AMD Sheds More Light on 128-GPU MI355X DLC Rack That Will Deliver 2.4 Exaflop at FP4 Precision – Here's How It Compares with Nvidia's Flagship Vera Rubin
AMD has unveiled exciting details about its 128-GPU MI355X DLC rack, a powerhouse system capable of delivering 2.4 Exaflop at FP4 precision. The announcement came during Hot Chips 2025, highlighting how AMD’s MI350 series scales from node to rack while pushing the limits of high-performance computing. This new rack puts AMD directly in competition with Nvidia’s upcoming Vera Rubin flagship.
MI355X DLC Rack: A Closer Look
The MI355X DLC rack is AMD’s new 2OU solution featuring 128 GPUs, 36TB of HBM3e memory, and a peak throughput of 2.6 exaflops at FP4 precision. AMD also offers a slightly smaller 96-GPU EIA variant with 27TB of HBM3e, providing flexible options for large-scale AI and HPC workloads.
The rack integrates 5th Gen Epyc CPUs, MI350 GPUs, and AMD Pollara NICs, adhering to OCP standards and UEC-supported networking. High-speed Infinity Fabric bandwidth reaches up to 1075GB/s, ensuring seamless GPU-to-GPU communication across the system.
How AMD’s MI355X Compares to Nvidia Vera Rubin
Nvidia’s Vera Rubin system, arriving next year, is also a massive-scale accelerator platform designed for AI and HPC tasks. While AMD’s MI355X focuses on FP4 precision workloads with massive HBM3e memory and Infinity Fabric bandwidth, Vera Rubin emphasizes maximum scale with Nvidia’s latest GPU architectures and interconnects.
Early comparisons show AMD’s 2.4 Exaflop FP4 throughput rivals Nvidia’s flagship capabilities, positioning MI355X as a competitive option for AI research, simulations, and next-gen computing centers.
Scalability and Future Prospects
AMD highlighted how MI350 series GPUs can scale from single-node setups to full racks like MI355X, making it ideal for both research labs and hyperscale data centers. Node-to-rack scalability is a key feature, supported by high-speed interconnects and efficient memory usage.
With the MI355X, AMD continues to push the envelope in HPC innovation, offering a high-performance alternative to Nvidia while preparing for a market increasingly hungry for exascale AI and compute workloads.
The 128-GPU MI355X DLC rack positions AMD at the forefront of high-performance computing, offering unmatched FP4 precision throughput and scalability. When compared with Nvidia’s Vera Rubin, it shows AMD’s commitment to delivering competitive, large-scale AI infrastructure ready for 2025 and beyond.
