- AMD MegaPod design emphasizes raw GPU count and interconnect efficiency
- Nvidia’s packaging method complicates direct comparisons with AMD’s approach
- Three racks will form the backbone of AMD’s MegaPod, and each compute tray may combine Verano CPUs with MI500 accelerators
AMD appears to be preparing a new large-scale computing system aimed at high-performance computing.
According to industry analysis, the so-called “MegaPod” is projected to house a large number of accelerators, positioning it as a direct response to Nvidia’s established “SuperPod.”
The first signs of its design point to an emphasis on raw GPU count and interconnect efficiency, although exact details remain speculative.
A system with 256 MI500 chips
Reports suggest that AMD will arrange the MegaPod across three separate racks.
The two outer racks are expected to hold 32 compute trays each, while the central rack will contain 18 trays for networking switches.
Within each compute tray, a single Verano CPU may be paired with four Instinct MI500 GPUs.
This layout would deliver 32 CPUs and 128 GPUs per rack, producing a total of 64 CPUs and 256 GPUs for the entire system.
The proposed design, sometimes referred to as “UAL256,” reflects AMD’s intent to expand capacity in ways meant to outmatch competitors.
The comparison between AMD’s MegaPod and Nvidia’s SuperPod is not straightforward.
On paper, the AMD system would offer 256 physical or logical GPU packages, while Nvidia’s Kyber VR300 NVL576 lists only 144.
However, Nvidia’s architecture organizes four GPUs per package, yielding 576 GPUs overall.
AMD’s reliance on single-GPU packages means its structure may appear larger in physical package count but not necessarily in total core density.
This complicates any attempt to declare one system as superior. For now, the MegaPod looks positioned more as a counterbalance than a clear leap ahead.
The central rack of the projected system is designed for switching infrastructure.
Analysts expect AMD’s new Vulcano network cards, derived from the Pensando line, to be deployed here.
The use of these cards could determine whether the MegaPod can deliver enough bandwidth to make full use of its GPU-heavy configuration.
While the physical layout of racks and trays seems straightforward, network latency and throughput will play as critical a role as raw GPU numbers.
The system is expected toward the end of 2027, placing it on the same development horizon as other major HPC installations and large-scale data centers, including the German Herder supercomputer.
AMD has acknowledged plans to combine Verano CPUs, MI500 accelerators, and Pensando Vulcano network cards, but exact details of the MegaPod remain unverified.
Via Computerbase (originally in German)
You might also like
https://cdn.mos.cms.futurecdn.net/koGpTooCEnSZDJLhKgXx54.jpg
Source link
