G4826Z5
4U 8x AMD Instinct™ MI355X Liquid-Cooled AI Training & Inference Server
The MiTAC G4826Z5 is a 4U dual-socket server built on the AMD EPYC™ 9755/ 9575F processors and engineered for AI Training, Inference, and HPC. Its defining feature is the 100% Direct Liquid Cooling (DLC) solution, which guarantees sustained peak performance from 8x AMD Instinct™ MI355X GPUs and high-TDP CPUs.
D928 mm
AMD EPYC™ 9755/ 9575F
PCIe Gen 5 x4 each
DDR5 6400 MT/s
PCIe Gen 5 x16
(1+1) 3,200W (12VDC)
Why G4826Z5U2BC-355X-575?
Zero-Throttling Direct Liquid Cooling
DLC ensures all high-TDP CPUs and GPUs run at peak frequency indefinitely, drastically accelerating AI job completion and ROI.
Extreme 8x AMD MI355X Compute Power
Integrate 8x top-tier GPUs with 288GB HBM3e each, providing the massive capacity required to train the world's largest LLMs.
High-Frequency EPYC Zen 5 Performance
The 5.0GHz clock speed guarantees lowest latency for sensitive AI/HPC, accelerating all single threaded I/O segments.
Superior DDR5-6400 Memory Bandwidth
24x DDR5 channels deliver ultra-high bandwidth to both the 256 cores and 8x GPUs, ensuring continuous, high-speed data flow.
Unrestricted PCIe Gen 5 I/O Fabric
12x PCIe 5.0 x16 slots offer unparalleled flexibility for integrating multiple 400G network cards to build low-latency clusters.
Ultra-Fast 8x NVMe U.2 Storage Access
8x hot-swap NVMe bays deliver critical low latency storage, accelerating model loading and data ingestion.
G4826Z5 Product Layout
Front View
Top View
Applications:
Low-Latency Financial Modeling & Trading
Optimized for latency-critical financial services requiring maximum single-core frequency and rapid risk analysis.
High-Speed AI Inference at Scale
Supports high-speed AI inference at scale with massive GPU compute for parallel model serving and ingestion.
HPC Workloads & Molecular Dynamics
Delivers exceptional CPU-to-GPU bandwidth via PCIe Gen 5 for time-sensitive HPC workloads and engineering research.
Database & In-Memory Analytics Acceleration
Ideal for massive-scale analytics acceleration with high memory bandwidth and rapid flash storage access.
