Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Dear lord people, you're all massively overthinking it.
EPYC was done, it's a 4 die package, with a specific socket, drivers, support, design and mobo manufacturer support. It's monumentally cheaper for AMD and the mobo manufacturers to reuse the same socket than to create an entirely new socket, which needs a full validation/certification cycle which is costly and takes a lot of time, to create a new socket. It also means for what is a very small overall market for Threadripper you increase the 'getting in' cost from the mobo manufacturers.
So you have a socket that supports more than 1 die, you have a socket that can easily support quad channel and it's all ready to go, the manufacturing for sockets is in place, why would you create an entirely different socket for more money when you can reuse it?
That is literally all this is, to create a new socket and therefore have a third socket for no reason would cost 10s of millions in development when the socket you have already spent that money on, much much more actually, on a socket that will work just fine.
So they reuse the socket, save millions and reduce costs while also making partners happier to have one overall platform to work on. Silicon costs practically nothing, think about this, the cost of an actual wafer to be processed, which takes ~6-8 weeks these days on these processes, is around $5k, that is hundreds and hundreds of hours of processing and that is the overall cost, the actual cost of the silicon itself is tiny. Attaching things to organic packaging is absolutely trivial cost, using the spacers probably costs in the range of $1-2 total, silicon and attaching 2 more pads. Organic packaging is absolutely trivial, completely conquered, insanely high yield technology these days, a silicon interposer brings everything down in scale by a magnitude and is new tech with worse yields and higher cost, but the yields there are still good. AS in organic packaging is in effect 100% yield it's old and very easy technology with the way they are manufacturered. Interposers took till now to be used because yields aren't 100% and with every extra chip yield goes down meaning 4 chips may cause a drop in yields by 5%. Except even if this was a silicon interposer, the two dummy dies couldn't actually fail, they only need to be stuck together(trivial) not have thousands of physical connections made perfectly(not so trivial).
The cost of using the same organic packaging design and two extra silicon pads for spacers is absolutely trivial and insanely lower than the cost of producing an entirely new platform with a socket midway between AM4 and server. It's also hugely less work which at a time AMD is struggling for money and needs to get out a lot of products in a short space of time, would simply be a waste.
EPYC was done, it's a 4 die package, with a specific socket, drivers, support, design and mobo manufacturer support. It's monumentally cheaper for AMD and the mobo manufacturers to reuse the same socket than to create an entirely new socket, which needs a full validation/certification cycle which is costly and takes a lot of time, to create a new socket. It also means for what is a very small overall market for Threadripper you increase the 'getting in' cost from the mobo manufacturers.
So you have a socket that supports more than 1 die, you have a socket that can easily support quad channel and it's all ready to go, the manufacturing for sockets is in place, why would you create an entirely different socket for more money when you can reuse it?
That is literally all this is, to create a new socket and therefore have a third socket for no reason would cost 10s of millions in development when the socket you have already spent that money on, much much more actually, on a socket that will work just fine.
So they reuse the socket, save millions and reduce costs while also making partners happier to have one overall platform to work on. Silicon costs practically nothing, think about this, the cost of an actual wafer to be processed, which takes ~6-8 weeks these days on these processes, is around $5k, that is hundreds and hundreds of hours of processing and that is the overall cost, the actual cost of the silicon itself is tiny. Attaching things to organic packaging is absolutely trivial cost, using the spacers probably costs in the range of $1-2 total, silicon and attaching 2 more pads. Organic packaging is absolutely trivial, completely conquered, insanely high yield technology these days, a silicon interposer brings everything down in scale by a magnitude and is new tech with worse yields and higher cost, but the yields there are still good. AS in organic packaging is in effect 100% yield it's old and very easy technology with the way they are manufacturered. Interposers took till now to be used because yields aren't 100% and with every extra chip yield goes down meaning 4 chips may cause a drop in yields by 5%. Except even if this was a silicon interposer, the two dummy dies couldn't actually fail, they only need to be stuck together(trivial) not have thousands of physical connections made perfectly(not so trivial).
The cost of using the same organic packaging design and two extra silicon pads for spacers is absolutely trivial and insanely lower than the cost of producing an entirely new platform with a socket midway between AM4 and server. It's also hugely less work which at a time AMD is struggling for money and needs to get out a lot of products in a short space of time, would simply be a waste.