Dell PowerEdge XE9640 vs XE8640
Dell PowerEdge XE9640
Dell PowerEdge XE8640
The short answer: the XE9640 and XE8640 are the same four-way NVIDIA HGX SXM GPU platform in two thermal packages, so the deciding question is your facility, not the silicon. Choose the XE9640, a 2U direct-liquid-cooled node, when you have liquid infrastructure and want maximum density and sustained performance at cluster scale. Choose the XE8640, a 4U air-cooled node, when you need the same GPU horsepower to drop into standard air-cooled racks with no plumbing. Uniqcli configures and delivers both, so the right pick is the one that matches your data center, workload, and deployment timeline.
Side by side
| Dell PowerEdge XE9640 | Dell PowerEdge XE8640 | |
|---|---|---|
| Cooling architecture | Direct liquid cooling (DLC), designed for cold plates and facility water | Air-cooled, high-airflow fan design, no liquid required |
| Form factor and density | 2U chassis, more accelerated nodes per rack | 4U chassis, larger footprint per node |
| GPU platform | Four-way NVIDIA HGX SXM with NVLink for training and HPC | Four-way NVIDIA HGX SXM with NVLink, same accelerated baseboard class |
| Facility requirements | Needs a coolant distribution unit, rack manifolds, and facility water | Drops into standard air-cooled racks, no plumbing to plan |
| Sustained thermal headroom | Liquid holds higher sustained clocks on long runs, quieter, lower cooling energy | Proven air cooling, ample headroom within standard rack airflow limits |
| Best deployment context | Greenfield or liquid-ready halls, large scale-out AI clusters | Existing air-cooled data centers, edge, colocation, first clusters |
| Time to stand up | Higher, since liquid infrastructure must be provisioned first | Lower, since it fits current facility support as-is |
| Host platform | Intel Xeon Scalable, high-speed memory and networking for GPU feeds | Intel Xeon Scalable, comparable CPU, memory, and I/O class |
| Federal readiness | TAA-compliant, available via NASA SEWP V, GSA, and GPC through Uniqcli | TAA-compliant, available via NASA SEWP V, GSA, and GPC through Uniqcli |
Shop these now
Live configurations from our catalog with partner pricing. Add to your cart to request a firm quote, or build a full BOM.
Dell PowerEdge XE9640
Dell PowerEdge XE8640
Choose Dell PowerEdge XE9640 when...
Pick the XE9640 when you are building for density and sustained AI training throughput. It packs the four-way NVIDIA HGX SXM GPU platform into a 2U direct-liquid-cooled chassis, so you fit more accelerated nodes per rack and hold higher sustained clocks under long training runs. This is the right node for greenfield or liquid-ready facilities with a coolant distribution unit and rack manifolds in place, for large scale-out clusters where rack count and PUE matter, and for teams standardizing on liquid to lower cooling energy at scale. If your data center already supports facility water, or you are designing one that will, the XE9640 gives you the tightest footprint and the best thermal headroom. Build the exact node and cluster on /bom, then send it to /quote.
Choose Dell PowerEdge XE8640 when...
Pick the XE8640 when you want the same four-way HGX SXM GPU horsepower without adding liquid infrastructure. It runs the four-GPU platform air-cooled in a 4U chassis, so it drops into standard high-airflow racks with no CDU, manifolds, or facility water to plan around. That makes it the faster path to stand up an accelerated node in an existing air-cooled hall, the simpler choice for a proof of concept or a first AI cluster, and the lower-complexity option for edge or colocation sites where liquid is not available. You still get NVLink-connected GPUs for training and inference, just in an envelope your current facility already supports. Spec it on /bom and price it through /quote when you are ready.
Neither node wins outright, because they answer different facility questions with the same GPU platform. If your hall supports liquid, or you are designing one that will, the XE9640 gives you tighter density, higher sustained performance, and better cooling efficiency across a large cluster. If you need that GPU horsepower now, in racks you already run, the XE8640 gets you there without a plumbing project. Many buyers land on a mix: XE8640 nodes to move fast today, XE9640 nodes as the facility goes liquid-ready. Uniqcli sells and supports both, is Dell-authorized and vendor-neutral, and can scope the node count, networking, and cooling path to your workload. Model both configurations side by side on /bom, then send the one that fits to /quote."
Talk to a specialistFrequently asked
What is the core difference between the XE9640 and XE8640?
Cooling and density. Both carry the same four-way NVIDIA HGX SXM GPU platform, but the XE9640 is a 2U direct-liquid-cooled node and the XE8640 is a 4U air-cooled node. The XE9640 needs facility water and packs more nodes per rack. The XE8640 runs in standard air-cooled racks with no liquid infrastructure.
Do I need liquid cooling in my data center to use the XE9640?
Yes. The XE9640 is built for direct liquid cooling, so it expects a coolant distribution unit, rack manifolds, and facility water. If your hall is air-cooled today and you are not ready to add liquid, the XE8640 delivers the same GPU platform in an envelope your current racks already support.
Do both run the same GPUs for AI training?
Both use the four-way NVIDIA HGX SXM accelerated platform with NVLink, so both target AI training, fine-tuning, and HPC. The practical difference at scale is sustained performance and density: liquid cooling on the XE9640 helps hold clocks on long training runs and fits more nodes per rack. Uniqcli confirms the exact supported GPU configuration on your quote.
Which is better for a federal AI cluster?
Both are TAA-compliant and available to federal buyers through Uniqcli via NASA SEWP V, GSA, and GPC. The choice comes down to the site: liquid-ready facilities and large clusters favor the XE9640, while existing air-cooled or edge sites favor the XE8640. We can align the configuration to your contract vehicle and facility on /quote.
Can I mix XE9640 and XE8640 nodes in one deployment?
Yes, and many buyers do. A common path is to deploy XE8640 air-cooled nodes now for speed, then add XE9640 liquid-cooled nodes as the facility becomes liquid-ready. Build both configurations side by side on /bom so you can compare rack density, cooling requirements, and total node count before you commit.
Build your Dell bill of materials.
Send us the requirement, the project, or an existing quote to beat. We come back with a validated, TAA-compliant Dell configuration and a real price, often below list.
[email protected] · Chicago, IL