RDP AI Infrastructure Solutions

Fixed reference configurations for AI-GPU compute, storage, fabric, and operations — quote-ready models for Private AI Factory deployments

AI-POD System SKUs

Complete 1-rack AI Factory solutions with fixed BOM mapping — GenAI Inference, Vision AI, and Training pods

Model No. AI-POD-7101
RDP GenAI Inference Pod (1 Rack)
Compute: 3× AI-GPU-72201 (4-GPU Inference Core | Intel)
Hot Storage: 1× AI-STG-6301B (Hot NVMe HA Storage | Core)
Fabric: 1× AI-FAB-6401 (100GbE Lossless Fabric HA Pair)
Ops: AI-OPS-6501 + 6502 recommended
Best for Private copilots, RAG, embeddings, enterprise inference services (CIOs, GCCs, BFSI, manufacturing HQ)
Model No. AI-POD-7102
RDP Vision AI Pod (1 Rack)
Compute: 4× AI-GPU-72101 (4-GPU Inference Entry | Intel)
Retention: 1× AI-STG-6302A (Capacity + Throughput Storage | Entry)
Optional Hot: + AI-STG-6301A (recommended for faster indexing)
Fabric: 1× AI-FAB-6401 (100GbE Lossless Fabric HA Pair)
Best for Video analytics, factory safety, quality inspection, surveillance intelligence (manufacturing plants, warehouses, campuses)
Model No. AI-POD-7103
RDP Training / Fine-tuning Pod (1 Rack)
Compute: 2× AI-GPU-75101 (8-GPU Training Entry | Intel)
Hot Storage: 1× AI-STG-6301B (Hot NVMe HA Storage | Core)
Dataset Lake: + AI-STG-6302A/6302B (recommended)
Fabric: 1× AI-FAB-6402 (200GbE Lossless Fabric HA Pair)
Best for Fine-tuning, training runs, batch AI pipelines, model iteration (ISVs, product engineering teams, research units)

AI-GPU (8 Models)

Fixed reference configurations for AI-GPU compute nodes — price-only decision for CIOs

A) 4-GPU Inference Entry (48GB VRAM class)
Model No. AI-GPU-72101
4-GPU Inference Node (Entry | Intel)
Chassis: 4U GPU server, front-to-back airflow
CPU: 2× Intel Xeon Gold 6430 (32C each)
Memory: 512GB DDR5 ECC (16×32GB, 4800 MT/s)
GPU: 4× NVIDIA L40S 48GB (PCIe Gen5 x16)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 2× 3.84TB NVMe
Network: 1× ConnectX-6 Dx dual-port 100GbE
Power: 2× 3000W Titanium redundant (~2.5–3.5 kW)
Best for RAG, copilots, embeddings, departmental inference, PoCs — works standalone or inside AI-POD-7101/7102
Model No. AI-GPU-72102
4-GPU Inference Node (Entry | AMD)
Chassis: 4U GPU server, front-to-back airflow
CPU: 2× AMD EPYC 9354 (32C each)
Memory: 512GB DDR5 ECC (16×32GB, 4800 MT/s)
GPU: 4× NVIDIA L40S 48GB (PCIe Gen5 x16)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 2× 3.84TB NVMe
Network: 1× ConnectX-6 Dx dual-port 100GbE
Power: 2× 3000W Titanium redundant (~2.5–3.5 kW)
Best for Same as 72101 with AMD EPYC platform — for AMD-only accounts or preference
B) 4-GPU Inference Core (80GB VRAM class)
Model No. AI-GPU-72201
4-GPU Inference Node (Core | Intel)
Chassis: 4U GPU server, redundant fans, redundant PSUs
CPU: 2× Intel Xeon Gold 6454S (32C each)
Memory: 768GB DDR5 ECC (12×64GB, 4800 MT/s)
GPU: 4× NVIDIA H100 80GB (PCIe Gen5 x16)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 4× 3.84TB NVMe
Network: 1× ConnectX-6 Dx dual-port 100GbE
Power: 2× 3000W Titanium redundant (~3.0–4.5 kW)
Best for Enterprise concurrency, heavier RAG, multi-tenant inference, higher VRAM workloads — default for AI-POD-7101
Model No. AI-GPU-72202
4-GPU Inference Node (Core | AMD)
Chassis: 4U GPU server, redundant fans, redundant PSUs
CPU: 2× AMD EPYC 9454 (48C each)
Memory: 768GB DDR5 ECC (12×64GB, 4800 MT/s)
GPU: 4× NVIDIA H100 80GB (PCIe Gen5 x16)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 4× 3.84TB NVMe
Network: 1× ConnectX-6 Dx dual-port 100GbE
Power: 2× 3000W Titanium redundant (~3.0–4.5 kW)
Best for Same as 72201 with AMD EPYC platform — higher core count option
C) 8-GPU Training Entry (80GB training class)
Model No. AI-GPU-75101
8-GPU Training Node (Entry | Intel)
Chassis: 8U HGX-class 8-GPU platform
CPU: 2× Intel Xeon Platinum 8480+ (56C each)
Memory: 1TB DDR5 ECC (16×64GB, 4800 MT/s)
GPU: NVIDIA HGX H100 (SXM) — 8× 80GB
Interconnect: NVLink/NVSwitch (HGX platform)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 4× 3.84TB NVMe
Network: 1× ConnectX-7 dual-port 200GbE
Power: 4× 3000W Titanium redundant (~8–12 kW)
Best for Fine-tuning, training runs, batch AI pipelines, ISVs/R&D teams starting training — starter node for AI-POD-7103
Model No. AI-GPU-75102
8-GPU Training Node (Entry | AMD)
Chassis: 8U HGX-class 8-GPU platform
CPU: 2× AMD EPYC 9654 (96C each)
Memory: 1TB DDR5 ECC (16×64GB, 4800 MT/s)
GPU: NVIDIA HGX H100 (SXM) — 8× 80GB
Interconnect: NVLink/NVSwitch (HGX platform)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 4× 3.84TB NVMe
Network: 1× ConnectX-7 dual-port 200GbE
Power: 4× 3000W Titanium redundant (~8–12 kW)
Best for Same as 75101 with AMD EPYC platform — higher core count for parallel workloads
D) 8-GPU Training Pro (high-memory training class)
Model No. AI-GPU-75301
8-GPU Training Node (Pro | Intel)
Chassis: 8U HGX-class platform
CPU: 2× Intel Xeon Platinum 8480+ (56C each)
Memory: 2TB DDR5 ECC (16×128GB, 4800 MT/s)
GPU: NVIDIA HGX H200 (SXM) — 8× 141GB
Interconnect: NVLink/NVSwitch (HGX platform)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 8× 3.84TB NVMe
Network: 1× ConnectX-7 dual-port 200GbE
Power: 4× 3000W Titanium redundant (~10–14 kW)
Best for Heavier fine-tuning, larger models, longer sustained training, faster checkpointing — Pro/core node for AI-POD-7103
Model No. AI-GPU-75302
8-GPU Training Node (Pro | AMD)
Chassis: 8U HGX-class platform
CPU: 2× AMD EPYC 9654 (96C each)
Memory: 2TB DDR5 ECC (16×128GB, 4800 MT/s)
GPU: NVIDIA HGX H200 (SXM) — 8× 141GB
Interconnect: NVLink/NVSwitch (HGX platform)
Storage: 2× 1.92TB SATA SSD (RAID-1) + 8× 3.84TB NVMe
Network: 1× ConnectX-7 dual-port 200GbE
Power: 4× 3000W Titanium redundant (~10–14 kW)
Best for Same as 75301 with AMD EPYC platform — maximum CPU cores for data-heavy training

AI-STG (4 Models)

Hot NVMe HA storage and capacity retention tiers for AI workloads

1) AI Hot Tier (Vector DB, Hot corp, Checkpoints, Active datasets)
Model No. AI-STG-6301A
Hot NVMe HA Storage (Entry | Dual-Controller | 2U)
Form Factor: 2U, 24× U.2 NVMe bays (front)
Controllers: Dual active-active (HA), hot-swappable
Media: 24× 3.84TB U.2 NVMe (enterprise)
Capacity: ~92TB raw / ~60–70TB usable
Connectivity: 8× 100GbE ports total (QSFP28)
Protocols: NFSv3/v4, SMB, iSCSI, NVMe/TCP
Features: Thin provisioning, snapshots, QoS, HA failover
Power: Dual redundant PSUs (~800W–1.6kW)
Best for RAG/vector DB, hot corp store, model cache, checkpointing for smaller training pods — default hot tier for AI-POD-7101
Model No. AI-STG-6301B
Hot NVMe HA Storage (Core | Dual-Controller | 2U)
Form Factor: 2U, 24× U.2 NVMe bays (front)
Controllers: Dual active-active (HA), hot-swappable
Media: 24× 7.68TB U.2 NVMe (enterprise)
Capacity: ~184TB raw / ~120–140TB usable
Connectivity: 8× 100GbE ports total (QSFP28)
Protocols: NFSv3/v4, SMB, iSCSI, NVMe/TCP
Features: Thin provisioning, snapshots, QoS, HA failover
Power: Dual redundant PSUs (~800W–1.6kW)
Best for Larger RAG corp, bigger vector stores, faster checkpoint bursts, multi-team AI pods — core hot tier for AI-POD-7101, default for AI-POD-7103
2) AI Capacity Tier (Dataset lake, Vision retention, Long retention + replay)
Model No. AI-STG-6302A
Capacity + Throughput Storage (Entry | 4U 60-Bay | HA Controllers)
Form Factor: 4U, 60× LFF bays (3.5")
Controllers: Dual controllers (HA), hot-swappable
Media: 60× 18TB NL-SAS HDD + 4× 3.84TB SSD cache
Capacity: ~1080TB raw / ~750–850TB usable
Connectivity: 8× 25GbE or 4× 100GbE (select at order)
Protocols: NFSv3/v4, SMB (optional), S3 (optional)
Features: Snapshots, quotas, tiering, replication-ready
Power: Dual redundant PSUs (~1.0–2.5kW)
Best for Vision/video retention, AI dataset lake, long retention + replay — default retention tier for AI-POD-7102, optional dataset lake for AI-POD-7103
Model No. AI-STG-6302B
Capacity + Throughput Storage (Core | 4U + Expansion | HA)
System: Base (4U 60-bay) + 1× expansion (4U 60-bay)
Controllers: Dual controllers (HA), hot-swappable
Media: 120× 18TB NL-SAS HDD + 4× 3.84TB SSD cache
Capacity: ~2160TB raw / ~1.5–1.7PB usable
Connectivity: 8× 25GbE or 4× 100GbE (select at order)
Protocols: NFSv3/v4, SMB (optional), S3 (optional)
Features: Same as 6302A, higher capacity + better parallelism
Power: Dual redundant PSUs (~1.0–2.5kW)
Best for Bigger multi-site retention, larger dataset lake, "grow without redesign" — core retention tier for larger AI-POD-7102, recommended dataset lake for AI-POD-7103

AI-FAB (Fabric Options)

Lossless fabric infrastructure for AI workloads — pod-scale and training-scale options

Model No. AI-FAB-6401
100GbE Lossless Fabric (HA Pair | Pod Scale)
Design: 2× Top-of-Rack switches (HA pair), MLAG/vPC
Ports: 32×100GbE QSFP28 per switch
Features: Lossless Ethernet (PFC/ECN), LACP/MLAG, QoS templates
Telemetry: Port health, link flaps, error counters
Cabling: DAC/optics for up to 4× AI-GPU nodes + 1× AI-STG hot storage
Management: 1× mgmt port per switch, RBAC, audit logs
Best for Stable throughput between GPU nodes + hot storage for inference & vision pods — default fabric for AI-POD-7101 and AI-POD-7102
Model No. AI-FAB-6402
200GbE Lossless Fabric (HA Pair | Training Scale)
Design: 2× ToR switches (HA pair), MLAG/vPC
Ports: 32×200GbE QSFP112 per switch
Features: Lossless tuning (PFC/ECN), QoS for storage/checkpoint traffic
Telemetry: Fabric health baselines
Cabling: For up to 2× 8-GPU training nodes + 1× hot NVMe storage
Management: 1× mgmt port per switch, RBAC, audit logs
Best for Training pods need higher east–west throughput and checkpoint bursts — default fabric for AI-POD-7103
Model No. AI-FAB-6403
400GbE Spine Option (Multi-rack Scale | Optional)
Design: 2× spine switches (400GbE)
Integration: Integrates with 6401/6402 as leaf layer
Use Case: When customer scales to 2+ racks and wants leaf-spine fabric
Best for Multi-rack scale (3–10 racks), no redesign later — optional for future expansion

AI-OPS (Ops/Acceptance Options)

Deployment readiness and lifecycle management for AI infrastructure

Model No. AI-OPS-6501
AI Factory Acceptance Pack (Mandatory)
Includes: Rack installation validation: power, PDU, thermal mapping
GPU Burn-in: Multi-hour GPU stress + PCIe checks + error logs
Storage: Throughput + latency baseline + failover test
Fabric: Lossless profile applied, link tests, error counters, redundancy failover
Inventory: Serials, MACs, firmware versions captured
Outputs: Acceptance report + baseline performance sheet + handover pack
Best for "Go-live confidence" with validation and sign-off — applies to every AI-POD, recommended for standalone AI-GPU
Model No. AI-OPS-6502
Day-2 Ops Runbook Pack (Recommended)
Monitoring: GPU health, temps, fan, power, ECC errors, NVMe wear, fabric errors
Alerts: Thresholds + escalation matrix
Baselines: Driver/firmware baseline (known-good versions)
Patch Cadence: Quarterly recommended + rollback method
Spares: PSUs, fans, NVMe, optics recommendation
RMA: Workflow + response SLAs (as per warranty tier)
Outputs: Runbook PDF + monitoring checklist + escalation contacts
Best for Ensure customer can run AI factory without chaos — recommended for all AI-PODs
Model No. AI-OPS-6503
Managed Monitoring + Patch Cadence (Optional)
Includes: Monthly health dashboard (GPU, storage, fabric)
Patch Window: Planning + firmware/driver upgrade support
Incident: Triage support (within SLA)
Best for RDP + SI can deliver "light managed services" around the pod — CIOs who want single accountability without full managed hosting
Model No. AI-OPS-6504
Annual Health Check + Firmware Baseline (Optional)
Includes: Full diagnostics + firmware baseline refresh
Assessment: Performance drift assessment
Checks: Thermal/power check + recommendations
Best for Yearly preventive maintenance + trust — optional for customers who want ongoing assurance