LLM Series Edge Servers
Short-Depth Rackmount Servers for On-Prem AI Inference
The LLM Series delivers short-depth rackmount edge computing for Private AI and Physical AI deployments. Built with scalable x86 compute, NVIDIA RTX™ PRO GPU acceleration, and flexible network connectivity, it enables dedicated on-prem AI infrastructure for local inference, multimodal AI, and intelligent edge applications without the scale and complexity of traditional data center servers.
Products:
Scalable x86 Processing
Choose Intel® Core™ or AMD Ryzen™ & EPYC™ platforms to match your AI workload.
Intel® Core™ Processors
LLM-1U-RPL- 14th / 13th / 12th Gen Intel® Core™
- Up to 24 cores / 32 threads
- Hybrid Performance and Efficient cores
- Intel® Thread Director workload optimization
AMD Ryzen™ & EPYC™ Processors
LLM-2U-AM5 · LLM-3U-AM5- AMD Ryzen™ 9000 Series
- AMD EPYC™ 4005 / 4004
- Up to 16 cores / 32 threads
- PCIe Gen 5 and DDR5 scalability
LLM Series Model Comparison
| Category | LLM-1U-RPL | LLM-2U-AM5 | LLM-3U-AM5 |
|---|---|---|---|
| Form Factor | 1U 19-inch Short-Depth Rackmount | 2U 19-inch Short-Depth Rackmount | 3U 19-inch Short-Depth Rackmount |
| Processor | 14th / 13th / 12th Gen Intel® Core™ i9/i7/i5/i3, up to 65W | AMD Ryzen™ 9000 & EPYC™ 4005/4004, up to 170W | AMD Ryzen™ 9000 & EPYC™ 4005/4004, up to 170W |
| Memory | Up to 64GB DDR4-3200 | Up to 192GB DDR5-5600 | Up to 192GB DDR5-5600 |
| PCIe Expansion | 1x PCIe Gen4 x16 or 2x Gen4 x8 | 2x PCIe Gen5 x16 1x PCIe x4 (x1, open-end) 1x PCIe x4 (x4, open-end, shared with M.2) |
2x PCIe Gen5 x16 1x PCIe x4 (x1, open-end) 1x PCIe x4 (x4, open-end, shared with M.2) |
| GPU Positioning | Compact to professional GPU acceleration | Single GPU, balanced AI performance | Single GPU with greater physical expansion |
| Networking | 3x 2.5GbE | 2x 10GbE SFP+, 2x GbE | 2x 10GbE SFP+, 2x GbE |
| NVMe Storage | 1x M.2 M-Key | 2x M.2 M-Key | 2x M.2 M-Key |
| SATA Storage | 2x 2.5" hot-swap SATA, RAID 0/1 | 2x 2.5" internal SATA, RAID 0/1 | 1x 2.5" + 1x 3.5" internal SATA, RAID 0/1 |
| OOB Management | — | 1x RJ45 EBIO-OOB-I (Optional) | 1x RJ45 EBIO-OOB-I (Optional) |
| Power Supply | 2x 600W, 1+1 redundant | 1x 850W | 1x 850W |
| Industrial Fan | 3x hot-swap dual smart fans + CPU fan | 4x system fans + CPU fan | 3x system fans + CPU fan |
| Operating Temperature | 0°C to 35°C* | 0°C to 35°C* | 0°C to 35°C* |
| Best Fit | Compact Edge AI inference | Balanced private AI, LLM and RAG | Higher-performance AI and expansion-heavy deployments |
From Cloud to Edge: A New Tier of AI Infrastructure
As AI moves closer to the edge, organizations do not always need high-density enterprise server infrastructure. The LLM Series fills the gap between GPU workstations and enterprise servers with short-depth, single-node, GPU-accelerated rackmount systems built for Private AI, RAG, and on-prem inference.
Benefits of LLM Series Rackmount Edge Servers
Improve TCO
Deploy dedicated AI compute without the cost and complexity of oversized enterprise server infrastructure.
Optimize Space-Constrained Footprints
Short-depth rackmount design brings GPU-accelerated AI into space-constrained environments
Bring AI Closer to Your Data
Run private AI and inference locally for greater data control, lower dependency on cloud connectivity, and faster access to operational data.
Case Study
Powering Next-Gen Robotic Picking: Real-Time Tactile & Vision AI at the Industrial Edge
A leading warehouse robotics provider deployed Premio’s LLM Series Edge AI Server to run real-time computer vision, tactile force feedback, and grasp planning directly at the robotic workcell—eliminating cloud latency to reliably pick complex, overlapping items.
Short-Depth 19-Inch Rackmount Design
Compact 1U, 2U, and 3U rackmount designs bring dedicated AI compute into space-constrained edge and on-prem environments, with short-depth options that reduce installation space versus conventional full-depth servers.
GPU Acceleration for Gen AI Workloads
Supports a scalable range of NVIDIA RTX PRO™ industrial-grade GPUs, from compact SFF cards to full-height, full-length GPUs for varying AI performance needs.
Explore Industrial GPU Computers
Supported GPUs for LLM Series
Validated NVIDIA professional GPUs
- Supports LLM-1U-RPL
- Supports LLM-2U-AM5
- Supports LLM-3U-AM5
Flexible Local Storage
NVMe and SATA storage options with RAID 0/1 support provide flexible capacity and data redundancy for model files, vector databases, application data, and local AI processing close to the edge.
Industrial Network Connectivity
The LLM Series supports RJ45 LAN connectivity across 1U, 2U, and 3U configurations, with dual 10GbE SFP+ available on select models for scalable on-prem AI deployments.
High-Speed Memory
Massive memory capacity and bandwidth to store heavy model weights and manage dynamic KV caches for long contexts. Because modern generative AI workloads are heavily memory-bound rather than compute-bound, memory performance directly dictates system latency, concurrency, and real-time generation speeds.
World-Class Safety Certifications
LLM Series servers are tested to UL, CE, and FCC standards for electrical safety and electromagnetic compliance, simplifying regulatory approvals and accelerating deployment worldwide. Combined with hardware-level security and industrial-grade components, they are built for long-term reliability in on-prem AI environments.
Remote Management for 24/7 Server Operations
Premio Remote Management solutions provide centralized access and hardware-level control for LLM-2U-AM5 and LLM-3U-AM5 systems, helping maintain server availability, accelerate issue recovery, and reduce onsite service.