High-performance AI computing with air-cooled flexibility
Powered by AMD Instinct™ MI355X GPUs with EPYC™ processors
-
AI Computing
Performance8 x MI355X GPUs
-
Flexible
DeploymentAir cooled
-
Enterprise
ScalabilityDual AMD EPYC™ CPUs,
PCIe 5.0 & modular design
A New Generation of AMD Instinct™ Performance
AMD Instinct™ MI355 vs. MI325
-
+33% VS. MI325GPU-to-GPU Bandwidth
-
+12.5% VS. MI325HBM3E Memory Capacity
-
1.9 X VS. MI325FP8 Dense Performance
PCIe 5.0 Expansion
High-speed I/O for storage and networking
288GB HBM3E and 8TB/s memory bandwidth per OAM (Open Accelerator Module)
1:1:1 connectivity between storage, GPU and NIC
Modular, tool-free design for simplified maintenance
Purpose-built 10U air-cooled architecture for 8 MI355X GPUs
15 x 80*80 Fan For AMD MI355X 8-GPU Baseboard
8 x 2.5" NVMe
2x 10Gb RJ45 Lan Port
1x Mgmt Port
1x VGA port
4x USB3.2 Gen1 ports
1x Power Button
GPU Sled
(6U)
CPU Sled
(4U)
2 x 2.5"NVMe/SATA*/SAS*
*RAID/HBA card required.
1 x PCIe Gen5 x 8 FHHL + 1 x PCIe Gen5 x 16, FHHL
6+2 3200W 54V CRPS
1x Locate Button
1x Power Button
GPU Sled Handle
8 x PCIe Gen5 x 16, LP
1 x PCIe Gen5 x 16, FHHL
Eight AMD Instinct MI355X GPUs
Built for large-scale AI inference and training
Balanced GPU-to-NIC Connectivity
1:1:1 connection between storage, GPU and NIC
AMD ROCm™ open software platform
An open ecosystem for AI development and deployment
Air Cooling Solution
10U air-cooled architecture for flexible data center deployment
Serviceability
Simplified maintenance for efficient IT operations
-
60 fan module
-
Rear riser module
-
Nic riser module
-
Nic riser module
-
Front IO module
-
80 fan module
-
Plungers
Management
Comprehensive IT infrastructure solutions
BMC
Remote server management
IT infrastructure-management software
Streamline IT operations via a single dashboard
Trusted Platform Module 2.0
Performance
Dual AMD EPYC™ 9005/9004 Series processors
-
128
Max CPU Cores
-
384
Max L3 Cache
-
128
PCIe Lanes per CPU
FAQs
-
What are the key features of the AMD Instinct™ MI355X GPUs in ASUS XA AM3A-E13?What are the key features of the AMD Instinct™ MI355X GPUs in ASUS XA AM3A-E13? ASUS XA AM3A-E13 features eight AMD Instinct™ MI355X GPUs built on AMD CDNA™ 4 architecture. Each GPU offers 288GB of HBM3E memory, up to 8TB/s memory bandwidth, and support for AI-optimized precision formats including MXFP4 and MXFP6. The eight-GPU platform provides 2.3TB of aggregate GPU memory for large-scale AI inference, training, and HPC workloads.
-
What is the difference between AMD Instinct™ MI325X and MI355X GPUs?What is the difference between AMD Instinct™ MI325X and MI355X GPUs? Compared with MI325X, AMD Instinct™ MI355X introduces AMD CDNA™ 4 architecture, increases HBM3E memory from 256GB to 288GB per GPU, and boosts memory bandwidth from 6TB/s to 8TB/s. MI355X also adds MXFP4 and MXFP6 support, offering greater flexibility for low-precision AI computing and increasingly complex generative AI workloads.
-
Does ASUS XA AM3A-E13 require liquid cooling for its eight AMD MI355X GPUs?Does ASUS XA AM3A-E13 require liquid cooling for its eight AMD MI355X GPUs? No. ASUS XA AM3A-E13 features a purpose-built 10U air-cooled architecture that accommodates eight AMD Instinct™ MI355X GPUs and dual AMD EPYC™ processors without requiring direct liquid cooling. This design offers an alternative for organizations seeking to deploy advanced AI computing capabilities within existing air-cooled data centers, subject to facility power and thermal requirements.
-
What AI workloads is ASUS XA AM3A-E13 designed for?What AI workloads is ASUS XA AM3A-E13 designed for? ASUS XA AM3A-E13 is designed for generative AI, large language model (LLM) inference, AI training, and high-performance computing (HPC). Its eight-GPU architecture, substantial HBM3E memory capacity, and PCIe 5.0 connectivity provide the computational and data-handling resources needed for memory-intensive models, scientific simulations, and enterprise AI applications.
-
What makes ASUS XA AM3A-E13 easier to maintain?What makes ASUS XA AM3A-E13 easier to maintain? ASUS XA AM3A-E13 features a modular, service-friendly design that simplifies hardware maintenance. Clearly marked service access points help IT technicians quickly identify components, while tool-free access to fan modules, NIC risers, and front I/O modules streamlines replacement and servicing. This design helps reduce maintenance complexity and improve operational efficiency in demanding data center environments.