Yowzers!
 

ddogg

Golden Member
May 4, 2005
1,864
361
136
I just saw that debeau8er video and it's really interesting. If those chips could be unlocked to 32C it might be the greatest CPU ever released :) I guess this also gives AMD a good upgrade path to potentially release a 24/32C threadripper in the future at reasonable prices.
 
  • Like
Reactions: lightmanek

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Two dead dies, or four partially functional ones?

Video says 2 dead dies, 2 fully functional. So it behaves just like a 2 die package would.

You only have quad channel memory on thread-ripper, and you have 2 memory channels per die.

This implies something else to me. It seems likely that the same two dies are inactive in every threadripper.

And that either:
1: Dies are tested before packaging and bad dies are placed in unused slots.
2: Dies are tested after packaging, but many good dies are simply disabled.

Either way, this is very strange.
 

EXCellR8

Diamond Member
Sep 1, 2010
4,045
889
136
of coure someone's already delidded the ripper. disposable income :cool:

EDIT: sorry vid was already linked...
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
1,540
136
Evidently enough. But that only makes economic sense if the unused dies are non-functional or otherwise substandard. So, what am I missing here?

I don't think anyone has ever seen a product designed around multiple dysfunctional dies before. There have been lots of dual die products, but they all used dual dies. Not two functional and two dysfunctional ones.

Also, it seems lucky to have dysfunctional dies line up with only the memory channels used in Threadripper.

That implies just as many dysfunctional dies line up in the wrong place and are garbage, or these aren't dysfunctional and they are just disabled to create the parts.

It's a very strange situation.
 

krumme

Diamond Member
Oct 9, 2009
5,956
1,595
136
Evidently enough. But that only makes economic sense if the unused dies are non-functional or otherwise substandard. So, what am I missing here?
That the marginal production cost of a full fat functional 8c die is only say 30 usd?
 

Yakk

Golden Member
May 28, 2016
1,574
275
81
If AMD are using partially disabled/bad dies, on top of small cores to begin with. I can see how they can sell TR for relatively cheap and still make substantial profits compared to an Intel binned full monolithic core.

All the while keeping up performance and keeping a possible upgrade path to a 32 core CPU on the same socket, nicely done AMD.
 
  • Like
Reactions: MangoX and ddogg

Space Tyrant

Member
Feb 14, 2017
149
115
116
That the marginal production cost of a full fat functional 8c die is only say 30 usd?
Yes. In addition to that expense, there is also the potential for *substantial* opportunity cost. Imagine a scenario where fab production falls behind demand. To me, that's not inconceivable.
 

Pick2

Golden Member
Feb 14, 2017
1,058
1,507
91
Maybe , they need two dies for spacers with two good dies to properly support that Massive IHS
 

Space Tyrant

Member
Feb 14, 2017
149
115
116
https://twitter.com/IanCutress/status/890593099952664576

Seems likely that it's just because it's an engineering sample and not the final product.
Yes, that would explain it.

Edit: I suppose it's also possible that there was never a specific 'threadripper' MCM at all. Rather, that all TR's are just EPYC's that failed to match the requirements for any higher-end EPYC SKU and instead wound up with 2 chips disabled and branded as TR.
 
Last edited:
  • Like
Reactions: ajc9988