AMD Threadripper confirmed to have 4 actual Zepplin dies.

[TheRandomness]

7 minutes ago, TheCherryKing said:

Xeon Phi processors are not scalable. 

But you can just get more of them

[TheCherryKing]

Just now, TheRandomness said:

But you can just get more of them

How?

[TheRandomness]

Just now, TheCherryKing said:

How?

More PCIe cards.

[TheCherryKing]

Just now, TheRandomness said:

More PCIe cards.

Not more sockets. 

[TheRandomness]

Just now, TheCherryKing said:

Not more sockets. 

And? You're still essentially getting more cores, albeit having to handle them a little differently.

[kiska3]

Here are the die shots from der8auer

So which ones of the 4 are active?

[Misanthrope]

Not sure why they put modules that are basically off, maybe it has something to do with the interposer and getting all lanes? I.e. the core is defective to actually do calculations but it can still interconnect for the pci-e lanes and such.

[Dylanc1500]

1 hour ago, TheRandomness said:

Or you know, buy a Xeon Phi and get 64 cores and 256 threads at 1.2GHz with LGA 3647.

How about 96 cores 768 threads at 4.0GHz with Power9

[TheRandomness]

1 minute ago, Dylanc1500 said:

How about 96 cores 768 threads at 4.0GHz with Power9

But that isn't standard x86/64 though... right?

[Streetguru]

I think the real question is weather or not linus will be using his delidding tool for the TR delid video.

[Dylanc1500]

1 minute ago, TheRandomness said:

But that isn't standard x86/64 though... right?

No, it's POWER architechure. However in these situations you are having to run custom variants of Linux and can still perform the same "workloads".

[TheRandomness]

1 minute ago, Streetguru said:

I think the real question is weather or not linus will be using his delidding tool for the TR delid video.

Der8aur already beat him to it, with a custom delid die-mate.

Just now, Dylanc1500 said:

No, it's POWER architechure. However in these situations you are having to run custom variants of Linux and can still perform the same "workloads".

But for the sake of convenience a Xeon Phi would be easily to 'scale' and test

[Streetguru]

Just now, TheRandomness said:

Der8aur already beat him to it, with a custom delid die-mate.

But for the sake of convenience a Xeon Phi would be easily to 'scale' and test

That's what I meant, if Linus is going to contact him about getting a tool, or if he's going to use his method which will surely kill the CPU. The 2nd option will likely get more views though...

[Dylanc1500]

16 hours ago, TheRandomness said:

Der8aur already beat him to it, with a custom delid die-mate.

But for the sake of convenience a Xeon Phi would be easily to 'scale' and test

Benchmarking might be a slight issue. unless someone specifically built a program to take full advantage of both of them for a similar workload. I technically could do it given a bit of time. However as far as scalabilty they scale up to as many as you can fit in one system.

 

Oh forgot to mention Power9 has PCIe 4.0 

[dcgreen2k]

Stupid question, but what's a Zeppelin die? Is it from one of AMD's old AM3 CPUs?

[WMGroomAK]

14 minutes ago, thegreengamers said:

Stupid question, but what's a Zeppelin die? Is it from one of AMD's old AM3 CPUs?

The Zeppelin die is basically a System on Chip layout with two CCXes for the Zen micro-architecture and each Ryzen processor consists of a single die, while Threadripper has two active dies and two dead/deactive dies and Epyc has four active dies.  

[S w a t s o n]

4 hours ago, TorqueS said:

Haven't this been known for a while?

From what I recall, I watched der8auer's video and I believe the 2 dummy dies have no circuitry in them. They're there so that the surface area of the IHS is uniform.

 

You didn't watch the video did you? The entire point of der8auer making this follow up was to confirm that the "dummy dies" do have transistors and such. They are complete Zepplin dies (each one is effectively a Ryzen 7).

[dacheezypuffs]

8 hours ago, Swatson said:

So it appears that der8auer has just confirmed that the 2 extra "dummy dies" in Threadripper are not in fact just silicon spacers but disabled or at least non-functional Zeppelin dies that do contain transistors.

 

 

Personally, this didn't surprise me as this was my theory all along. Silicon and packaging aren't free and even though Zepplin die yield is supposed to be 80%+ it would still make sense to use real dies even if you have to disable working ones.

 

This also supports my theory that we might see a 32 core Threadripper down the line, I assume this will be slowly rolled out to compete with Intel's HCC/XCC core i9 chips.

I thought we all knew this...

[S w a t s o n]

50 minutes ago, dacheezypuffs said:

I thought we all knew this...

Did you read the post I made literally above yours?

 

Edit: In the case you really thought we all knew this, no. No one knew the dies were actual dies, this is the first confirmation in the world.

[vanished]

This is a lot more satisfying of an explanation.  I could never believe they were just blanks, it seemed so stupid

[SpaceGhostC2C]

Good guy AMD put 2 "shadow dies" to handle all the spying from NSA, CIA, FBI, MS, FB, Google, etc at the hardware level. Because when you go high end, you deserve to have all the CPU resources are your disposal

 

 

On 9/15/2017 at 10:04 PM, Streetguru said:

That's what I meant, if Linus is going to contact him about getting a tool, or if he's going to use his method which will surely kill the CPU. The 2nd option will likely get more views though...

Linus wants it alive. His goal is to transplant the engraved IHS in his decorative dead CPU into a working Threadripper. 

[TorqueS]

11 hours ago, Swatson said:

You didn't watch the video did you? The entire point of der8auer making this follow up was to confirm that the "dummy dies" do have transistors and such. They are complete Zepplin dies (each one is effectively a Ryzen 7).

No, I didn't because I thought it was the same video.

[mariushm]

Why cut dies out of a perfectly good wafer when you can just re-use dies that have failed binning (for example dies that can't even be sold as Ryzen 3 1200) ?

 

My guess is they're simply dies that have some critical errors like maybe several bad spots in the L3 cache memory (which would make them unable to get even 4 MB of  cache memory to sell these as Ryzen 3) or maybe some failure in the chip's ddr4 controller or other such critical points.

 

The socket is recycled from EPYC so in theory it has the pins for extra memory channels but i would suspect on Threadripper, they'd repurpose those pins as grounds and voltage (for improved power delivery) so most likely if they'd ever make such processors with more than 2 dies active, they wouldn't work in current motherboards.

 

No, I really doubt they'd make Threadripper with more than 2 dies, there's EPYC processors for that, and they already have 8 core / 16 threads EPYC cpus for just 50$ extra or something like that.  Where would they squeeze more powerful Threadripper cpus?

 

 

[Guest]

I think that the implementation of the two unactive dies may have something to do with cooling, and the heat dissipation from the CPU within a larger surface area. I think that the larger surface area on the CPU helps dissipate more heat from the CPU, reducing the temperatures. Maybey this is something AMD was trying to achieve because of the higher TDP.

[Bananasplit_00]

wow no shit its 4 dies, did anyone actiually think they would cut un-used silicon just for dummies?