Datacenter flagship · Hopper · launched 2024 · prices checked 2026-09-03
Rent NVIDIA H200 NVL — 141 GB, $2.567/hr on-demand
- VRAM 141 GBHBM3e
- FP16 tensor 835TFLOPS
- PowerScore 588RTX 3090 = 100
- Configs 1–8×NVLink
- Online now 12 2 regions
Billed per second, price locked at deploy. Storage $0.08/GB/mo · bandwidth $0.01/GB — the whole fee schedule. Next weekly market re-check: 2026-09-10.
Datacenter flagship · Hopper architecture
The H200 NVL is the PCIe form of the H200: 141 GB HBM3e in a standard server slot, bridgeable in pairs or quads with NVLink. It brings H200-class memory to hosts without an SXM baseboard, which is why it rents for less per hour than the SXM part.
This is training-grade silicon: HBM3e memory feeding tensor cores at multi-TB/s, NVLink for scaling past one card, and the reliability profile of Tier-III datacenter hosts. Teams rent it for pre-training, long fine-tunes and high-throughput inference where batch size is money.
NVIDIA H200 NVL specs: VRAM, TFLOPS, bandwidth
| GPU model | NVIDIA H200 NVL | Architecture | Hopper (2024) |
|---|---|---|---|
| VRAM | 141 GB HBM3e | Memory bandwidth | 4,800 GB/s |
| FP16 tensor perf. | 835 TFLOPS | FP32 perf. | 60.0 TFLOPS |
| CUDA cores | 16,896 | TDP | 600 W |
| PowerScore (RTX 3090 = 100) | 588 | PCIe generation | Gen 5.0 |
| Multi-GPU | 1× – 8× · NVLink | Max instance storage | 16,000 GB NVMe |
| Network up to | 1,000 Mbps | CUDA | 12.4 – 13.0 |
Bandwidth, CUDA cores, TDP and FP32 are public NVIDIA figures; FP16 tensor is the dense (non-sparsity) number. Machine-level values come from live inventory.
H200 NVL price per hour: on-demand, interruptible, reserved
One public rule sets every price on this page: the marketplace median for the H200 NVL ($3.67/hr, snapshot 2026-09-03) × 0.70, rounded down — so on-demand is $2.567, 30% below market. Interruptible halves it; a 3-month reservation takes another 35% off.
| Mode | Per GPU-hour | Per day (24 h) | Per month (730 h) | What you get |
|---|---|---|---|---|
| On-demand | $2.567 | $61.61 | $1,874 | Guaranteed capacity, price locked at deploy, stop anytime |
| Interruptible | $1.283 | $30.79 | $937 | Flat −50%; may pause under capacity pressure, disk kept, auto-requeue |
| Reserved (3 months) | $1.668 | $40.03 | $1,218 | −35% on on-demand, rate locked for the term, capacity held |
Per GPU: an 8× machine costs exactly 8× — no multi-GPU premium. Estimate a full month with storage and bandwidth in the GPU cost calculator.
What you can run on a H200 NVL (141 GB VRAM)
With 141 GB of HBM3e, a single card holds a ~49B-parameter LLM in FP16 or up to ~141B parameters quantized to 4-bit, with room for KV-cache at practical context lengths. Scale to 8× GPUs on one machine with NVLink for bigger models or bigger batches — the per-GPU price stays $2.567.
- One-click template: vLLM on a H200 NVL
- One-click template: PyTorch NGC on a H200 NVL
- One-click template: Axolotl — Fine Tuning on a H200 NVL
- Sizing help: LLM VRAM requirements guide
H200 NVL availability by region
12 × H200 NVL across 2 machines, live from inventory:
Ashburn, VA
Tokyo
H200 NVL vs alternatives: price per TFLOP
| GPU | VRAM | FP16 | On-demand | $ / TFLOP-hr |
|---|---|---|---|---|
| H200 NVL this card | 141 GB | 835 | $2.567 | $3.07‰ |
| H100 PCIE | 80 GB | 756 | $2.147 | $2.84‰ |
| H100 NVL | 80 GB | 835 | $1.877 | $2.25‰ |
| A100 PCIE | 80 GB | 312 | $0.662 | $2.12‰ |
| L40S | 48 GB | 362 | $0.466 | $1.29‰ |
‰ = dollars per 1,000 TFLOP-hours of FP16 — a rough value-for-compute yardstick across cards.
- H200 vs H200 NVL — specs, price per hour, which to rent
- All GPU comparisons
Renting a H200 NVL: frequently asked questions
How much does it cost to rent an NVIDIA H200 NVL per hour?
$2.567 per GPU-hour on-demand — a fixed price set at least 30% below the current market median of $3.67. Interruptible capacity costs $1.283/hr and a 3-month reservation $1.668/hr. Around $1,874/month if you keep one running non-stop, billed per second.
What can a H200 NVL with 141 GB VRAM run?
In LLM terms, roughly a 49B-parameter model in FP16 or up to ~141B parameters 4-bit quantized on a single card, with context headroom. Multi-GPU instances (up to 8× on current inventory) multiply that; diffusion and rendering workloads fit comfortably at this VRAM class.
Is the H200 NVL available to rent right now?
Yes — 12 GPUs across 2 machines in 2 regions are listed as we render this page. Configurations go from 1× to 8× with NVLink on multi-GPU chassis. Deploy from the console and it is running in about 30 seconds.
How do I deploy a H200 NVL?
Create an account (email + password, no card, no KYC), top up in crypto, open the console, filter by H200 NVL, pick a machine and a template such as PyTorch, vLLM or ComfyUI. The same deploy is one command with the CLI: powergpu launch --gpu h200-nvl --template pytorch.
Why is the H200 NVL cheaper here than on GPU marketplaces?
We price from the public marketplace median and fix our on-demand rate at least 30% below it, rounded down. The price is re-checked weekly (next check 2026-09-10) and published — no auctions, no per-host roulette, no bidding.