Friday, June 11th 2021

AMD Confirms CDNA2 Instinct MI200 GPU Will Feature at Least Two Dies in MCM Design

Today we've got the first genuine piece of information that confirms AMD's MCM approach to CDNA2, the next-gen compute architecture meant for ML/HPC/Exascale computing. This comes courtesy of a Linux kernel update, where AMD engineers annotated the latest Linux kernel patch with some considerations specific for their upcoming Aldebaran, CDNA2-based compute cards. Namely, the engineers clarify the existence of a "Die0" and a "Die1", where power data fetching should be allocated to Die0 of the accelerator card - and that the power limit shouldn't be set on the secondary die.

This confirms that Aldebaran will be made of at least two CDNA2 compute dies, and as (almost) always in computing, one seems to be tasked with general administration of both compute dies. It is unclear as of yet whether the HBM2 memory controller will be allocated to the primary die, or if there will be an external I/O die (much like in Zen) that AMD can leverage for off-chip communication. AMD's approach to CDNA2 will eventually find its way (in an updated form) for AMD's consumer-geared next-generation graphics architecture with RDNA3.
Source: Videocardz
Add your own comment

5 Comments on AMD Confirms CDNA2 Instinct MI200 GPU Will Feature at Least Two Dies in MCM Design

#1
DeathtoGnomes
It is unclear as of yet whether the HBM2 memory controller will be allocated to the primary die
there was a PR about Hbm3e recently, could it be within the same time frame?
Posted on Reply
#2
Manoa
is not so hard to throw more chip in group for computational problems, it's hard when you wanne do graphics :x
koduri did it too with the "more flops" (lel) so what
Posted on Reply
#3
95Viper
Stay on topic... the topic is "AMD Confirms CDNA2 Instinct MI200 GPU Will Feature at Least Two Dies in MCM Design"
Posted on Reply
#4
Jism
Manoais not so hard to throw more chip in group for computational problems, it's hard when you wanne do graphics :x
koduri did it too with the "more flops" (lel) so what
Well, this looks like they are going to pair the computational cards based on the same stuff as RDNA3 is for graphics. It's just that graphics seem to get still one chip which is obviously better.
Posted on Reply
#5
Imsochobo
Manoais not so hard to throw more chip in group for computational problems, it's hard when you wanne do graphics :x
koduri did it too with the "more flops" (lel) so what
it's actually harder, they depend more on other threads than graphics.
Posted on Reply