AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

Which DRAM manufacturer would be ready to invest in GPU memory optimized for silicon interposers or stacking though? For mobiles I can see one of them stepping up to the plate, GPUs ... dunno. Without Qimonda I doubt we would have even had GDDR5 and we know what happened to them.
 
Samsung and Micron have stacked memory initiatives.
Intel is rumored to be looking into using a silicon interposer for a CPU.
Intel and Micron seem to be joined tightly on stacked memory, so I do not know if Micron would partner with AMD on this.
Maybe this is a test chip on a Samsung/IBM/GF initiative for interposers and stacked memory?
 
Maybe it's for mobile shit GPUs?
Some of the value chips have memory controllers that can interface with multiple memory types besides GDDR5.
 
The bottom of that shot has some text that looks like "Amkor Confidential".
Actually, Amkor has a marketing shot of its 3D TSV tech that bears a strong resemblance to this one, although that example only had one set of stacked memory.
 
OK I'm gonna say it, "console"
Smile-hiding.gif
 
It could be for Wii U too as suggested by someone at Muropaketti, even though it's not "far future" really.
 
OK I'm gonna say it, "console"
Smile-hiding.gif

That's what I thought too. We already know the Wii U has an AMD GPU, but it's a kind of RV770-ish thing, and this looks bigger (assuming 40 nm or smaller).

So maybe the next Xbox or Playstation.
 
That's what I thought too. We already know the Wii U has an AMD GPU, but it's a kind of RV770-ish thing, and this looks bigger (assuming 40 nm or smaller).

So maybe the next Xbox or Playstation.
I think anything tagged with "far future" should be 28nm (or below but I can't see how they could get a prototype with below).
Wii U sounds like not "far future" to me, but OTOH I don't think there would be anything close to a gpu prototype for the next Xbox or Playstation at this point.
 
I dare say a key point here is that ultra-high bandwidth courtesy of this RAM arrangement obviates EDRAM as seen with Xenos.

Additionally, power levels should be "reasonable", rather than 250W+ as seen with discrete. So cooling won't be a nightmare for the early iterations. Though this layout is more forgiving of GPU power (at least in engineering terms) than a variant with memory stacked atop the GPU die.

Amkor%201.jpg
 
The site appears down... does anyone mind explaining the general design of the interposer to me? (From what I can gather from this thread the link has info about kind of chip-to-chip interface, possibly some stacked memory?)
 
My understanding is its like a tiny on-die motherboard, allowing you to put the CPU on one side and RAM on the other, without having to embed the RAM eDRAM-style. Basically, you get the low latency/high bandwidth like eDRAM but you don't have to do special masks and processes like eDRAM, the RAM can be made seperately.
 
The site appears down... does anyone mind explaining the general design of the interposer to me?
It's simply a new form of multi-chip-module, but instead of using a glorified PCB to put the chips for the module on they use another HUGE chip (made in an older process to keep costs down).
 
Which DRAM manufacturer would be ready to invest in GPU memory optimized for silicon interposers or stacking though? For mobiles I can see one of them stepping up to the plate, GPUs ... dunno. Without Qimonda I doubt we would have even had GDDR5 and we know what happened to them.
If you're a GPU maker, what features of DRAM as we currently know it, are desirable in a memory subsystem enabled by an interposer?

It seems to me that this is an opportunity to create something with the performance and granularity of "L4 cache" but sized as entire system memory.

In other words, why wouldn't a GPU manufacturer make the memory chips too? AMD and NVidia already know how to build caches...
 
Back
Top