AI & GPU Accelerated Server for Deep Learning, Training & Inference
The HGR-AI-1 is our flagship AI-optimized server, purpose-built for deep learning training, large language model fine-tuning, and high-performance inference. Powered by Dual Intel Xeon Gold 6448Y processors and 4x NVIDIA A100 80GB GPUs connected via NVLink, this server delivers unmatched AI/ML performance with 320GB of total GPU memory and 512GB of system RAM.
Enterprise-grade components designed for demanding AI and deep learning workloads
2x Intel Xeon Gold 6448Y with 32 cores each, delivering 64 total cores and 128 threads. High single-thread and multi-thread performance for data preprocessing and CPU-bound AI tasks.
Four NVIDIA A100 Tensor Core GPUs with 80GB HBM2e each, totaling 320GB of GPU memory. NVLink interconnect enables fast GPU-to-GPU communication for distributed training.
Massive system memory for loading large datasets and model weights. DDR5 delivers higher bandwidth for CPU-GPU data transfers.
Ultra-high-speed networking for distributed training across multiple nodes and fast dataset loading from network storage.
| Model | 2x Intel Xeon Gold 6448Y |
| Cores / Threads | 64 Cores / 128 Threads (Total) |
| Base Frequency | 2.1 GHz |
| Max Turbo | 4.1 GHz |
| Cache | 60MB L3 Cache per CPU |
| TDP | 225W per CPU |
| Architecture | Intel 4th Gen Sapphire Rapids |
| PCIe Lanes | 80 PCIe 5.0 Lanes per CPU |
| Model | 4x NVIDIA A100 80GB |
| GPU Memory | 80GB HBM2e per GPU (320GB Total) |
| GPU Memory Bandwidth | 2TB/s per GPU |
| CUDA Cores | 6,912 per GPU (27,648 Total) |
| Tensor Cores | 432 per GPU (1,728 Total) |
| FP64 Performance | 9.7 TFLOPS per GPU |
| FP32 Performance | 19.5 TFLOPS per GPU |
| TF32 Tensor | 156 TFLOPS per GPU |
| FP16 Tensor | 312 TFLOPS per GPU |
| INT8 Tensor | 624 TOPS per GPU |
| NVLink | NVLink 3.0, 600 GB/s |
| TDP | 300W per GPU |
| Interface | PCIe Gen 4.0 x16 / SXM4 |
| Installed RAM | 512GB DDR5 ECC |
| Speed | DDR5-4800 MHz |
| Channels | 8 Channels |
| DIMM Slots | 16 |
| Max Memory | 2TB (16x 128GB) |
| ECC | Yes |
| Primary Storage | 4x 3.84TB NVMe SSD |
| Total Capacity | 15.36TB Raw |
| Interface | NVMe PCIe Gen 4.0 x4 |
| Read Speed | Up to 7,000 MB/s per drive |
| Write Speed | Up to 5,000 MB/s per drive |
| RAID Support | Software RAID 0, 1, 5, 10 |
| Hot Swap | Yes |
| Public Network | 1x 100GbE |
| Private Network | 1x 100GbE |
| Management | 1x 1GbE IPMI/BMC |
| Public Bandwidth | Unlimited @ 100Gbps |
| Private Bandwidth | Unlimited (vRack) |
| IPv4 | 1 Included (/32) |
| IPv6 | Included (/128) |
| Additional IPs | Up to 256 Failover IPs |
| InfiniBand | Optional HDR 200Gb/s |
| Anti-DDoS | Included (VAC Technology) |
| Remote Management | IPMI 2.0 / KVM over IP |
| Secure Boot | UEFI Secure Boot |
| TPM | TPM 2.0 |
| Hardware Firewall | Optional |
| Backup Storage | 500GB Included |
| Form Factor | 4U Rack Mount |
| Power Supply | 2x 3000W Titanium (Redundant) |
| Power Consumption | ~2,400W Typical |
| Cooling | Redundant Hot-Swap Fans |
| Dimensions | 175mm x 482mm x 812mm |
| Weight | ~55kg |
Customize your server to match your specific requirements
Server deployed within 120 seconds
The HGR-AI-1 excels in demanding AI, deep learning, and GPU-accelerated environments
Train large neural networks, transformers, and foundation models with multi-GPU parallelism.
Fine-tune large language models like LLaMA, Mistral, and custom models with 320GB GPU memory.
Train and deploy image classification, object detection, and segmentation models at scale.
Molecular dynamics, climate modeling, and physics simulations with GPU acceleration.
Build and deploy Stable Diffusion, GANs, and other generative models.
Serve multiple AI models simultaneously with high throughput and low latency.
Get started in minutes with our automated provisioning. 24/7 expert support included.