Uvation MarketplaceMarketplace

Supermicro AS-8126GS-NB3RT GPU Server

Air-cooled 8U server pairing eight NVIDIA Blackwell Ultra GPUs with dual AMD EPYC 9005/9004 processors for AI training, inference and HPC.

GPUs
8×
GPUs
HBM3e
2.3 TB
HBM3e
CPU cores
384
CPU cores

AMD EPYC Power for Blackwell Ultra AI

Supermicro’s AS-8126GS-NB3RT pairs eight Blackwell Ultra GPUs with dual AMD EPYC processors in one air-cooled 8U chassis for AI training, inference and HPC.

Up to 384 CPU Cores per System

Dual AMD EPYC 9005 processors deliver up to 384 cores and 768 threads, supporting CPUs up to 500 W TDP.

Big Memory and Storage for Large Models

Up to 6 TB of DDR5-6400 across 24 DIMMs and 8 hot-swap E1.S NVMe bays keep datasets close.

Eight GPUs in One Air-Cooled Chassis

Eight GPUs linked at 1.8 TB/s each stay cool with up to 12 heavy-duty fans, with no liquid loops required.

8 GPUs. One NVLink Domain.
One Server.

Fifth-generation NVLink and NVSwitch connect all eight GPUs into one unified memory and compute domain.

Supermicro AS-8126GS-NB3RT 8U air-cooled GPU server
8× GPUs
NVIDIA HGX B300 Blackwell Ultra
2.3 TB
HBM3e GPU memory
1.8 TB/s
NVLink bandwidth per GPU
14.4 TB/s
Total NVLink bandwidth

Workloads

From LLM training to real-time inference and research, this server handles demanding AI and HPC workloads.

Large Language Model Training at Scale

72 petaFLOPS of sparse FP8 compute and 2.3 TB of HBM3e help teams train larger models faster.

Conversational AI and Real-Time Inference

108 petaFLOPS of dense FP4 compute speeds up reasoning, chatbots and long-context responses.

Drug Discovery and Scientific Research

Up to 384 AMD EPYC cores and 6 TB of DDR5 feed simulation, analytics, and research alongside AI.

Engineered to dominate

Blackwell Ultra Meets AMD EPYC

Eight Blackwell Ultra GPUs, fifth-generation NVLink and 800 Gb/s networking per GPU combine with AMD EPYC compute for peak AI performance.

Cutaway view of the air-cooled AS-8126GS-NB3RT chassis
144 PFLOPS
FP4 sparse (108 dense)
72 PFLOPS
FP8/FP6 sparse compute
2x
Attention vs. HGX B200
800 Gb/s
8× ConnectX-8 ports
Cutaway view of the air-cooled AS-8126GS-NB3RT chassis
144 PFLOPS
FP4 sparse (108 dense)
72 PFLOPS
FP8/FP6 sparse compute
2x
Attention vs. HGX B200
800 Gb/s
8× ConnectX-8 ports

The rack, itemized.

SUBSYSTEM

SPECIFICATION / CONFIGURATION DETAILS

Processor

2× AMD EPYC 9005 or 9004 processors

Operating System

ESXi 8.0U3, ESXi 9.0, Oracle Linux 9.6, RHEL 10.0

Chipset

System on Chip (SoC)

Accelerators / GPU

NVIDIA SXM: HGX B300 8-GPU (288GB)

Memory

24 DIMMs; up to 6 TB

Primary Storage

8 × front hot-swap E1.S NVMe drive bays

Secondary Storage

2 × M.2 NVMe boot drives

Security

Trusted Platform Module (TPM) 2.0; Silicon Root of Trust (RoT) – NIST 800-193 Compliant

Chassis

CSE-GP807TS-R000NP

Power Supply

6x 6600W Redundant (3 + 3) Titanium Level (96%) power supplies

Ports

8× OSFP; 2× 10GbE; USB; VGA; Mini DisplayPort

Slots / PCIe Architecture

2 PCIe 5.0 x16 FHHL slots

Bezel

Not specified (A separate front security bezel is not specified by Supermicro for this system)

Form Factor

8U Rackmount

Network Card

8x ConnectX®-8 SuperNICs BlueField®-3 DPU

Warranty

3-Year Labor, 1-Year Parts & 1-Year Advance Cross-Shipment Limited Manufacturer Warranty

Supermicro AS-8126GS-NB3RT server with 48V DC busbar power

Power and Networking Built for AI Scale

8 × 1U DC power shelves

Six 3+3 redundant supplies at 96% efficiency help keep the server online through a supply failure.

8× 800 Gb/s ConnectX-8 SuperNIC ports

Eight 800 Gb/s ConnectX-8 SuperNICs give each GPU its own path for fast, low-latency multi-node scale-out.

2× PCIe 5.0 x16 FHHL expansion slots

Two PCIe 5.0 x16 slots add room for BlueField-3 DPUs or extra network and storage cards.

Nuclear-Ready, Plug and Play AI Factories

Bypass grid constraints with self-contained, high-density inference ready infrastructure directly adjacent to clean, base load nuclear power.

Global Deployment Engine

Any Infra, Anywhere in the world.

Global Shipping and Deployment Capability treats advanced infrastructure as mobile, modular freight.

Packaged complex computing nodes, cooling manifolds, and power distribution units into certified, standard form factors.

360° Service Ecosystem

Unifying Legacy with Next-Gen Compute.

Vertically integrated service ecosystem that serves as an operational bridge between legacy environments and next-generation, AI-native infrastructure.

Finance and Leasing

Capital-as-a-Service for High-Density Compute

Convert upfront capital expenditures (CapEx) into predictable operating expenses (OpEx).

Uvation allows enterprises and sovereign entities to deploy multi-megawatt AI factories and advanced GPU clusters without liquidity constraint.