Specifications
GPU Option A
8× NVIDIA HGX B300 NVL8, 270 GB, 1,100 W, SXM6, NVLink + NVSwitch
GPU Option B
8× AMD Instinct MI355X, 288 GB, 1,400 W, OAM, Infinity Fabric
Processors
2× 5th Gen AMD EPYC 9005, up to 192 cores each
Memory
24 DDR5 DIMM slots, up to 6 TB RDIMM, 6,400 MT/s
Front Storage (E3.S)
Up to 16× E3.S NVMe Gen5 (245.76 TB)
Front Storage (U.2)
Up to 10× U.2 NVMe Gen5 (153.6 TB)
Boot Storage
Embedded BOSS-N1, 2× M.2 NVMe (HWRAID 0/1)
Form Factor
10U rack
Networking (B300)
8× ConnectX-8 OSFP embedded
OCP Slot
1× OCP 3.0, Gen5 x8
PCIe (MI355X)
Up to 12× Gen5 x16 FHHL
PCIe (B300)
Up to 4× 150 W Gen5 x16 FHHL
Power Supplies
12× 3,200 W Titanium, 200–240 VAC hot-swap redundant
Cooling
Air cooled; 15 GPU hot-swap fans + 5 CPU cold-swap fans
Dimensions
17.30 × 18.98 × 40.27 in (without bezel)
Weight
379.86 lb (MI355X) / 360.79 lb (B300)
Managed By
Dell iDRAC10
Operating Systems
Ubuntu Server LTS, Red Hat Enterprise Linux, SUSE Linux Enterprise Server
Overview
The Dell PowerEdge XE9785 is a 10U rack-scale platform built for large-scale generative AI, training and HPC. It accepts either 8 NVIDIA HGX B300 NVL8 GPUs (270 GB each, fully interconnected by NVLink with NVSwitch) or 8 AMD Instinct MI355X OAM accelerators (288 GB each with Infinity Fabric), so a single SKU serves both NVIDIA and AMD deployments.
Each configuration pairs the eight-way GPU complex with two 5th Generation AMD EPYC 9005 processors supporting up to 192 cores per socket, 24 DDR5 DIMM slots for up to 6 TB of 6,400 MT/s memory, and both E3.S and U.2 Gen5 NVMe storage options.
Power is delivered by twelve 3,200 W Titanium hot-swap redundant supplies, and cooling uses fifteen hot-swap GPU fans plus five cold-swap CPU fans. On the B300 configuration, eight embedded ConnectX-8 OSFP ports provide the high-bandwidth fabric needed for multi-node AI clusters.
Key Benefits
Dual-vendor GPU choice in one platform — deploy NVIDIA HGX B300 or AMD MI355X without re-architecting the rack. 2.1 TB of aggregate GPU memory (B300) or 2.3 TB (MI355X) supports the largest models, while NVLink/NVSwitch and embedded ConnectX-8 networking sustain the bandwidth multi-node training demands. Enterprise manageability comes via iDRAC10 and a full choice of Linux operating systems.
Applications
Large language model training and fine-tuning, generative AI inference at scale, HPC simulation, AI research clusters, and rack-scale AI factory deployments that require an 8-GPU NVLink domain.
Request a Quote — DELL POWEREDGE XE9785 — 8-WAY NVIDIA HGX B300 OR AMD MI355X AI SERVER
QS Compute — global B2B supply of AI computing hardware, edge AI systems and accelerators. Volume pricing, 15-day sample lead time.
Get Your Quote →