The 5-step pipeline.
How energy turns into a deployed, financed, offtake-matched cluster, and what SLYD does at every step.
How SLYD works →Platform
SLYD Cloud
New and recovered GPU systems.
NVIDIA and AMD systems through documented manufacturer and qualified channel supply, with financing and deployment coordinated on the same platform.
Explore GPU hardware →By accelerator
Infrastructure and services
Compute, hardware, and power in one book.
Browse available GPU capacity by accelerator, configuration, region, and price, or bring supply to qualified demand.
Open marketplace →Compute
Bring supply
For buyers
Start with an indicative structure.
Tell us deal size, structure, and offtake. Any range is preliminary and subject to underwriting, diligence, and documentation.
Open Configure →For lenders
GPU market trends and deployment playbooks.
Infrastructure best practices, hardware comparisons, and industry analysis from the SLYD team.
Read the blog →Company
HPE AI servers
HPE brings two lineages to AI infrastructure: the ProLiant Compute enterprise line and the Cray supercomputing line. They overlap on accelerators and diverge on fabric, cooling, and operating model.
What are HPE AI servers?
HPE offers GPU-accelerated ProLiant Compute and Cray systems for enterprise AI and HPC workloads. Selection depends on the exact current model, accelerator platform, scale, interconnect, storage, cooling, power, software, consumption model, and service requirements.
Two lineages, one accelerator catalog
The most useful first question with HPE is not which accelerator, but which line. That decision carries the fabric, the management model, and the operational expectations with it.
HPE ProLiant Compute
HPE's enterprise server line. ProLiant Compute XD685 is published as an 8-way GPU server built for the growing demands of enterprise AI, with flexibility in fabric, memory, storage, and operating system.
HPE Cray
HPE's supercomputing line. Cray XD670 is published as a 5U chassis with a single 2-CPU node and 8 NVIDIA H200 SXM5 GPUs, with PCIe Gen 5 slots supporting Slingshot 11, InfiniBand NDR, and Ethernet.
HPE AI systems and their accelerator options
Reproduced from HPE's published product pages. Only systems whose accelerator options could be read from an HPE source on the review date are listed. Confirm the current lineup and configuration with HPE.
| System | Form factor | Accelerators, as published by HPE |
|---|---|---|
| HPE ProLiant Compute XD685 | 8-way GPU server, 5U for direct liquid cooling and 6U for air | Choice of eight NVIDIA Blackwell Ultra (B300 HGX), NVIDIA Blackwell (B200), NVIDIA H200 Tensor Core GPUs, or AMD Instinct MI355X accelerators, with 5th Generation AMD EPYC processors |
| HPE Compute XD690 | Published within NVIDIA AI Computing by HPE | Eight NVIDIA Blackwell Ultra (B300 HGX) GPUs |
| HPE Cray XD670 | 5U chassis, single 2-CPU node, direct liquid cooling option | 8 NVIDIA H200 Tensor Core SXM5 GPUs |
HPE publishes further AI systems beyond these three. This page lists only what could be sourced to a current HPE product page at review time, on the principle that an unverified model list is worse than a short one. Ask for the current HPE QuickSpecs for any model under consideration.
The cooling choice changes the rack math
HPE publishes ProLiant Compute XD685 at 5U for direct liquid cooling and 6U for air. That is a concrete illustration of something that catches out a lot of rack plans: the same 8-way platform is a physically different system depending on how it is cooled, and the rack unit count moves with it.
The practical sequence is to establish what the facility can actually support, then choose the cooling variant, then count racks. Doing it the other way around produces a plan that does not survive the site survey. See cooling infrastructure and power infrastructure for what each option asks of the building.
Fabric choice is where the Cray line earns its place
HPE publishes Cray XD670 with eight PCIe Gen 5 half-height half-length slots supporting Slingshot 11, InfiniBand NDR, and Ethernet, providing direct switchable connections between the high-speed fabric, GPUs, NVMe drives, and CPUs.
For clusters where collective communication dominates, that flexibility is the reason to look at the Cray line rather than a general-purpose enterprise server. For estates that are mostly inference or mixed workloads, it is usually unnecessary complexity. The networking page works through how to make that call.
HPE services and SLYD process are separate things
Manufacturer services and consumption models
HPE publishes its own support, service, and consumption offerings, including its as-a-service and private cloud models. Scope, term, regional eligibility, and commercial structure are defined by HPE agreements and should be evaluated with HPE directly.
Sourcing process
SLYD evaluates a stated requirement, identifies which documented sourcing paths are available for the requested system, and confirms configuration, seller authority, provenance, condition, availability, and commercial terms for the specific transaction.
SLYD is an authorized reseller for Dell Technologies, HPE, Supermicro, Lenovo, Cisco and IBM, and sources through authorized distribution including Arrow, Ingram Micro and TD SYNNEX. Configuration availability and lead times are quoted at time of order. SLYD makes no claim of factory-direct status, does not sell or resell HPE consumption offerings, and publishes no universal support, warranty, or response-time commitment.
To request a configuration
- Workload and scale
- Preferred accelerator, if already chosen
- Node count and expected growth
- Deployment location
- Whether the site supports direct liquid cooling
- Available rack power
- Fabric preference or existing fabric
- Target timeline
- Compliance and export constraints
- Capital structure preference
HPE AI server questions
Which HPE systems support the accelerator we are evaluating?
HPE publishes ProLiant Compute XD685 as an 8-way GPU server with a choice of eight NVIDIA Blackwell Ultra (B300 HGX), NVIDIA Blackwell (B200), NVIDIA H200 Tensor Core GPUs, or AMD Instinct MI355X accelerators. HPE publishes Compute XD690 with eight NVIDIA Blackwell Ultra (B300 HGX) GPUs, and HPE Cray XD670 as a 5U chassis with eight NVIDIA H200 Tensor Core SXM5 GPUs.
How do air-cooled and liquid-cooled HPE configurations differ?
They are physically different systems. HPE publishes ProLiant Compute XD685 in a standard 5U form factor for direct liquid cooling and 6U for air, so the same 8-way platform occupies different rack space depending on the cooling choice. HPE Cray XD670 is published with a direct liquid cooling option. Rack planning has to follow the cooling decision, not precede it.
What is the difference between HPE ProLiant Compute and HPE Cray systems?
They come from different lineages and suit different jobs. ProLiant Compute is HPE's enterprise server line, and the XD685 sits there as a general-purpose 8-way AI server. HPE Cray comes from HPE's supercomputing line, and XD670 reflects that with high-speed fabric options including Slingshot alongside InfiniBand and Ethernet. Both can run AI workloads; the fabric and operational model differ.
Can SLYD sell HPE GreenLake or Private Cloud AI?
No. Those are HPE consumption and service offerings governed by HPE agreements, and no current record establishes a SLYD role in selling, reselling, or delivering them. Buyers evaluating those models should engage HPE directly. SLYD's involvement in any transaction is limited to what is written into executed documents for that project.
Does a product name on this page mean SLYD has it in stock?
No. HPE model names are reproduced from HPE's published product pages to explain the architecture choices available. Whether a specific configuration can be obtained, in what quantity, for what geography, and on what terms is confirmed at the transaction level against current records.
How are warranty, service, delivery, and commercial terms verified?
At the transaction level, for the specific equipment, quantity, and geography, with the confirmation date and scope stated. HPE service offerings are HPE's, with scope, term, and regional eligibility defined by HPE. SLYD does not resell, extend, or guarantee any HPE service and publishes no universal support or warranty claim.
What information is needed to request an HPE configuration?
The workload and scale, preferred accelerator if already chosen, node count, deployment location, available rack power, whether the site supports direct liquid cooling, target timeline, fabric preference, and any compliance or capital constraints that shape the design.
Can HPE systems be evaluated for equipment financing?
SLYD originates and arranges eligible equipment financing and operating lease opportunities with third-party capital providers. Approval, pricing, collateral, term, and documentation depend on the borrower, equipment, deployment, repayment source, and transaction structure. Financing is not guaranteed, and it is separate from any HPE consumption offering.
Sources for the HPE information on this page
- HPE, ProLiant Compute XD685 product page. Reviewed 18 August 2026. hpe.com ProLiant Compute XD685
- HPE, Compute XD690 product page. Reviewed 18 August 2026. hpe.com Compute XD690
- HPE, Cray XD670 product page. Reviewed 18 August 2026. hpe.com Cray XD670
- HPE, AI servers overview. hpe.com AI servers
Model names, form factors, and accelerator options are HPE's, reproduced as published on the review date. They are not SLYD guarantees of configuration, service, or availability. HPE is a trademark of its owner and is used here descriptively.
Request an HPE configuration
Share the workload, node count, site, cooling capability, and fabric preference. Those settle the line and the model faster than starting from the accelerator.
Page updated: August 18, 2026