RDP GX4 4-GPU Server Max
Liquid-cooled 4-GPU training & inference node — trains & serves up to 70B, NVIDIA-Certified and Make-in-India.
Overview
RDP GX4 4-GPU Server Max is a production, liquid-cooled GPU server for model training and high-throughput inference, built for CIOs, CTOs, heads of infrastructure and neocloud operators who need datacenter-grade performance with Make-in-India economics.
Four NVLink-connected H100 GPUs and 320 GB HBM3 make it a strong single-node home for 70B training, fine-tuning and low-latency inference — with a clear path from pilot to a multi-node cluster.
Key specifications
| GPUs | 4× NVIDIA H100 80GB SXM5 |
| GPU memory | 320 GB HBM3 (4× 80 GB) |
| GPU interconnect | NVLink (900 GB/s) |
| CPU | 2× AMD EPYC 9554 (128 cores) |
| System memory | 1.5 TB DDR5 ECC (up to 3 TB) |
| Storage | 2× 1.92 TB NVMe + 4× 3.84 TB NVMe |
| Networking | 2× 400G (ConnectX-7) · RoCEv2 / IB |
| Cooling | Direct liquid cooling (DLC), CDU-ready |
| Form factor | 4U rackmount |
| Power | 4× 3000W redundant (2+2) |
| Management | BMC · Redfish · IPMI 2.0 |
| OS / stack | Ubuntu 22.04, RHEL 9 · NVIDIA AI Enterprise |
| Warranty | 3-year pan-India onsite |
Spec values are representative for this preview and bind to the corrected live product data at wiring.
Workload fit
70B training & fine-tuning, 13B–34B at speed, low-latency 70B inference, and large-context RAG. Scales to multi-node with the 400G fabric and a head node.
Why RDP GPU Mart
- Make in India OEM — 28,000 sq ft Hyderabad facility, 14 years, 300,000+ devices shipped.
- Sovereign-ready: India data residency (DPDP), MeitY-recognised, ISO 27001 / SOC 2 paths.
- INR-transparent: GST invoice, HSN 8471, CGST/SGST or IGST, pan-India onsite SLA.
- Available on GeM for government and PSU procurement.
FAQ
How is it billed and supported?
INR with GST (HSN 8471), CGST/SGST or IGST by state, EMI/lease/GPUaaS options, and 3-year pan-India onsite support. Available on GeM.
Can I start as a pilot and scale to a cluster?
Yes — the fabric and management stack let you add nodes and a head node later. Talk to an architect for the reference design.
What software stack is included?
Ubuntu 22.04 / RHEL 9 with NVIDIA AI Enterprise, drivers and CUDA validated; Slurm/Kubernetes on request.
Other GPU Servers in this line
| GX4 4-GPU | GX4 Plus | GX4 Max | GX8 HGX | |
|---|---|---|---|---|
| GPUs | 4× L40S | 4× H100 PCIe | 4× H100 SXM | 8× H100 SXM |
| Model fit | up to 34B | 34B–70B | 70B | 70B–405B |
| Cooling | Air | Air | Air / DLC | Direct liquid |
| Price | Price on request | Price on request | Request a Quote | On request |
| View | View | View | Quote |
Pair it with
Designing a GPU cluster, not just one server?
Talk to our solution architects — multi-node fabric design, financing & GPU-as-a-Service, India-onsite SLA.