Dell Blackwell lineup: PowerEdge XE9680L, XE9780, and XE9685L systems.

When a regulated enterprise or a large account standardizes on a single GPU server platform, the decision is rarely about peak FLOPS. It is about who underwrites the deployment for the next five years. That is the case Dell makes with its PowerEdge XE Blackwell line. The silicon inside an XE9680L, an XE9780, or an XE9685L is the same NVIDIA HGX board you can buy from a dozen other OEMs. What is different is the wrapper: global ProSupport, factory L11 rack integration, validated reference architectures, and a balance sheet large enough to finance a multi-rack buildout. For the buyer who has to satisfy a procurement committee and an audit trail, that wrapper is the product.

This is a vendor spotlight, not a partnership announcement. Dell is one of the qualified OEMs whose complete systems Rillor can source as standardized forward contracts. We profile the PowerEdge XE Blackwell lineup here so a buyer evaluating Dell can map a specific chassis to a specific Rillor SKU, understand the air-versus-liquid tradeoff, and see where rack-scale GB200 and GB300 delivery fits.

The node-level lineup

Dell's Blackwell node story is three chassis that cover the two relevant GPU generations and the two relevant CPU vendors. They share a 4U HGX form factor and an 8-way NVLink topology, and they differ on cooling, GPU generation, and head-node silicon.

PowerEdge XE9680L: liquid-cooled HGX B200

The XE9680L is a 4U liquid-cooled server supporting 8x NVIDIA HGX B200 (180GB, 1000W SXM) GPUs interconnected over NVLink, paired with dual 5th Gen Intel Xeon Scalable CPUs and up to 4TB of DDR5 across 32 DIMM slots. It is the direct-to-chip cooled member of the B200 family, and it is the one Dell put on the record with benchmarked results: the XE9680L posted the first MLPerf Training v5.0 numbers on the liquid-cooled HGX B200 platform. For a buyer who needs a validated reference rather than a datasheet, that distinction matters. The XE9680L is the node behind a published, third-party-comparable training result, not a spec sheet promise.

PowerEdge XE9685L: HGX B200 on AMD EPYC Turin

The XE9685L is the AMD answer in the same generation. It pairs two 5th Gen AMD EPYC 9005 series (Turin) processors, up to 192 cores each, with 8x NVIDIA HGX B200 GPUs in a 4U direct-liquid-cooled chassis. It took the 2025 CRN Product of the Year award in the AI Infrastructure category. The reason a buyer reaches for the XE9685L over the XE9680L is core count and memory bandwidth on the head node. Turin's 192-core parts give data-loading and preprocessing pipelines more room before the CPU becomes the bottleneck feeding eight Blackwell GPUs. If your workload is I/O-heavy on the host side, the EPYC variant earns its place. The deeper version of that argument is a CPU pairing question (Granite Rapids versus EPYC Turin for GPU server head nodes), and it turns almost entirely on how much host-side preprocessing your pipeline demands.

PowerEdge XE9780: air-coolable HGX B300 Blackwell Ultra

The XE9780 is the Blackwell Ultra node. It uses two 6th Gen Intel Xeon Scalable processors (Granite Rapids generation, up to 86 cores each) with 8x NVIDIA HGX B300 NVL8 (270GB, 1100W SXM6) GPUs, 32 DDR5 DIMM slots up to 4TB at 6400 MT/s, and twelve 3200W Titanium power supplies. The headline is that Dell offers it as an air-coolable design. For a buyer whose existing data hall is not plumbed for direct liquid cooling, the air-cooled XE9780 is the path to Blackwell Ultra without a facility retrofit. Dell quotes up to four times faster LLM training on the 8-way HGX B300 versus the prior generation. The liquid-cooled XE9780L and XE9785L variants exist for buyers who do have liquid in the hall and want the higher per-rack density.

If you are still deciding between generations, the node-level question is concrete: HBM3e capacity (180GB on B200, 270GB on B300), SXM power envelope (1000W versus 1100W), and what that delta buys you per dollar of forward commitment. For a buyer pricing both, the generational gap is real but the standardized contract is identical, which is what makes the two directly comparable on Rillor.

Air versus direct-liquid cooling, the Dell way

The cleanest way to read the Dell lineup is by cooling, because that is the decision that most often gates a deployment. The B200 nodes (XE9680L, XE9685L) are direct-liquid-cooled. The B300 node (XE9780) is offered air-coolable, with liquid-cooled XE9780L and XE9785L variants for higher density.

The reason this matters for a forward contract is lead time and site readiness. An air-cooled XE9780 can land in a conventional rack today; a liquid-cooled XE9780L or an IR7000 deployment requires manifolds, CDUs, and facility water. When you commit forward delivery, you are also committing to having the hall ready on the delivery date. The facility tradeoff (does the room reject heat to air or to a water loop) decides which delivery month you can actually accept, so the cooling choice is upstream of the contract, not downstream of it.

Rack-scale: IR7000 and Integrated Rack Scalable Systems

Node-level systems are how you buy eight GPUs. Rack-scale is how you buy seventy-two. For GB200 NVL72 and GB300 NVL72, Dell delivers through its Integrated Rack Scalable Systems program built on the IR7000 rack.

The IR7000 is an OCP-based 21-inch rack purpose-built for native direct liquid cooling. It cools deployments of up to 480kW per rack and captures nearly 100% of generated heat, which is what makes a 72-GPU NVLink domain thermally feasible. Dell's liquid-cooled XE9780L and XE9785L support up to 192 Blackwell Ultra GPUs with direct-to-chip cooling, and the IR7000 can be configured with up to 256 Blackwell Ultra GPUs per rack.

What separates the Dell rack-scale offering from buying loose nodes is the integration program. Dell Integrated Rack Scalable Systems combine L11 factory integration, L12 solution testing, and expert on-site deployment to deliver turnkey, plug-and-play rack-scale infrastructure, including configurations for GB200 NVL72 and GB300 NVL72. L11 means the rack is built, cabled, and burned in at the factory; L12 means the full solution is tested as a unit before it ships. For a buyer standing up a 72-GPU coherent domain, the alternative is integrating that yourself on the data hall floor, which is where schedules slip. If you are weighing the two generations at rack scale, see GB200 NVL72 versus GB300 NVL72.

The fabric and the supporting silicon

The networking is the standard Blackwell stack, and Dell sells the full path. The compute fabric uses ConnectX-7 (400G NDR) and ConnectX-8 (800G) NICs at the node. For the cluster fabric, Dell expanded its portfolio with PowerSwitch SN5600 Ethernet switches on the NVIDIA Spectrum-X platform and NVIDIA Quantum-X800 InfiniBand switches delivering up to 800 gigabits per second, alongside the Quantum-2 generation for existing InfiniBand fabrics. The choice between Ethernet and InfiniBand is workload-dependent and worth its own decision: lossless InfiniBand favors tightly synchronized large-model training, while Spectrum-X Ethernet favors operators who want one fabric across storage, management, and compute.

On the supporting silicon, the XE nodes use Samsung DDR5 RDIMMs (up to 4TB, 6400 MT/s on the XE9780) and Micron 9550 PRO NVMe for local storage. None of this is exotic. The point is that Dell ships a coherent, validated combination rather than a parts list, and every layer is covered by the same support contract.

What genuinely differentiates Dell

Strip away the silicon, which Dell shares with every HGX OEM, and the real differentiators are operational.

• Global ProSupport. A single support relationship that spans nodes, switches, and racks, with deployment services attached. For a regulated buyer, having one accountable vendor for the whole stack is often a hard requirement, not a preference.
• Factory L11 rack integration and L12 testing. The rack arrives built and tested, not as a pallet of components. This compresses the gap between delivery and first training run, which is the gap that actually costs money.
• Validated reference architectures. The XE9680L MLPerf Training v5.0 result is the visible tip of this. Dell publishes designs that have been benchmarked as a system, so a buyer is sizing against measured numbers.
• Financing scale. Dell can underwrite a multi-rack capex commitment in a way few OEMs can. For a buyer matching a forward delivery schedule to a financing schedule, that matters.

This is also where the forward-market case for Dell is strongest. A buyer who values Dell's lifecycle support is, by definition, planning years ahead. That is the buyer who benefits most from locking forward delivery at a known price instead of negotiating spot each quarter. The argument for treating capacity as a forward commitment, rather than a waitlist, is laid out in why serious GPU buyers need a forward market.

Mapping Dell systems to Rillor SKUs

Rillor lists standardized forward contracts on complete OEM GPU systems, and Dell's PowerEdge XE line maps cleanly onto the Rillor SKU catalog. The mapping is by GPU generation and form factor, not by cosmetic chassis differences:

• XE9680L and XE9685L (8x HGX B200) map to RIL-GX-B200-2T, the 2-ton-class B200 node SKU. The Intel and AMD head-node variants settle into the same Rillor SKU because the contract standardizes on the GPU system, with CPU and cooling captured as configuration.
• XE9780 and XE9780L (8x HGX B300) map to RIL-GX-B300-2T, the Blackwell Ultra node SKU.
• GB200 NVL72 racks map to RIL-NVL72-GB200, and GB300 NVL72 racks built on the IR7000 map to RIL-NVL72-GB300.

Every contract is a bilateral OTC forward with intent of physical delivery: 10% deposit at execution, balance at delivery, an independent escrow agent holding the deposit, a seller performance bond, and NVIDIA channel compliance with the end-customer-of-record captured for the OEM's NVIDIA business unit. These are forwards, not exchange-listed futures, and they always settle in hardware, never in cash. If a buyer's plans change before delivery, the contract can be transferred to another KYC'd buyer with Rillor and OEM approval. None of that is unique to Dell; it is the standard Rillor contract, applied to Dell systems. Buyers can see how the whole flow works on the buyer page.

One guardrail worth stating plainly. Rillor lists Dell systems because Dell is a qualified OEM in the ecosystem Rillor can source, not because of any signed or announced partnership between the two companies. If you want indicative pricing on a specific Dell configuration, the SKU mapping above is the place to start.

The XE9780 and the broader PowerEdge XE Blackwell line are a strong default for buyers who weigh lifecycle support and integration as heavily as raw silicon. On Rillor, that preference becomes a forward contract at a known price, with physical delivery and escrow on every line. Browse the live marketplace to see indicative forward pricing across the SKUs above.