Rent the H200 if…
you serve LLMs — 70B-class models, long contexts or large batches above all — or your training run is limited by memory and would otherwise be sharded across two H100s.
Deploy an H200 · $1,690/mo
NVIDIA H200141 GB HBM3e · Hopper
vs
NVIDIA H100 SXM80 GB HBM3 · Hopper
NVIDIA H200 vs H100: which GPU should you rent?
The same Hopper compute with more and faster memory: 141 GB of HBM3e at 4.8 TB/s on the H200 against 80 GB of HBM3 at 3.35 TB/s on the H100 SXM — for 1.4× the monthly price.
- H200, dedicated$1,690per month · $2.32/h effective
- H100, dedicated$1,249per month · $1.71/h effective
- Price ratio1.4×the H200 costs 1.4× the H100 per month
The short answer
Which one should you rent?
The H200 and the H100 SXM share the same Hopper GPU: identical tensor throughput, the same FP8 Transformer Engine, the same CUDA kernels. What changes is memory — 1.76× the capacity and 1.43× the bandwidth. Rent the H200 when your job is limited by memory: 70B-class inference on a single GPU, long contexts whose KV cache spills past 80 GB, large batches. Rent the H100 when the model fits comfortably in 80 GB and the work is compute-bound, as most training is: you get the same speed for $441 less a month.
- Compute. Identical: about 989 TFLOPS of dense FP16/BF16 and about 1,979 of FP8 on both. A compute-bound job runs at the same speed.
- Memory capacity. 141 GB against 80 GB: a 70B model at FP8 — about 70 GB of weights — fits on one H200 with room for its KV cache; on H100s it is more comfortable on two.
- Memory bandwidth. 4.8 TB/s against 3.35 TB/s, 43% more. Token generation is bandwidth-bound, and that is where the H200 pulls ahead.
- Price per GB. $11.99 per GB of GPU memory a month on the H200 against $15.61 on the H100: the H200 is the cheaper way to buy memory.
Rent the H100 if…
your models fit in 80 GB with room to spare and the job is compute-bound, as most pre-training and fine-tuning is: the same throughput at a lower monthly price.
Deploy an H100 · $1,249/moSpecifications
H200 vs H100: the numbers side by side
NVIDIA's published figures for each part, tensor throughput without sparsity. The last column is how the H200 compares, as a multiple of the H100.
| Specification | NVIDIA H200 | NVIDIA H100 SXM | H200 vs H100 |
|---|---|---|---|
| Architecture | Hopper · 2024 | Hopper · 2022 | — |
| GPU memory | 141 GB HBM3e | 80 GB HBM3 | 1.8× |
| Memory bandwidth | 4.8 TB/s | 3.35 TB/s | 1.4× |
| FP16 / BF16 tensor (dense) | ~989 TFLOPS | ~989 TFLOPS | Same |
| FP8 tensor (dense) | ~1,979 TFLOPS | ~1,979 TFLOPS | Same |
| FP4 tensor (dense) | Not supported | Not supported | — |
| Maximum power | Up to 700 W | Up to 700 W | Same |
| Form factor | SXM5 | SXM5 | — |
| GPU-to-GPU link | NVLink 4 · 900 GB/s | NVLink 4 · 900 GB/s | — |
| Multi-Instance GPU (MIG) | Up to 7 instances | Up to 7 instances | — |
| ECC memory | Yes | Yes | — |
| Video engines | 7 NVDEC · 7 NVJPG | 7 NVDEC · 7 NVJPG | — |
| CUDA compute capability | 9.0 | 9.0 | — |
| Dedicated host here | 24 vCPU · 256 GB RAM · 3 TB NVMe | 24 vCPU · 256 GB RAM · 3 TB NVMe | — |
| Price per month here | $1,690 | $1,249 | 1.4× |
Sources: NVIDIA datasheets and product pages for each GPU. Tensor figures are dense (NVIDIA often prints them doubled “with sparsity”). Prices are this site's flat monthly rates for a dedicated server with the host listed, in USD excluding VAT.
Value
What each dollar buys
Price per month divided into the two resources most jobs run out of first — memory and memory bandwidth — and what the same GPU costs kept on for a month at the market-median hourly rate.
Per GB of GPU memoryH200 $11.99 · H100 $15.61per month, dedicated host included
Per TB/s of bandwidthH200 $352 · H100 $373per month — what decode-heavy inference pays for
Effective hourly rateH200 $2.32/h · H100 $1.71/hflat monthly price over 730 hours
Same GPU on demand, 24/7H200 $3,270 · H100 $2,446market-median hourly rate × 730 h
Model sizes
What fits on one H200 and one H100
The largest model whose weights fit on a single GPU, keeping 10% of the memory for the runtime, activations and a small KV cache. Long contexts and large batches need more — size them with the GPU memory calculator.
| Weights precision | NVIDIA H200 | NVIDIA H100 SXM | Typical use |
|---|---|---|---|
| 16-bit (BF16 / FP16) | up to ~63B parameters | up to ~36B parameters | Full-precision inference; the base weights of a LoRA fine-tune |
| 8-bit (FP8 / INT8) | up to ~130B parameters | up to ~72B parameters | FP8 serving with vLLM or TensorRT-LLM on GPUs that support it; INT8 weights elsewhere |
| 4-bit (GPTQ, AWQ, Q4) | up to ~230B parameters | up to ~130B parameters | Quantised inference on a budget; QLoRA fine-tuning |
Parameters in billions, weights only: 2 bytes per parameter at 16-bit, 1 at FP8 or INT8, about 0.55 at 4-bit (GPTQ, AWQ, GGUF Q4). Mixture-of-experts models need memory for all their experts, not just the active ones.
Prices
H200 and H100 price per month, by term
Per dedicated GPU server, in USD excluding VAT — CPU cores, RAM, NVMe, 20 TB of transfer and 24/7 support included.
| Term | NVIDIA H200 | NVIDIA H100 SXM | Difference per month |
|---|---|---|---|
| Monthly | $1,690 | $1,249 | $441 |
| 3 months · −5% | $1,606 | $1,187 | $419 |
| 6 months · −10% | $1,521 | $1,124 | $397 |
| 12 months · −15% | $1,437 | $1,062 | $375 |
| Dedicated host | 24 vCPU · 256 GB RAM · 3 TB NVMe · 25 Gbps | 24 vCPU · 256 GB RAM · 3 TB NVMe · 25 Gbps | included |
H200 vs H100: questions answered
Is the H200 faster than the H100?
Not in raw compute — both have the same Hopper tensor cores. It is faster when memory is the bottleneck: its 4.8 TB/s speeds up token generation, and its 141 GB avoids splitting a model across two GPUs.
Can I run Llama 70B on one H200?
Yes, at FP8: the weights take about 70 GB, leaving roughly 55 GB for the KV cache and the runtime. At 16-bit the weights alone are about 140 GB, which needs two H200s.
Is H100 code compatible with the H200?
Yes. Same architecture, same CUDA compute capability (9.0), same FP8 Transformer Engine: containers and kernels built for the H100 run unchanged.
Are the H200 and the H100 in stock?
9 of 32 H200 units were free at the last stock update; 14 of 64 H100 units were free at the last stock update (18 September 2026). Both are provisioned automatically, in under 10 minutes after payment.
Can I rent the H200 or the H100 without KYC?
Yes. Both are rented with an email address and paid from a prepaid balance in Bitcoin, Ether or USDT — $1,690 and $1,249 a month, dedicated, with no identity check and no card.
Other GPU comparisons
Both are in stock, by the month
Dedicated, single-tenant and live in under 10 minutes: the H200 at $1,690 or the H100 at $1,249 a month, paid in BTC, ETH or USDT with no KYC.