Dual Intel Xeon Scalable | Up to 20 GPUs | 24x DDR4 DIMMs | PCIe 4.0 Architecture
The H3C UniServer R5300 G3 is a groundbreaking 4U rackmount GPU server independently developed for AI and HPC. It is the industry's first AI server to adopt PCIe 4.0 link design, delivering double the bandwidth of PCIe 3.0. Ideal for deep learning training, inference, HPC, and big data analytics.
| Feature | Description |
|---|---|
| PCIe 4.0 Architecture | 16 GT/s per lane – ultra-low latency, high-speed NVMe and networking |
| 8 GPU Module | Up to 8 double-width GPUs for deep learning training |
| 20 GPU Module | Up to 20 single-width GPUs for high-throughput inference |
| Massive Storage | Up to 288TB capacity with 24x 3.5" drives |
| Category | Specification |
|---|---|
| Form Factor | 4U rackmount |
| Processors | 2x Intel Xeon Scalable or Montage Tsingma (up to 28 cores, 205W TDP) |
| Chipset | Intel C621 |
| Memory | 24x DDR4 DIMMs, up to 2933MT/s, 3.0TB max; supports Intel Optane DCPMM |
| GPU Support (8 GPU) | 8 double-width GPUs |
| GPU Support (20 GPU) | 20 single-width GPUs |
| Front Drive Bays | 24x 3.5" SAS/SATA/NVMe (max 8 NVMe) OR 12x 3.5" (max 4 NVMe) |
| Expansion (8 GPU) | 8x double-width x16 + 4x single-width x16 + OCP 3.0 + RAID slot |
| Expansion (20 GPU) | 20x single-width x16 + OCP 3.0 + RAID slot |
| Network | Optional OCP 3.0 network card |
| Cooling | 8x 8056 fans, N+1 hot-plug redundant |
| Power Supply | 4x 2000W/2400W Platinum, N+N redundancy |
| Management | HDM agent-free + H3C FIST |
| Dimensions | 174.8 x 447 x 807 mm |
| Operating Temp | 5°C to 40°C |
| Location | Ports |
|---|---|
| Front | 2x USB 2.0 (left), 1x USB 3.0 (right) |
| Rear | 2x USB 3.0, 1x VGA, 1x Serial, 1x HDM management |
Power: N+N redundant Platinum PSUs
Cooling: N+1 hot-plug redundant fans
RAID: 0/1/10/5/6/50/60 support
Management: Remote monitoring, proactive alerting
Microsoft Windows, Linux distributions, VMware, H3C CAS
The H3C UniServer R5300 G3 combines industry-first PCIe 4.0 technology, flexible GPU configurations (up to 20 GPUs), massive 288TB storage, and enterprise-grade reliability. Whether training deep learning models, performing real-time inference, or running HPC simulations, it delivers the performance and scalability required for AI-driven organizations.