B200 · H200 · H100
Large-model training and fine-tuning, multi-GPU distributed jobs and high-throughput inference for production LLMs.
Rent NVIDIA B200, H200, H100, A100, L40S and L4 GPUs for AI training, fine-tuning and inference — no hardware to buy, with simple per-GPU-hour pricing on a 1-year commitment.
Dedicated 8-GPU servers on a 1-year commitment, priced per GPU per hour — from L4 inference nodes to B200 training clusters.
Availability: GPU capacity is provisioned on order: we do not currently own GPU hardware, and GPUs are delivered within 30–45 days of LOI / contract signing.
| GPU | vRAM | vCPUs | RAM | Per GPU / hour | |
|---|---|---|---|---|---|
| NVIDIA B200Flagship | 192 GB | 32 | 400 GB | ₹600$6.30/hr | Request → |
| NVIDIA H200Popular | 141 GB | 30 | 375 GB | ₹345$3.60/hr | Request → |
| NVIDIA H100Popular | 80 GB | 26 | 250 GB | ₹230$2.43/hr | Request → |
| NVIDIA RTX PRO 6000 | 96 GB | 32 | 170 GB | ₹165$1.71/hr | Request → |
| NVIDIA A100 80GB | 80 GB | 16 | 115 GB | ₹171$1.89/hr | Request → |
| NVIDIA A100 40GB | 40 GB | 16 | 115 GB | ₹162$1.79/hr | Request → |
| NVIDIA A40 | 48 GB | 16 | 100 GB | ₹87$1.30/hr | Request → |
| NVIDIA L40S | 48 GB | 60 | 220 GB | ₹92$1.08/hr | Request → |
| NVIDIA A30 | 24 GB | 16 | 90 GB | ₹81$1.22/hr | Request → |
| NVIDIA L4Best value | 24 GB | 25 | 110 GB | ₹45$0.52/hr | Request → |
Prices are per GPU, per hour, on a 1-year (annual) commitment, exclusive of applicable taxes (GST). GPUs are supplied as dedicated servers of 8 GPUs each; vCPU and RAM figures are per GPU. Annual cost per server = rate × 8 GPUs × 8,760 hours — e.g. an 8× H100 server at ₹230 per GPU-hour is ₹1,61,18,400 per year. Delivery within 30–45 days of LOI / contract signing. Short-term and on-demand usage, multi-GPU nodes, volume and enterprise pricing on request.
A quick guide to matching workloads with GPUs on our rate card.
Large-model training and fine-tuning, multi-GPU distributed jobs and high-throughput inference for production LLMs.
Deep learning research, computer vision, NLP and mid-size model training for labs and data science teams.
Cost-efficient inference, video analytics, rendering and visualisation — the L4 starts at ₹45 per hour on a 1-year commitment.
Many teams start on cloud GPUs and move steady workloads on-premise — we deliver both.
| GPU cloud (annual plan) | On-premise AI lab | |
|---|---|---|
| Upfront cost | None — per-GPU-hour pricing on an annual plan | Capital purchase of servers, networking and storage |
| Time to start | Fast — no procurement cycle | Procurement, installation and commissioning |
| Best for | Pilots, training bursts, variable workloads | Steady, high-utilisation workloads and data sovereignty |
| Scaling | Add GPUs as needed | Plan capacity in advance |
| Support | Bench Mark team | Bench Mark commissioning, training and 3-year AMC |
Bench Mark's rate for the NVIDIA H100 (80 GB) is ₹230 per GPU per hour on a 1-year commitment, exclusive of GST. GPUs are supplied as dedicated 8-GPU servers, so an 8× H100 server costs ₹1,840 per hour, or ₹1,61,18,400 per year. Short-term and volume pricing is available on request. GPUs are provisioned on order and delivered within 30–45 days of LOI / contract signing.
The NVIDIA L4 (24 GB) starts at ₹45 per GPU-hour on a 1-year commitment, suitable for inference, video and lighter AI workloads.
NVIDIA B200, H200, H100, RTX PRO 6000, A100 (80 GB and 40 GB), A40, L40S, A30 and L4 — priced per GPU-hour on a 1-year commitment.
No. Listed prices are per GPU, per hour, on a 1-year commitment, and exclusive of applicable taxes (GST). GPUs are supplied as dedicated 8-GPU servers, so a server's annual cost is the rate × 8 GPUs × 8,760 hours.
Yes. Multi-GPU nodes, dedicated clusters, short-term or on-demand usage and enterprise pricing are quoted by our team — contact us with your workload.