AMD Clarifies Why Threadripper Uses 4 Silicon Dies

[DiaSin]

Surprised this hasn't been posted yet.. 

 

So, for some time rumors have flown that Threadripper is made of "failed" Epyc chips, as there are only two of the four dies in use, but AMD has finally spoken up to clarify things. As it turns out, two of the dies are inert and do not even contain transistors. Basically, they are just spacers, there for structural integrity and nothing else.

 

Quote

Many thought that Threadripper simply used 4 CPU cores from each die, a point that has been long disproven by AMD, while others thought that these were "failed" Ryzen CPU dies that simply remained inactive. This left many to assume that Threadripper was simply "failed EPYC" though, in reality, nothing can be further than the truth, as AMD has now explained. 

 

It is true that AMD's Threadripper TR4/SP3r2 socket has the same layout as AMD's EPYC SP3 socket, using the same substrate design and having an identical external aesthetic. Beyond that things become very different, as EPYC uses four active CPU dies whereas Threadripper only has two, which are in a diagonal configuration to allow heat to more evenly dissipate on the CPU's IHS. 

  
The question that many have been asking is what the purpose of the other two dies are, with the answer simply being structural. These extra dies prevent imbalance and allow for simple cooler mounting without any chance of damaging the CPU. These two extra dies are not active or even contain working transistors, they are blank and as such are not "wasted Ryzen CPU dies".

Source: https://www.overclock3d.net/news/cpu_mainboard/amd_clarifies_why_threadripper_uses_4_silicon_dies/1

 

 

Edited August 3, 2017 by DiaSin
I had to re-type the quote by hand for font color to take, but I did get this fixed for the night theme... 

[Guest]

Huh. That's kinda cool.

[Lord Nicoll]

Those dies cost money, the wafers they're cut from are not cheap, the indium soldering process is also expensive, indium only wets to gold, gold won't wet to silicon either, you have to vapour deposit I think it's vanadium oxide then titanium and then gold to get the indium to stick to the silicon die, and the IHS needs a gold plating, and indium isn't cheap either, that is a lot of expense for two dummy dies, I'm not totally convinced. 

[DrMacintosh]

As long as I get them multi threaded tasks for less than what Intel charges I'm happy idc.

[Billy_Mays]

That's pretty cool

[DrMacintosh]

Just now, Lord Nicoll said:

I'm not totally convinced. 

Of? What do you need to be convinced of? 

[2FA]

2 minutes ago, Lord Nicoll said:

Those dies cost money, the wafers they're cut from are not cheap, the indium soldering process is also expensive, indium only wets to gold, gold won't wet to silicon either, you have to vapour deposit I think it's vanadium oxide then titanium and then gold to get the indium to stick to the silicon die, and the IHS needs a gold plating, and indium isn't cheap either, that is a lot of expense for two dummy dies, I'm not totally convinced. 

Still cheaper than wasting a Ryzen die when yields are high.

[kiska3]

Please fix for night theme users

[Jito463]

14 minutes ago, Lord Nicoll said:

Those dies cost money, the wafers they're cut from are not cheap, the indium soldering process is also expensive, indium only wets to gold, gold won't wet to silicon either, you have to vapour deposit I think it's vanadium oxide then titanium and then gold to get the indium to stick to the silicon die, and the IHS needs a gold plating, and indium isn't cheap either, that is a lot of expense for two dummy dies, I'm not totally convinced. 

It's possible they're using lower quality wafers for the dummy dies.  Just as with the dies produced from a wafer - with some of them being higher quality than others (hence the "binning" process) - it's not unreasonable to assume some of the wafers could be low quality and not worth using.  This would allow GloFo to recoup some of the costs from the low quality wafers, while also simultaneously giving AMD the dummy dies they need for TR.  If so, that sounds like a win-win to me.

 

Of course, that's all predicated on the assumption that they're being cut from low quality wafers, but as I said, that's not an unreasonable assumption to make.

[S w a t s o n]

25 minutes ago, Lord Nicoll said:

Those dies cost money, the wafers they're cut from are not cheap, the indium soldering process is also expensive, indium only wets to gold, gold won't wet to silicon either, you have to vapour deposit I think it's vanadium oxide then titanium and then gold to get the indium to stick to the silicon die, and the IHS needs a gold plating, and indium isn't cheap either, that is a lot of expense for two dummy dies, I'm not totally convinced. 

2 minutes ago, Jito463 said:

It's possible they're using lower quality wafers for the dummy dies.  Just as with the dies produced from a wafer - with some of them being higher quality than others (hence the "binning" process) - it's not unreasonable to assume some of the wafers could be low quality and not worth using.  This would allow GloFo to recoup some of the costs from the low quality wafers, while also simultaneously giving AMD the dummy dies they need for TR.  If so, that sounds like a win-win to me.

 

Of course, that's all predicated on the assumption that they're being cut from low quality wafers, but as I said, that's not an unreasonable assumption to make.

Yea I saw der8auer talking about it and based on what he said and what I can find on the whole die making process, those are almost certainly "real" dies that are either dead or disabled, as it would be quite expensive to make spacers rather than use failed dies.

[ConnorDahne]

Is this why der8auer got his video pulled?

[HalGameGuru]

Could they be using wafers that were incapable of being used for dies? Or failed wafers? I doubt these are materials that could have been used for saleable products

[DiaSin]

15 minutes ago, kiska3 said:

Please fix for night theme users

What's wrong with it? Also.. PLEASE tell me how to switch to that theme.. I'm totally missing something and haven't been able to figure out where I saw the setting.

[S w a t s o n]

1 minute ago, ConnorDahne said:

Is this why der8auer got his video pulled?

He explained that AMD Germany gave him permission but AMD global wasn't cool with it and rescinded the permission

[Jito463]

1 minute ago, DiaSin said:

What's wrong with it? Also.. PLEASE tell me how to switch to that theme.. I'm totally missing something and haven't been able to figure out where I saw the setting.

For the first question, edit your post, highlight all text, hit the A button in the toolbar and select 'Automatic' for the text color.

 

For the second question, scroll all the way to the bottom and click on 'THEME'.

[AnonymousGuy]

My guess is they didn't have the funding to get board partners to put TR on a separate socket, so they had to do this ghetto workaround to make it work with the same substrate as Epyc.

[DiaSin]

 

23 minutes ago, kiska3 said:

Please fix for night theme users

4 minutes ago, Jito463 said:

For the first question, edit your post, highlight all text, hit the A button in the toolbar and select 'Automatic' for the text color.

 

For the second question, scroll all the way to the bottom and click on 'THEME'.

It's actually already set to automatic for font color, I don't know why it isn't working. Thanks for reminding me where the theme switcher is though Jito.. 

[S w a t s o n]

Honestly, I'm not sure why AMD doesn't just admit that Threadripper will eventually have SKU's with more than 16 cores to compete with intel's x299 18 core+ chips, using 3 or all 4 dies. It seems fairly obvious to me.

[kiska3]

5 minutes ago, DiaSin said:

 

It's actually already set to automatic for font color, I don't know why it isn't working. Thanks for reminding me where the theme switcher is though Jito.. 

Thai is what it looks like to for me

[mr moose]

2 minutes ago, Swatson said:

Honestly, I'm not sure why AMD doesn't just admit that Threadripper will eventually have SKU's with more than 16 cores to compete with intel's x299 18 core+ chips, using 3 or all 4 dies. It seems fairly obvious to me.

Maybe they didn't think of that, maybe they just had lots of glue left over from some other project and decided this was a good use. 

 

Or maybe that + this is enough of an explanation:

 

7 minutes ago, AnonymousGuy said:

My guess is they didn't have the funding to get board partners to put TR on a separate socket, so they had to do this ghetto workaround to make it work with the same substrate as Epyc.

 

[Dash Lambda]

I think what they mean by 'they don't have working transistors' is that they're nonfunctional, not that they're blank.

I can't imagine that they would avoid using failed chips, since spacers on something like this is the perfect way to actually use nonfunctional parts, but at the same time I can't imagine they're only using failed Ryzen dies if their yields are good. So I'd guess they're both using failed dies and some sort of dummies, but that also seems convoluted...

Really, what I'd love for AMD to clear up, is exactly how they sort out and differentiate their dies.

[DiaSin]

2 minutes ago, kiska3 said:

Thai is what it looks like to for me

I'm aware. It does the same for me, now that I'm on the night theme, but it IS set to automatic. Like I said, I don't know why it isn't working. 

[ConnorDahne]

11 minutes ago, Swatson said:

He explained that AMD Germany gave him permission but AMD global wasn't cool with it and rescinded the permission

I must have misunderstood his update video then.. I thought he had permission, then lost it..
I assumed they didn't want people to know about the four dies until announced.

[Dash Lambda]

1 minute ago, DiaSin said:

I'm aware. It does the same for me, now that I'm on the night theme, but it IS set to automatic. Like I said, I don't know why it isn't working. 

Highlight the text in the quote and remove formatting.

[S w a t s o n]

1 minute ago, ConnorDahne said:

I must have misunderstood his update video then.. I thought he had permission, then lost it..
I assumed they didn't want people to know about the four dies until announced.

Well he did have permission and then lost it. He explained the whole amd germany vs amd global in his livestream Q/A video.