Fun fact: Aldebaran is the brightest star in the constellation Taurus. It is 44 times the size of the sun and 400 times more luminous. It is also the namesake of AMD’s upcoming Instinct Accelerator.
AldebaranThe computer part has appeared several times in the Linux patch notes. Its two-array design was half-involved in February and confirmed last month: “In Aldebaran, only the primary array gets valid power data. It shows the power / energy values as 0 in the secondary array.” a note reads.
It is the first AMD Accelerator / GPU to take advantage of an MCM (Multi-Chip Module) design, a technique that has been included in patents for years but has only just begun to materialize. Using multiple closely connected chips / dies allows for greater scalability than traditional monolithic designs, but also reduces the performance per unit of compute on a card, particularly in poorly parallelized workloads.
Nvidia is expected to challenge Aldebaran with its own MCM designs based on the Hopper architecture next year. Intel, meanwhile, is about to launch the Ponte Vecchio accelerator based on the Xe HPC architecture and an early MCM implementation.
Like Ponte Vecchio, Aldebaran leverages MCM to expand memory capacity, the patch notes indicate. A. qualification since last week describes Aldebaran has two matrices, four UMC per matrix and eight channels connected to 2 GB of HBM per UMC. Or 128GB, in total.
The patch notes also mention support for a “new type of HBM2 memory”, presumably HBM2e. If AMD is using SK Hynix’s new HBM2e memory modules with 3.6 Gbps of bandwidth and a 4096-bit bus on each die, then Aldebaran could have 3.6 TB / s of memory bandwidth.
For comparison, Aldebaran’s predecessor, the Instinct MI100 (previously named Arcturus, another star) has 32GB of HBM2 and 1.2TB / s of memory bandwidth. Nvidia’s A100 Accelerator can be configured with up to 80GB of 3.2Gbps HBM2e memory for 2TB / s of bandwidth.
AMD CEO Dr. Lisa Su has already said that the CDNA2 architecture, that is, Aldebaran, will be released later this year. Presumably, it will be released as the MI200 Accelerator.
“This year, as I said, we are putting together our next-generation CDNA architecture. Actually, this is a key component that allowed us to win the largest supercomputer tenders in the US,” Su saying at a conference in May. “… And we will launch the next generation of that architecture, in fact, at the end of this year. We are very excited about it. I think it has progressed extremely well.”
Header of Timothy dykes