GLabs GPU Servers
Shared Infrastructure for Organizations With an Existing Data Center
Rack-mount GPU servers for teams that already operate a data center and need shared, multi-tenant GPU capacity — rather than a pre-integrated standalone cluster.
Overview
| Spec | Detail |
|---|---|
| 4-GPU configuration | G-Series GL-G2U-4G (AMD EPYC 9004, PCIe Gen5) |
| 8-GPU configuration | G-Series GL-G4U-8G (Intel Xeon 6, 5th Gen Intel Xeon Scalable, or AMD EPYC 9004) |
| 10-GPU configuration | G-Series GL-G4U-10G (Intel Xeon 6, 5th Gen Intel Xeon Scalable, or AMD EPYC 9004 / 9005) |
| Throughput | 14,000+ TFLOPS aggregate (top configuration) |
| GPU options | NVIDIA H200 NVL, H100 NVL, L40S, L40, RTX PRO 6000 Server Edition |
| Software | GLIDS base image on every deployment, plus the GLabs AI Stack for AI workloads, GenesysBench™ validated |
| Deployment | GLIDS-validated — rack-and-stack into an existing facility |
All 70 Systems
Every model in the range has its own page: host platforms, previous generations, specifications and photos. Open a series to filter by height and compare up to four systems side by side.
By Height, 1U to 8U
Six series, from a single-socket 1U management node to an eight-GPU HGX system. Configured to the workload, validated with GenesysBench™ and deployed through GLIDS, with the GLabs AI Stack added for AI workloads.
| Height | Series and Models |
|---|---|
| 1U | C-Series (GL-C1U-4H-R, GL-C1U-4H, GL-C1U-2S-FIO, GL-C1U-2F) · S-Series (all-flash, hybrid and capacity) · N-Series (GL-N1U-2NG) |
| 2U | G-Series (GL-G2U-4G) · S-Series (all-flash and hybrid) · N-Series (GL-N2U-4N-FIO) |
| 4U | G-Series (GL-G4U-8G, GL-G4U-10G) · X-Series (GL-X4U-B200-L) · M-Series (GL-M4U-8G, GL-M4U-8G-CX8) · S-Series (GL-S4U-36H6F, GL-S4U-60H8F) · N-Series (GL-N4U-18N) |
| 6U | X-Series (GL-X6U-H200) · M-Series (GL-M6U-CX8-L) |
| 8U | X-Series (GL-X8U-B300, GL-X8U-B200) |
| Part | Meaning |
|---|---|
| Series | C General Purpose · G Scale-Out GPU · X Scale-Up GPU · M NVIDIA MGX · S Storage · N Multi-Node |
| Height | Rack units, 1U to 8U |
| Capacity | G GPUs · N nodes · H 3.5" HDD bays · S 2.5" SATA/SAS bays · F NVMe bays · or the HGX platform (B300, B200, H200) |
| Option | L liquid-cooled · CX8 NVIDIA ConnectX-8 SuperNIC · FIO front I/O · R redundant power |
So GL-X8U-B300 is an X-Series 8U system on NVIDIA HGX B300, and GL-S4U-60H8F is an S-Series 4U storage server with 60 HDD and 8 NVMe bays. The host CPU, Intel Xeon or AMD EPYC, is chosen per configuration.
C-Series: General Purpose
Single-socket 1U Intel servers for the machines that surround your GPUs: management, services and everyday workloads.
| Model | Height | Drive Bays | Network |
|---|---|---|---|
GL-C1U-4H-R | 1U | 4× 3.5" SATA | 2× 1GbE + 2× 10GbE, redundant PSU |
GL-C1U-4H | 1U | 4× 3.5" SATA | 2× 1GbE + 2× 10GbE, or per configuration |
GL-C1U-2S-FIO | 1U | 2× 2.5" SATA | Per configuration, front I/O |
GL-C1U-2F | 1U | 2× NVMe | Per configuration |
G-Series: Scale-Out GPU
PCIe GPU servers with 1 to 10 cards. Choose the GPU, then the host: Intel Xeon 6, 5th Gen Intel Xeon Scalable, or AMD EPYC 9005 / 9004.
| GPU | Model | Host CPU | Best For |
|---|---|---|---|
| RTX PRO 6000 Blackwell Server Edition | GL-G4U-8G | Intel Xeon 6 | Inference, vision, visual computing |
| H200 NVL | GL-G4U-10G | Xeon Scalable · Xeon 6 · EPYC 9005 | Inference and fine-tuning of larger models |
| H100 NVL | GL-G4U-10G · GL-G4U-8G | Xeon Scalable · EPYC 9004 · Xeon 6 | Inference and fine-tuning |
| L40S | GL-G2U-4G · GL-G4U-10G | EPYC 9004 · Xeon Scalable | Video AI, rendering, 3D |
| L40 | GL-G2U-4G · GL-G4U-10G | EPYC 9004 · Xeon Scalable | Rendering, visualization |
X-Series: Scale-Up GPU
Eight GPUs on one NVLink domain: NVIDIA HGX platforms for training and very large model inference.
| Platform | GPUs | Memory | Model |
|---|---|---|---|
| HGX B300 | 8× Blackwell Ultra SXM | Up to 2.1 TB | GL-X8U-B300 |
| HGX B200 | 8× Blackwell SXM | 1.4 TB | GL-X8U-B200 · GL-X4U-B200-L (liquid-cooled) |
| HGX H200 | 8× H200 SXM | 1.1 TB | GL-X6U-H200 |
Host CPU per configuration: Intel Xeon 6 (including Xeon 6900P on the B300), 5th Gen Intel Xeon Scalable, or AMD EPYC 9005.
Blackwell HGX platforms carry fifth-generation NVLink: 1.8 TB/s GPU-to-GPU and 14.4 TB/s total per baseboard.
M-Series: NVIDIA MGX
NVIDIA's modular GPU server design. MGX platforms follow NVIDIA's reference server architecture, on Intel Xeon 6 or AMD EPYC 9005 hosts.
| Model | GPUs Supported | Host CPU |
|---|---|---|
GL-M4U-8G | H200 NVL · H100 NVL on Intel Xeon 6; H200 NVL · L40S on AMD EPYC 9005 | Intel Xeon 6 or AMD EPYC 9005 |
GL-M4U-8G-CX8 | RTX PRO 6000 Blackwell Server Edition on either host; H200 NVL on Intel Xeon 6 | Intel Xeon 6 or AMD EPYC 9005 |
GL-M6U-CX8-L | Per configuration (liquid-cooled) | Intel Xeon 6 |
| On request | NVIDIA GH200 Grace Hopper | NVIDIA Grace |
S-Series: Storage
Storage that keeps the GPUs fed, in three tiers: all-flash for hot data, hybrid, and capacity for cold data.
| Tier | Role | Models |
|---|---|---|
| All-flash | Hot tier | GL-S1U-12F, GL-S1U-24F, GL-S1U-8F, GL-S2U-24F, GL-S2U-16F |
| Hybrid | Flash and disk | GL-S1U-4H4F, GL-S1U-8S4F, GL-S1U-12H4F, GL-S2U-12H4F, GL-S2U-12H8F |
| Capacity | Cold and bulk | GL-S1U-4H, GL-S4U-36H6F, GL-S4U-60H8F |
The model code spells out the bays: GL-S1U-12F = 12 NVMe. GL-S4U-36H6F = 36 HDD + 6 NVMe. GL-S4U-60H8F = 60 HDD + 8 NVMe. Single- or dual-socket hosts on Intel Xeon 6 or AMD EPYC 9004, per model.
N-Series: Multi-Node
More nodes per rack unit: several independent nodes in one chassis, for dense clusters and CPU-bound workloads.
| Model | Nodes | Density | Platform |
|---|---|---|---|
GL-N2U-4N-FIO | 4 | 2 nodes/U | AMD EPYC 9004, front I/O |
GL-N1U-2NG | 2, one GPU each | 2 nodes/U | Intel C252 or AMD Ryzen (AM5) |
GL-N4U-18N | 18 | 4.5 nodes/U | Intel C266 or C252 |
GPU and platform names are trademarks of their respective owners.
See also: GLabs HPC · GLabs Vault · GLabs POD · GLabs AI Factory
Where It Shows Up
Where It Shows Up
- Self-Hosted LLM InferenceShared GPU capacity for multi-tenant model serving instead of per-token cloud APIs.
- Computer Vision & Video AI at ScaleGPU servers behind an edge fleet for aggregated inference and retraining.
- 3D/VFX Render Farms & Game Dev Build FarmsScalable GPU capacity behind individual AI Workstations.
Used across: AI Solutions (GLabs AI Stack), Pro Creators, Engineering & Design
Tell us what you're building.
A solutions architect responds directly — no ticket queue.