B300
NVIDIA · Blackwell Ultra
288 GBHBM3e
8 TB/sBandwidth
4.5 PFFP8 dense
$4,899 /GPU/mo$6,552 · −25%
NVIDIA · Hopper · SXM · 8-GPU HGX
Rent the NVIDIA H200 SXM by the month.
The NVIDIA H200 is Hopper with 141 GB of HBM3e at 4.8 TB/s: 76% more memory and 43% more bandwidth than an H100, on the most mature CUDA software stack available. It is our best-value GPU for fine-tuning and FP8 inference.
141 GBHBM3e per GPU
4.8 TB/sMemory bandwidth
1.98 PFFP8 dense
NVLink 4Interconnect
Why the H200
What the H200 is best at
- 141 GB HBM3e. A 70B model at FP8 on one GPU with room for the KV cache, or in BF16 on two.
- 4.8 TB/s bandwidth. 1.4× an H100 for memory-bound inference.
- Same software as H100. Every CUDA kernel, container and recipe written for Hopper runs unchanged.
- LLM fine-tuning (LoRA, QLoRA, full)
- FP8 inference with vLLM or TensorRT-LLM
- RAG and embedding pipelines
An 8× H200 server gives 1.1 TB of HBM3e on NVLink 4 (900 GB/s per GPU).
Pricing
H200 monthly price by server size
One flat price per GPU, the same from 1 to 8 GPUs. The market median is the public on-demand list price of specialist GPU clouds × 730 hours.
| Server | CoreGPU / month | Market median / month | You save / month | You save / year | ≈ per GPU-hour |
|---|---|---|---|---|---|
| 1× H200 | $2,699 | $907 | $10,884 | $3.70 | |
| 2× H200 | $5,398 | $1,814 | $21,768 | $3.70 | |
| 4× H200 | $10,796 | $3,628 | $43,536 | $3.70 | |
| 8× H200 | $21,592 | $7,256 | $87,072 | $3.70 |
Prices in USD, excluding taxes. Market median checked Nov 4, 2026 across specialist GPU clouds; hyperscalers excluded. All GPU prices
Compare
H200 vs H100
Same compute as H100 but 141 vs 80 GB of memory. If your model or batch does not fit in 80 GB, H200 avoids splitting it across more GPUs.
| H200 | H100 | |
|---|---|---|
| Architecture | Hopper | Hopper |
| GPU memory | 141 GB HBM3e | 80 GB HBM3 |
| Memory bandwidth | 4.8 TB/s | 3.35 TB/s |
| FP8 dense | 1.98 PF | 1.98 PF |
| FP4 dense | n/a | n/a |
| Interconnect | NVLink 4 · 900 GB/s | NVLink 4 · 900 GB/s |
| Power (TDP) | 700 W | 700 W |
| Price per GPU / month | $2,699 | $1,999 |
| Price per GB of memory | $19.14 | $24.99 |
| Price per FP8 PFLOPS | $1,363 | $1,010 |
| Availability | 34 available | Only 3 left |
Specifications
NVIDIA H200 SXM technical specs
Per-GPU figures from the vendor datasheet. Tensor throughput is dense (without sparsity).
| Model | NVIDIA H200 SXM |
|---|---|
| Architecture | Hopper |
| Form factor | SXM · 8-GPU HGX |
| GPU memory | 141 GB HBM3e |
| Memory bandwidth | 4.8 TB/s |
| FP8 Tensor (dense) | 1.98 PF |
| FP4 Tensor (dense) | Not supported |
| Interconnect | NVLink 4 · 900 GB/s |
| Power (TDP) | 700 W |
| GPUs per server | 1 to 8 |
| Availability | 34 available |
FAQ
H200 rental: common questions
Another question? Our engineers answer within one business day.
Contact usHow much does it cost to rent an NVIDIA H200 SXM per month?
$2,699 per GPU per month at CoreGPU, about $3.70 per GPU-hour, versus a market median of $3,606 ($4.94/hour × 730 h). A full 8-GPU server is $21,592 per month, taxes excluded.
Is the H200 available right now?
Yes: 34 available in the current allocation of 128 GPUs. In-stock servers are delivered with root SSH access in under 60 seconds after payment.
How many H200 GPUs can I rent?
From 1 to 8 GPUs per server, at the same price per GPU. Servers with 8 GPUs come with the full NVLink 4 mesh. Need more? Rent several servers or ask for a private allocation.
H200 or H100?
Same compute as H100 but 141 vs 80 GB of memory. If your model or batch does not fit in 80 GB, H200 avoids splitting it across more GPUs. See the full H200 vs H100 comparison.
What do I get with a H200 server?
Dedicated GPUs (never shared or sliced), root SSH access, your choice of a clean OS or a ready image with CUDA, PyTorch and vLLM, local NVMe scratch, and month-to-month billing with no setup fee.
Alternatives
Other GPUs in stock now
Deploy your H200 server in under 60 seconds
Pick 1 to 8 GPUs in the configurator above. Month to month, cancel anytime.