vacancies advertise contact news tip The Vault
facebook rss twitter

AMD Threadripper: 2 of the 4 dies are 'basically rocks' says exec

by Mark Tyson on 19 September 2017, 09:31

Tags: AMD (NYSE:AMD)

Quick Link: HEXUS.net/qadluo

Add to My Vault: x

The story of the relationship between Threadripper and Epyc CPUs gained another chapter last week. However AMD's James Prior stepped in this weekend to clear things up. In summary, though you can see four regular looking dies, just like an Epyc CPU, if you delid a Threadripper CPU - two dies work, other two have no path to operation. As Mr Prior summed up, the two non-operational dies are "basically rocks".

The Threadripper / Epyc relationship was originally exposed by well known OC expert der8auer, who delidded a Threadripper processor on his YouTube channel back in July. That escapade revealed what looked like four processor dies under the Threadripper lid.

Chip surface was ground down to various levels, hence the variation in appearance

In further investigations late last week der8auer bought a new retail 1950X CPU (EUR 1000) and then ground the surface off the four dies to see the silicon structures inside (see above). He concluded that "there is no such thing as a dummy die" as all four looked the same and appeared to contain eight processor cores. Der8auer went on to guess that AMD would therefore one day release a 32 core Threadripper.

Since the extra two dies didn't look like 'dummy' dies or simple structural slabs of silicon, as had previously been indicated, AMD's dummy die assertion was brought into question. Thankfully, to address the confusion AMD's James Prior, Tweeted and responded to a query or two this weekend.

On Sunday Mr Prior Tweeted that "Threadripper is not a Epyc processor. Different substrate, different dies. 2 dies work, other 2 have no path to operation. Basically rocks." In response AnandTech's Ian Cutress replied with a flow diagram, and Prior agvreed that he was on the right track.

Further clarification came from Prior later in the day when he added that "Yes, exactly why they're not described as inactive, but dummy. Doesn't matter if they were dead, or active, they're not going to work." Inactive would hint at the ability to (re)activate the dies, so AMD carefully avoided describing them as such.



HEXUS Forums :: 25 Comments

Login with Forum Account

Don't have an account? Register today!
I do like a good flow chart/diagram :D
if you need a 32 core ThreadRipper SIMPLY buy the 32 core Epyc.
lumireleon
if you need a 32 core ThreadRipper SIMPLY buy the 32 core Epyc.

Quite. I guess people really wanted a 360W TDP :D
DanceswithUnix
Quite. I guess people really wanted a 360W TDP :D

I rather suspect people wanted 16 free cores, despite the platform clearly not being able to support 4-die MCMs (can you imagine the NUMA mess of having four dies, but only two of them having memory access…?!).

As I've said previously, it's a sensible use of AMD's limited resources to keep the same physical socket and packaging; which inevitably means having to stick something in their to balance the heatspreader. Why not use dies that failed qualification - they'd end up in the bin otherwise…
scaryjim
I rather suspect people wanted 16 free cores, despite the platform clearly not being able to support 4-die MCMs (can you imagine the NUMA mess of having four dies, but only two of them having memory access…?!).

Would suck for SuperPi, but might work OK for other tasks.

In fact as someone who used to have to use 286 DOS PCs back in the day with their upper memory, high memory, low memory and extended memory it sounds fine :D

edit: Oh and expanded memory, which on a 386 could be emulated using extended memory. *shudder*