Dell PowerEdge GPU / AI: XE9680, XE9640, XE8640, R760xa
Dell’s dedicated AI line runs from the 2U R760xa, which fits PCIe accelerators alongside two mainstream Xeon processors, up to the 6U XE9680, which packs eight NVIDIA HGX, AMD Instinct MI300X or Intel Gaudi 3 accelerators on a shared baseboard. XE9640 (4U) and XE8640 (4U) sit between the two on GPU baseboard density and processor generation.
Technical specification
| Model | Form factor | Socket / processor generation | Max memory | Drive bays | GPU support | Power supply | Network |
|---|---|---|---|---|---|---|---|
| XE9680 | 6U | 2x 5th Gen Intel Xeon Scalable, up to 64 cores | 32x DDR5, 4 TB max | – | Up to 8x NVIDIA HGX (H100/H200/H20) or AMD Instinct MI300X or Intel Gaudi 3 | – | – |
| XE9640 | 4U | 2x 5th Gen Intel Xeon Scalable, up to 64 cores | 32x DDR5, 4 TB max | – | NVIDIA HGX baseboard GPUs | – | – |
| XE8640 | 4U | 2x 4th Gen Intel Xeon Scalable, up to 56 cores | 32x DDR5, 4 TB max | – | NVIDIA HGX baseboard GPUs | – | – |
| R760xa | 2U | 2x 4th/5th Gen Intel Xeon Scalable, accelerator-optimised | – | – | PCIe GPU accelerators | – | – |
Source of the values: Dell PowerEdge XE9680/XE9640/XE8640 spec sheets (delltechnologies.com). R760xa also appears in the Intel R-series family above; Dell lists it under both the mainstream and the AI/GPU line depending on configuration. Drive bay, power supply and network figures were not confirmed in the sources we checked for this GPU-focused line. An en dash means the manufacturer does not publish that value in the source we hold; we do not fill it in with an estimate. Confirm every figure against the current manufacturer document before it is written into a contract.
The codes we quote in this family are XE9680, XE9640, XE8640 and R760xa. Send us the workload and the constraints, and we will confirm availability, lead time and the exact configuration.
Which one should you choose?
HGX baseboard density is the real differentiator, not the U height
XE9680 and XE9640 both carry a shared NVIDIA HGX baseboard for up to eight GPUs with full NVLink interconnect between them; the interconnect bandwidth between GPUs matters more for large-model training than raw GPU count alone. R760xa’s PCIe accelerators do not share this baseboard interconnect.
XE8640 runs the previous Xeon generation, not a lesser feature set
XE8640 uses 4th generation Xeon Scalable (up to 56 cores) while XE9680/XE9640 use 5th generation (up to 64 cores); the GPU baseboard capability is otherwise comparable. Choose XE8640 only when the processor generation genuinely does not matter to the workload, since newer XE9640 stock is the safer long-term bet.
R760xa is the entry point, not a compromise
For inferencing or moderate acceleration rather than large-model training, R760xa’s 2U chassis and standard rack power draw are far easier to deploy into an existing row than a 6U HGX chassis with its higher power and cooling demands. Do not default to the biggest GPU chassis when the workload does not need baseboard-level interconnect.
Power and cooling planning has to happen before the order, not after
An 8-GPU HGX chassis can draw several times the power of a standard 2U server and typically needs a dedicated high-density power feed and, in many rooms, liquid cooling support. Confirm the room’s power and cooling capacity against the specific GPU model chosen (H100, H200, MI300X; each draws differently) before committing to a chassis.
Where it fits, and where it does not
Fits AI training or large-scale inferencing where GPU-to-GPU interconnect bandwidth and accelerator count drive the decision. Does not fit a room without dedicated high-density power and cooling already planned, and does not fit a workload that only needs modest acceleration; the R series (Intel/AMD) with an optional GPU is the cheaper, easier match there.
What we need in order to quote it
- Specific accelerator model needed (H100, H200, MI300X, Gaudi 3)
- Whether baseboard NVLink interconnect is required or PCIe acceleration is enough
- Room power feed and cooling capacity, including liquid cooling if applicable
- Processor generation preference (4th vs 5th Gen Xeon)
Request a Quote
Tell us the usable capacity, the protocol and the delivery country. A priced quote comes back within 48 hours, usually with a second option so the trade-off is visible.
