- Megapod AMD design emphasizes GPU count in raw and interconnection efficiency
- The NVIDIA packaging method complicates direct comparisons with the AMD approach
- Three racks will form the spine of AMD megapod, and each computing tray can combine summer CPU with MI500 accelerators
AMD seems to be preparing a new large -scale computer system aimed at high performance computing.
According to the industry analysis, it is projected that the so -called “megapodite” will house a large number of accelerators, positioning it as a direct response to the established “superpod” of Nvidia.
The first signs of their design point to an emphasis on the raw GPU count and interconnection efficiency, although the exact details remain speculative.
A system with 256 mi500 chips
The reports suggest that AMD will organize the megapod in three separate racks.
The two external racks are expected to contain 32 computing trays each, while the central shelf will contain 18 trays for network switches.
Within each computer tray, a single summer CPU can match four Mi500 instinct GPUs.
This design would deliver 32 CPU and 128 GPU for shelf, producing a total of 64 CPU and 256 GPU for the entire system.
The proposed design, sometimes called “UAL256”, reflects AMD’s intention to expand the capacity destined to overcome competitors.
The comparison between the AMD megapod and the Nvidia superpod is not simple.
On paper, the AMD system would offer 256 physical or logical GPU packages, while the Kyber VR300 NVL57 of NVIDIA lists only 144.
However, Nvidia’s architecture organizes four GPU per package, producing 576 GPU in general.
AMD dependence on single GPU packages means that its structure may seem greater in the physical package count, but not necessarily in total central density.
This complicates any attempt to declare a system as superior. For now, the megapod is positioned more as a counterweight than a clear jump ahead.
The central frame of the projected system is designed to change infrastructure.
Analysts expect the new AMD Vulcan Network cards, derived from the line thinking, will be implemented here.
The use of these cards could determine if the megapod can deliver enough bandwidth to make complete use of its heavy GPU settings.
While the physical design of racks and trays seems simple, the latency and network performance will play as critical role as raw GPU numbers.
The system is expected towards the end of 2027 to place it on the same development horizon as other HPC main facilities and large -scale data centers, including the German shepherd supercomputer.
AMD has recognized plans to combine summer CPUs, MI500 accelerators and network cards thinking Vulcano, but the exact details of the megapod still do not verify.
Through computer (originally in German)
