https://www.tomshardware.com/news/amd-gpu-chiplet-patent Breaking Through the GPU Performance Barrier using chipletsOn the face of it, this move certainly makes sense. As you increase die sizes, yields tend to drop drastically due to defects in the silicon, leading to costly losses that impact prices. Semiconductor manufacturing equipment also has a reticle size limit, which essentially creates a barrier beyond which it simply isn't possible to make a bigger GPU. Coupled with the state of the process node (TSMC is currently at 7nm), there is a performance ceiling that foundries simply cannot manufacture beyond.
AMD Chiplet GPU Patent
However, breaking the big silicon die into multiple little ones addresses all these pain points.
And as such, AMD proposes that a chiplet-based GPU just needs a simple electrical path to be tied together to look like one big GPU, rather than a bunch of little ones addressed through some form of controller.
AMD Chiplet GPU Patent
The beauty of this GPU model is that it remains compatible with existing coding languages, and the GPU model doesn't change from the developers' point of view. A few things will need to be adjusted in the driver, naturally, to account for the new architecture, but existing software should run without any major changes.What I like about this is the new on chip communications fabric improvements is that it may well be able to flow back into the CPU side, making it possible for resources to be shared between CPU, GPU and the upcoming FPGA/AI functions.
And, having gaming level graphics boosted by programmable-on-the -fly AI
all on the same 14-20 layer chipset --- hey, what is not to like?
==================================================
Intel reacts to this serious announcement of a real AMD development project with BS about a 20 core Big Little with no graphics on board (Intel will use a separate GPU chipset if they ever get around to building this theoretical response).
AMD posts die shots of real engineering sample parts ---- Intel responds with some nebulous BS verbiage about an not yet designed Intel big little that would be built on a lithography node that Intel admits it cannot get to run right.
Which one do you think will be built inside a year from now ??????
Intel's stock price is getting simply hammered over shite like this ----- Pat Gelsinger, make the bad PR boys stop their endless lip dribbling as it is jest killing your image.
==================================================
Year end results paint an interesting picture.
AMD APUs were what was desired, but Intel Tiger Lake was what was available and that was what actually got bought at the end of the reporting period.
Supplies of full PC chipsets from everybody went dry at the end ---- so a lot of Chromebooks got bought in the last few weeks.
TSMC can no longer supply all of AMD's needs, so Samsung is getting rung into the mix on an ongoing fashion. TSMC can always get this volume back by contract terms, but only by supplying all of AMD's orders religiously on time and at the contracted price.
===================================================
TSMC has already published their 5nm expansion plan (nothing else is new, really, TSMC 5nm is already all built out and it is all set up and rolling product right now --- and it is all firmly allocated to Apple and AMD and Qualcomm, and the phone boys, the guys who paid to build it).
TSMC's 3nm expansion is about half built, but is already totally scheduled tight and full through 2024. Same major players paid to build this 3mn stuff and have first dibs on what will be a scarce allocated resource from the very 3nm beginning.
It is already known that Intel did not buy into either 5nm nor 3nm by the original cut off time and as such INTEL HAS NO ALLOCATION LAID IN FOR ANY CURRENT EXISTING 3nm TSMC PRODUCTION LINES THROUGH YEAR 2024.
Intel claims that they do have allocations though, so we shall see. We do have the two "anybody's production" 3nm plants going into Austin Texas, and perhaps that is what Intel is counting upon. And, Intel may be referring to Samsung instead of TSMC.