LLM training
Large-language-model pre-training and fine-tuning cluster — typical deployment for AI-native enterprises and research labs.
Dell PowerEdge Servers · Server Hardware
Dell PowerEdge XE9680
8-GPU AI training server — H100, H200 and AMD MI300X.
Quick answer
Dell PowerEdge XE9680 is a 6U AI training server supporting 8 NVIDIA H100 or H200 GPUs (with NVLink and NVSwitch interconnect) or 8 AMD Instinct MI300X GPUs. It is Dell's enterprise alternative to NVIDIA DGX for large-language-model training, retrieval-augmented-generation inference and HPC workloads.
Dell PowerEdge XE9680 is a 6U server designed for AI training. It hosts 8 NVIDIA H100 or H200 SXM5 GPUs interconnected via NVLink and NVSwitch — the same HGX board used by NVIDIA DGX H100 — or 8 AMD Instinct MI300X GPUs with Infinity Fabric topology. The AMD MI300X option provides 192 GB of HBM3 per accelerator, allowing very large LLMs to fit entirely in single-card memory for inference.
InfiniBand 400G networking, 4 TB of DDR5 host memory, ten PCIe Gen5 expansion slots, optional liquid cooling and enterprise-grade serviceability make XE9680 a credible Dell alternative to NVIDIA DGX for enterprise AI infrastructure across the EU, UK and US. Multi-node training with InfiniBand 400G interconnect supports tensor-parallel and pipeline-parallel distributed training at production scale.
Key features
What it offers
- 8x NVIDIA H100/H200 SXM5 or AMD Instinct MI300X GPUs
- NVLink + NVSwitch (NVIDIA) or Infinity Fabric (AMD) topology
- 10 PCIe Gen5 expansion slots
- 32 DDR5 DIMM slots, up to 4 TB of memory
- InfiniBand 400G networking support
- Air or liquid cooling option
Tech Summary
Important technical data
- GPU
- 8x NVIDIA H100/H200 SXM5 or AMD MI300X
- Memory
- Up to 4 TB
- Networking
- InfiniBand 400G
- Cooling
- Air or liquid
Use CasesAI University research Financial services
When would you choose this product?
Research HPC
University AI research laboratory and national HPC facility deployment.
AI risk modeling
Bank in-house AI risk and pricing model training.
Who is it for?
AI training, LLM and HPC customers across the EU, UK and US.
Frequently Asked Questions
Frequently asked questions
How does XE9680 compare with HPE Compute XD685?
Very similar — both platforms use the same NVIDIA HGX H100/H200 baseboard. The deciding factor is management experience (Dell iDRAC vs HPE iLO + OneView) and regional service network.
What is the advantage of the AMD MI300X option?
192 GB of HBM3 memory per accelerator — large LLMs fit entirely in single-card memory for inference without partitioning. The MI300X is a credible alternative to H100 for inference-heavy workloads.
What is the GPU lead time?
H100 and H200 supply remains constrained; typical order-to-delivery is 8-16 weeks. MI300X is generally faster to source. Sora Yazilim plans GPU procurement well in advance against customer pipeline.
Is liquid cooling required?
Air cooling is available, but with 8x H100/H200 the thermal load is high; liquid cooling improves data-center PUE and is increasingly the default for new builds.
Multi-node distributed training?
Yes — InfiniBand 400G interconnect supports tensor-parallel and pipeline-parallel distributed training across multiple XE9680 nodes.
Warranty?
Three years ProSupport Plus is strongly recommended for any AI training cluster — the platform value and operational criticality justify the top-tier SLA.
Vendor's official product page
Opens the vendor's original technical documentation and product page in a new tab.
Dell PowerEdge Servers — Dell PowerEdge XE9680 →From the same brand
Dell PowerEdge Servers family — other products
Dell PowerEdge R760
2U dual-socket flagship rack server — the enterprise workhorse.
DetailsDell PowerEdge R660
1U dual-socket — the dense data-center rack server.
DetailsDell PowerEdge R860
4-socket enterprise database server.
DetailsDell PowerEdge R7625
AMD EPYC 9004/9005 2U dual-socket — high-core-density rack.
DetailsComparable Solutions
Similar solutions from other brands
Dell PowerEdge XE9680 licensing + deployment + support
Sora Yazılım handles licensing, deployment, training and ongoing management — all from a single team.