MSI CG480-S6053 NVIDIA MGX 4U GPU Server Platform

The MSI CG480-S6053 is a 4U GPU server platform built on NVIDIA MGX modular architecture, engineered for high-performance computing and AI workloads. A switching PCIe topology supports up to eight dual-slot GPU accelerators in a single 4U rackmount chassis, driven by dual AMD EPYC 9005 Series processors at up to 500W TDP. The MGX modular design allows GPUs, CPUs, and DPUs to be integrated flexibly, so system configurations can be tailored to AI training, inference, and data analytics.
The platform provides 24 DDR5 DIMM slots across 12 channels per CPU (1DPC), accepting RDIMM and RDIMM-3DS modules up to 256GB each at 6400MT/s. Expansion covers eight PCIe 5.0 x16 FHFL double-wide slots and four single-wide slots through the PCIe switch, plus one x16 slot direct from CPU0. Storage is eight front hot-swap 2.5-inch U.2 PCIe 5.0 x4 NVMe bays with two M.2 2280/22110 PCIe 3.0 x2 slots; optional NVMe RAID 0/1 is available on the S6053G480RAU8R SKU. Networking is two 10GBase-T ports on an Intel X710-AT2 with NCSI support, alongside a dedicated 1000Base-T management port.
Management runs on an ASPEED AST2600 BMC with AMI MegaRAC firmware supporting IPMI 2.0 and DMTF Redfish, with dual BIOS, dual BMC, and eMMC local storage. Security includes chassis intrusion detection, on-board TPM 2.0, and optional Microchip CEC1736 hardware root of trust. Power is 3+1 redundant 3200W CRPS 80 PLUS Titanium, cooled by two EVAC air modules rated for 500W CPUs. The chassis measures 438.5 mm wide, 175 mm high, and 800 mm deep, and is FCC Class A and CE certified.
GPU Partner supplies this platform fully integrated. Two validated build configurations are offered: eight NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs with dual EPYC 9555 and 768GB of DDR5, or eight NVIDIA H200 NVL GPUs with dual EPYC 9575F and 1,536GB of DDR5. Every system ships after full L6-L10 assembly, burn-in, and validation testing with a 3-year system warranty.
- Form Factor
- 4U Rackmount
- Architecture
- NVIDIA MGX
- GPU Topology
- Switching, up to 8 GPUs
- Standard SKU
- S6053G480RAU8
- RAID SKU
- S6053G480RAU8R
- CPU Support
- Dual AMD EPYC 9005
- Socket
- SP5, dual-socket (2P)
- Max CPU TDP
- Up to 500W
- DIMM Slots
- 24x DDR5
- Channels
- 12 per CPU (1DPC)
- Module Types
- RDIMM / RDIMM-3DS
- Max Frequency
- 6400MT/s
- Max per DIMM
- 256GB
- Double-Wide
- 8x PCIe 5.0 x16 FHFL
- Single-Wide
- 4x PCIe 5.0 x16 FHFL
- Direct from CPU0
- 1x PCIe 5.0 x16 FHFL
- Routing
- Via PCIe switch
- GPU Cards Supported
- Up to 8
- Front Bays
- 8x hot-swap 2.5"
- Bay Interface
- U.2 PCIe 5.0 x4 NVMe
- Bay Routing
- Via PCIe switch
- M.2 Slots
- 2x 2280/22110
- M.2 Interface
- PCIe 3.0 x2 NVMe
- RAID
- NVMe RAID 0/1 (opt.)
- Onboard LAN
- 2x 10GBase-T
- LAN Controller
- Intel X710-AT2
- NCSI
- Supported
- Management LAN
- 1x 1000Base-T
- Management Port
- 1x 1000Base-T
- Ethernet
- 2x 10GBase-T
- USB
- 1x USB 3.0/2.0 Type-A
- Video
- 1x Mini DisplayPort
- Power LED Button
- 1
- UID LED Button
- 1
- BMC Chip
- ASPEED AST2600
- Firmware
- AMI MegaRAC
- Protocols
- IPMI 2.0, Redfish
- BIOS Redundancy
- Dual BIOS
- BMC Redundancy
- Dual BMC
- BMC Storage
- eMMC (local)
- Chassis Intrusion
- Detection
- TPM
- On-board TPM 2.0
- Root of Trust
- Microchip CEC1736
- RoT Availability
- Optional
- Power Supplies
- 3+1 Redundant
- Wattage
- 3200W CRPS
- Efficiency
- 80 PLUS Titanium
- Cooling Modules
- 2x EVAC air cooling
- Max CPU Cooling
- 500W
- Form Factor
- 4U Rackmount
- Width
- 17.2 in. (438.5 mm)
- Height
- 6.88 in. (175 mm)
- Depth
- 31.4 in. (800 mm)
- Operating Temp
- 10°C – 35°C
- Operating Humidity
- 8–85% non-cond.
- Storage Temp
- -30°C – 60°C
- Storage Humidity
- 8–90% non-cond.
- Certifications
- FCC Class A, CE
- AI Training
- Training, Fine-Tuning
- Inference
- LLM Inference
- Visualization
- NVIDIA Omniverse
- CPU
- 2x AMD EPYC 9555
- Architecture
- Zen 5 “Turin”
- Cores per CPU
- 64C / 128T
- Total Cores
- 128C / 256T
- Base Clock
- 3.2GHz
- Max Boost
- Up to 4.4GHz
- L3 Cache
- 256MB per CPU
- Total L3 Cache
- 512MB
- TDP
- 360W (320–400W cfg.)
- GPU
- 8x RTX PRO 6000
- Edition
- Blackwell Server
- GPU Memory
- 96GB GDDR7 ECC each
- Total GPU Memory
- 768GB
- Memory Bandwidth
- Up to 1,597GB/s
- CUDA Cores
- 24,064 per GPU
- Tensor Cores
- 752 (5th gen)
- RT Cores
- 188 (4th gen)
- FP32 Performance
- 120 TFLOPS peak
- Host Interface
- PCIe Gen5 x16
- Interconnect
- 128GB/s bidirectional
- NVLink
- Not supported
- GPU Power
- Up to 600W each
- GPU Cooling
- Passive
- System Memory
- 768GB DDR5 RDIMM
- ECC
- Yes
- Modules
- 12x 64GB
- Slots Populated
- 12 of 24
- Speed
- Up to 6400MT/s
- CPU
- 2x AMD EPYC 9575F
- Architecture
- Zen 5 “Turin”
- Cores per CPU
- 64C / 128T
- Total Cores
- 128C / 256T
- Base Clock
- 3.3GHz
- Max Boost
- Up to 5.0GHz
- All-Core Boost
- 4.5GHz
- L2 Cache
- 64MB per CPU
- L3 Cache
- 256MB per CPU
- Total L3 Cache
- 512MB
- TDP
- 400W (320–400W cfg.)
- GPU
- 8x NVIDIA H200 NVL
- Architecture
- Hopper
- GPU Memory
- 141GB HBM3e each
- Total GPU Memory
- 1,128GB
- Memory Bandwidth
- 4.8TB/s per GPU
- FP8 Performance
- 3,958 TFLOPS
- FP16/BF16
- 1,979 TFLOPS
- Sparsity
- With sparsity
- Host Interface
- PCIe Gen5 x16
- Interconnect
- 128GB/s per GPU
- NVLink Bridges
- 2x 4th-gen
- Bridge Topology
- 2-way / 4-way
- Bridged Bandwidth
- Up to 900GB/s
- Form Factor
- Dual-slot, air-cooled
- GPU TDP
- Up to 600W each
- System Memory
- 1,536GB DDR5 RDIMM
- ECC
- Yes
- Modules
- 24x 64GB
- Slots Populated
- 24 of 24 (all)
- Speed
- Up to 6400MT/s
- Integration
- Full L6-L10 assembly
- Testing
- Burn-in & validation
- Warranty
- 3-year system
- Certification
- Red Hat catalog
- RHEL Support
- RHEL 9.5+ / AI 1.5+