top of page
PAE Soltion Banner.jpg

HighPoint PCIe Accelerator Expansion (PAE) Solution Bundles

Deploy Multi-Slot Blackwell RTX PRO GPUs in Standard 1U & 2U Server Racks Using PAE Expansion Solutions

Enterprise workgroups across local 70B+ AI inference, medical 3D imaging (Cryo-EM, CT/MRI), 8K/12K virtual production, and CAE/CFD simulation rely on standard 1U and 2U rack servers for multi-user collaboration, remote management, and high-speed networking. While these data-intensive applications demand the 96 GB memory of the NVIDIA® RTX™ PRO 6000 Blackwell Workstation Edition (600W), standard server risers typically feature conventional single-slot spacing (20.32 mm / 0.8 inches). Attempting to install a multi-slot workstation GPU internally can block adjacent expansion slots, interfere with internal chassis components, generate high acoustic noise, or exceed server power and cooling limits.

 

HighPoint PCIe Accelerator Expansion (PAE) Solution Bundles eliminate this deployment barrier by externalizing workstation accelerators into a dedicated chassis. Using only a single-width server slot, the bundle connects your server to an external RocketStor 8631D enclosure over a high-speed CopprLink™ cable. This architecture leaves neighboring server slots completely open for 100GbE networking and storage, while the RocketStor 8631D enclosure independently powers and cools the 600W+ GPU at a quiet 65.9 dBA—protecting your server from heat, noise, and electrical strain.

Find Your PAE Solution

Select your deployment architecture to identify the appropriate PAE Solution Bundle:

Workflow Requirements
Recommended Solution
Do you need a direct path between NVMe storage (rackmount/mobile rack backplanes) with a GPU (GPU-Direct Storage)?
RocketStor 8631DH - Integrated MCIO x8 Ports + GDS Storage Ingestion
Do you need support for multiple GPUs with direct Peer-to-Peer (P2P) DMA?
RocketStor 8631DD - Dual-GPU Ready P2P Expansion Solution
Do you need to added dedicated GPU compute via single-width 1U/2U server risers?
RocketStor 8631DW - Single External GPU Expansion Solution

HighPoint PCIe Accelerator Expansion (PAE) Solution Bundles

Select Your PAE Bundle

Selection Criteria
RocketStor 8631DW   
RocketStor 8631DH   
RocketStor 8631DD   
(Single-GPU Offload Bundle)
(Compute & Storage Bundle)
(Dual-Port Scalable Bundle)

Application Deployment Ideas

• Local 70B+ LLM Inference (vLLM)   

• Edge Multi-Camera Video Analytics   

• Interactive 3D CAD/CAM Workstations

• 3D Medical Imaging (Cryo-EM, CT/MRI)   

• 8K/12K Virtual Production (UE5)   

• High-Throughput Model Checkpointing

• CAE / CFD Simulation (OpenFOAM, Ansys)   

• Distributed Multi-GPU AI Scaling   

• Large-Mesh Finite Element Solvers

Server Deployment Requirement

Standard Server with One PCIe Slot (Single Width or wider)

(Low-Profile or Full-Height Bracket Supported)

Standard Server with One PCIe Slot (Single Width or wider)

(Low-Profile or Full-Height Bracket Supported)

Standard Server with One PCIe Slot (Single Width or wider)

(Full-Height Bracket; Horizontal Riser Required for 1U)

GPU Form-Factor & Capacity

Accommodates 1x Multi-Slot GPU

(Up to 370 mm L × 170 mm H × 88 mm W)

Accommodates 1x Multi-Slot GPU

(Up to 370 mm L × 170 mm H × 88 mm W)

Accommodates 1x Multi-Slot GPU Out of Box

Scalable to 2x Multi-Slot GPUs (via 2nd chassis)

Internal NVMe Storage Expansion

None (Dedicated 1:1 GPU Compute)

2x MCIO x8 Ports (SFF-TA-1016)

Direct connection to 4x4 NVMe backplanes

None (Dedicated Dual-GPU Compute)

Primary Architectural Value

1:1 Dedicated PCIe Gen5 x16 Pipeline

Lowest-cost entry point to externalize a single multi-slot workstation GPU from a single-width server slot with dedicated 64 GB/s throughput

Direct GPU-to-Storage P2P Routing via PCIe Gen5 Switch Fabric

Dedicated Gen5 x16 connectivity for GPU and internal NVMe storage backplanes

Direct GPU-to-GPU P2P Routing via PCIe Gen5 Switch Fabric

Dual dedicated Gen5 x16 GPU connections for unshared multi-GPU scaling

Included Hardware in Box

• 1x Rocket 7634D Host AIC (LP & FH)   

• 1x RS8631D 1300W Enclosure (65.9 dBA)   

• 1x 1m CopprLink Cable + 12VHPWR Cable

• 1x Rocket 7638D Host AIC (LP & FH)   

• 1x RS8631D 1300W Enclosure (65.9 dBA)   

• 1x 1m CopprLink Cable + 12VHPWR Cable

• 1x Rocket 1638E Host AIC (FHHL)   

• 1x RS8631D 1300W Enclosure (65.9 dBA)   

• 1x 1m CopprLink Cable + 12VHPWR Cable

MSRP (USD)

$2,298.00

$2,798.00

$3,298.00 

(Optional 2nd Enclosure: $1,299.00)

图片1_edited.jpg

The Three Architectural Pillars of HighPoint PAE

Category table.png
PCIe 5.0 Switch.png

1. Broadcom 48-Lane Switching Host Adapters with Real-Time Telemetry

​

HighPoint PAE host Switch Adapters conform strictly to industry standard PCI-SIG CEM Low-Profile and Full-Height specifications, occupying strictly one single-width PCIe slot. They install cleanly into standard 1U/2U server riser sleds without chassis lid interference, preserving all adjacent riser slots for 100GbE networking or storage. 

  • Non-Blocking Switching Fabric: Powered by Broadcom 48-lane PCIe Gen5 switching silicon, each Switch Adapter delivers dedicated, non-blocking lanes to ensure maximum bus saturation without relying on motherboard slot bifurcation.

 

  • In-Band Hardware Telemetry: Through HighPoint’s MPT (Management & Performance Tool) CLI/GUI utility, administrators gain in-band hardware health monitoring, link-speed verification, real-time thermal diagnostics, and seamless firmware updates across Windows, Linux, and FreeBSD environments. 

图片1.avif

2. Native PCI-SIG® CopprLink™ Cabled Interconnect

 

The PAE architecture uses certified PCI-SIG® CopprLink™ (CDFP / SFF-TA-1032) cabled interconnects to extend the high-speed PCIe bus outside the server. 

  • Uncompromised 64 GB/s Throughput: Delivers pure, unencapsulated PCIe Gen5 x16 bandwidth across 16 high-speed differential pairs with zero protocol translation, network encapsulation, or packetization overhead.

​

  • Chassis Decoupling: Completely bypasses the mechanical depth, height, and thermal limits of 1U/2U server chassis while maintaining true end-to-end PCIe Gen5 physical-layer performance. 

ICON.png

3. RocketStor 8631D Retimer Accelerator Enclosure: Signal Conditioning, 1300W Isolated Power & Engineered Acoustic Design

 

The RocketStor 8631D unifies physical accommodation, signal regeneration, power delivery, and acoustic control into an independent expansion platform: 

  • Built-in Astera Labs PCIe 5.0 Active Retimer: Integrated directly onto the enclosure backplane and powered internally, the retimer eliminates jitter and compensates for signal attenuation across external CopprLink connections. Full SRIS (Separate Reference Independent Spread Spectrum Clocking) support locks link negotiation at line-rate 32 GT/s Gen5 speeds, eliminating down-training and packet re-transmissions.

​

  • Dedicated 1300W Industrial CRPS Power Delivery: Each RocketStor 8631D is equipped with an independent 1300W Common Redundant Power Supply (CRPS) equipped with native 12VHPWR / 12V-2x6 high-current wiring harnesses. This platform delivers sustained 600W+ capacity and absorbs peak dynamic current spikes up to 2x rated TDP, isolating accelerator power demands entirely from the server.

​

  • Acoustically Engineered Thermal Architecture (65.9 dBA): Designed specifically for lab, studio, and edge environments, the chassis integrates an acoustically tuned industrial PSU and intelligent high-static-pressure dual exhaust fans. The automated thermal curve ramps cooling efficiently based on real-time thermals—preventing GPU thermal throttling under sustained 100% matrix load while maintaining a controlled sound profile of 65.9 dBA.

​

  • Expansive Multi-Slot Clearance: Provides an open internal chamber supporting workstation-class and AI accelerators up to 370 mm in length, 170 mm in height, and multi-slot widths up to 88 mm.

Pre-Sales Compatibility & Verification Checklist

Server Riser Slot Clearance:

  • RocketStor 8631DW / 8631DH: Requires 1x PCIe Gen5 or Gen4 x16 slot supporting standard PCI-SIG CEM Low-Profile (68.9 mm) or Full-Height (110 mm) brackets (fits standard 1U and 2U server risers).

  • RocketStor 8631DD: Requires 1x PCIe x16 slot supporting a Full-Height Half-Length (FHHL) card (compatible with 1U horizontal/butterfly risers and standard 2U+ full-height riser sleds).

 

Internal Backplane Interface (RS8631DH Only):

  • Confirm that your server's front-panel NVMe drive cage supports standard SFF-TA-1016 (MCIO x8) cabling to link directly into the Rocket 7638D internal storage ports.

 

Dedicated Power Budget Allocation:

  • Ensure an independent 100V–240V AC line or rack PDU receptacle is available for each 1300W RocketStor 8631D enclosure. The accelerator draws 0W from the host server power plane.

 

Order Evaluation or Request Formal Quotation:

  • HighPoint PAE Solution Bundles ship late October 2026 through authorized global distributors (Arrow Electronics, ASI, TD SYNNEX, MA Labs, CDW, B&H, and Amazon Business).

Ready to Modernize Your AI Infrastructure?

Verify your server platform, riser configuration, and NVIDIA Blackwell Workstation GPU models with HighPoint’s engineering team before deployment.

Verify your server platform, riser configuration, and NVIDIA Blackwell Workstation GPU models with HighPoint’s engineering team before deployment.

Request a Complimentary Deployment Consultation Review

Frequently Asked Questions: HighPoint PAE Solution Bundles

  • Standard 1U and 2U server risers typically enforce a conventional single-slot pitch - 20.32 mm (0.8 inches). As a result, installing a multi-slot workstation accelerator would physically block the neighboring expansion slots, disrupt chassis airflow, and risk tripping server power supplies under heavy loads. HighPoint PAE bundles resolve this by placing a single-width host adapter into the server and externalizing the multi-slot card into an independent RocketStor 8631D enclosure over CopprLink™ cabling—leaving adjacent server slots free for 100GbE networking and storage.

  • Each bundle targets a distinct architecture based on connectivity and workload requirements:

    • RocketStor 8631DW: Dedicated 1:1 PCIe Gen5 x16 pipeline engineered for single-GPU offload at the lowest entry cost.

    • RocketStor 8631DH: Converged compute and storage bundle featuring 2x internal MCIO x8 ports alongside the GPU link to enable direct NVIDIA GPUDirect Storage (GDS) with front NVMe drive backplanes.

    • RocketStor 8631DD: Scalable dual-port compute bundle featuring dual dedicated Gen5 x16 GPU connections and direct switch-level Peer-to-Peer (P2P) DMA routing across two external enclosures. 

  • PAE bundles use native PCI-SIG® CopprLink™ (CDFP / SFF-TA-1032) cabling to transport pure, unencapsulated PCIe Gen5 x16 signals at full 64 GB/s line rate. To eliminate signal jitter and high-frequency attenuation over the external cable, the RocketStor 8631D enclosure integrates an onboard Astera Labs PCIe 5.0 active retimer with SRIS clocking. This maintains native bus speeds with sub-microsecond transit latency and zero protocol-translation overhead.

  • Yes. The Switch adapters used for the RocketStor 8631DW and RocketStor 8631DH feature standard PCI-SIG Low-Profile brackets that drop directly into standard 1U and 2U server risers. The RocketStor 8631DD uses a Full-Height Half-Length (FHHL) single-width host adapter, which installs seamlessly into enterprise 1U servers equipped with standard horizontal (butterfly) riser cages as well as standard 2U+ risers. 

  • The RocketStor 8631D features an independent 1300W industrial Common Redundant Power Supply (CRPS) equipped with native 12VHPWR / 12V-2x6 high-current wiring harnesses. The enclosure independently powers the accelerator and absorbs transient peak current spikes up to 2x rated TDP, drawing 0W from the host server power plane to eliminate over-current protection (OCP) shutdowns.

  • Yes. Unlike internal server installations that force chassis fans to run at piercing speeds (80+ dBA), HighPoint PAE isolates heat externally. The RocketStor 8631D enclosure pairs an acoustically engineered industrial PSU with automated smart dual cooling fans, sustaining full GPU boost clocks under 100% matrix load while maintaining a controlled acoustic profile of 65.9 dBA.

PAE Series Solutions

Cabling & Accessories: Coming Soon

bottom of page