Specifications
GPU Platform
NVIDIA HGX AI supercomputing baseboard, up to 8 SXM GPUs
GPU Options
NVIDIA H100, H200 or B100 (selectable at configuration)
GPU Interconnect
NVIDIA NVLink and NVSwitch — all-to-all among the 8 onboard GPUs
Scale-Out Fabric
Inventec switchboard for GPU Direct RDMA between servers at full bandwidth
Processors
Dual Intel Xeon Scalable (Eagle Stream) — 4th Gen and 5th Gen options
Memory
DDR5 SDRAM, 8-channel per socket
Expansion Bus
PCIe Gen 5.0
Power Architecture
Decoupled power sources — 12 V module for CPU, 54 V module for GPU
Cooling Options
Air-cooled and liquid-cooled configurations
Tray Design
CPU and GPU housed in separate trays for independent service and upgrade
Serviceability
Tool-less fan, power supply and NIC modules
Form Factor
Rackmount server, modular tray architecture
Management
Inventec BMC with IPMI 2.0 and Redfish
Compatibility
Designed to interoperate with adjacent server types in mixed data centre rows
Customisation
White-glove configuration programmes for specialised data centre requirements
Target Workloads
LLM training and inference, HPC simulation, video generation, agentic AI
Overview
The Inventec P8000IG6 is an eight-GPU server built on the NVIDIA HGX AI supercomputing platform. Customers choose H100, H200 or B100 SXM modules, and the baseboard's NVLink and NVSwitch fabric provides all-to-all bandwidth between the eight accelerators. Inventec's own switchboard extends that connectivity between servers, giving GPU Direct RDMA traffic a dedicated, full-bandwidth path rather than contending for host PCIe lanes.
The chassis is deliberately modular. The motherboard occupies a separate tray from the GPU assembly, so CPU generation refreshes and GPU upgrades can be staged independently instead of forcing a whole-server replacement. Fans, power supplies and NICs are tool-less for field service. Power is split into decoupled 12 V (CPU) and 54 V (GPU) modules, which improves both efficiency and fault isolation in a rack of eight-GPU systems.
Platform support spans dual 4th and 5th Gen Intel Xeon Scalable (Eagle Stream) processors with eight-channel DDR5 per socket and PCIe Gen 5.0 expansion, and both air-cooled and liquid-cooled variants are available depending on rack power and density targets.
Key Benefits
Full eight-GPU NVLink domain: NVSwitch gives every accelerator in the node a direct path to every other, which is essential for tensor-parallel training. Switchboard scale-out: GPU Direct RDMA between servers runs at full bandwidth instead of through the host. Independent tray upgrade: CPU and GPU refresh cycles are decoupled, extending the platform's useful life. Fault isolation on power: separate 12 V and 54 V modules limit blast radius from a PSU fault. Deployment flexibility: air or liquid cooling lets the same platform fit existing rows or new high-density zones.
Applications
Large language model pretraining and fine-tuning, AI inference serving clusters, scientific and engineering HPC simulation, generative video and image workloads, agentic AI systems, and ODM/OEM AI factory build-outs where a modular eight-GPU platform is required.
Request a Quote — INVENTEC P8000IG6 — 8-GPU NVIDIA HGX AI AND HPC SERVER PLATFORM
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →