1× H200141 GB VRAM
Location not stated · 15 vCPU · 267 GB RAM · 1024GB NVMe
Details
Stock and price checked 15 min ago.
You start it yourself once there is stock.
H200-S
Full price
€3.00
per hour, whole instance, 1 GPU
NVIDIA · Hopper
Rent H200
NVIDIA H200 Tensor Core GPU, 141 GB.LLM trainingLLM inferenceHPC
Full price from
—
in stock, per GPU-hour
Spot from
—
in stock, per GPU-hour
Offers in stock
0 of 6
full price or spot
Providers in stock
0 of 3
full price or spot
Offers (8)
View offer
2× H200141 GB each, 282 GB total
Location not stated · 30 vCPU · 534 GB RAM · 1536GB NVMe
Details
Stock and price checked 15 min ago.
You start it yourself once there is stock.
This GPU supports NVLink, but the provider doesn't say whether this server has it connected. Ask before relying on it.
H200-M
Full price
€6.00
per hour, whole instance, 2 GPUs
€3.00/h per GPU
4× H200141 GB each, 564 GB total
Location not stated · 60 vCPU · 1068 GB RAM · 2048GB NVMe
Details
Stock and price checked 15 min ago.
You start it yourself once there is stock.
This GPU supports NVLink, but the provider doesn't say whether this server has it connected. Ask before relying on it.
H200-L
Full price
€12.00
per hour, whole instance, 4 GPUs
€3.00/h per GPU
8× H200141 GB each, 1128 GB total
Location not stated · 127 vCPU · 2136 GB RAM · 4096GB NVMe
Details
Stock and price checked 15 min ago.
You start it yourself once there is stock.
This GPU supports NVLink, but the provider doesn't say whether this server has it connected. Ask before relying on it.
H200-XL
Full price
€24.00
per hour, whole instance, 8 GPUs
€3.00/h per GPU
8× H200141 GB each, 1128 GB total
United Kingdom · 2048 GB RAM · 8x400 Gbit InfiniBand · 2 x 960 GB NVMe, 6 x 3.84 TB NVMe
Details
Stock and price checked 15 min ago.
You start it yourself once there is stock.
This GPU supports NVLink, but the provider doesn't say whether this server has it connected. Ask before relying on it.
bm3-ai-1xlarge-h200-141-8
Full price
€25.45
per hour, whole instance, 8 GPUs
€3.18/h per GPU
Spot €21.82/h instance · €2.73/h per GPU · no stock data
8× H200141 GB each, 1128 GB total
Canada · 176 vCPU · 1800 GB RAM · 100GB root + 48000GB ephemeral
Details
Stock and price checked 15 min ago.
You start it yourself once there is stock.
The hardware type or the provider confirms a fast GPU-to-GPU link. Software still has to split the work across GPUs. They never appear as one big memory.
n3-H200-SXM5x8
Full price
€28.46
per hour, whole instance, 8 GPUs
€3.56/h per GPU
Full price
from €4.82
per GPU, per hour
Spot from €1.25
Full price
from €4.82
per GPU, per hour
Spot from €0.98
Price and availability history
Price per GPU-hour
Excluding VAT.
Providers with stock
Each provider counts once. Full price and spot are separate.
Not enough history in this range yet. We record every 10 minutes since 9 November 2026.
Share of offers in stock
Full price. Offers we can't say are really sold there are left out until they've been in stock once.
Not enough history in this range yet. We record every 10 minutes since 9 November 2026.
Units free
No provider reports a number for this GPU.
Not enough history in this range yet. We record every 10 minutes since 9 November 2026.
Stock by provider
In stock, yes or no, for every provider selling this GPU. A filled band means in stock: green at full price, orange as spot.
Not enough history in this range yet. We record every 10 minutes since 9 November 2026.
Using several together
Depends on the version. The SXM version is fully linked with NVLink at up to 900 GB/s. The PCIe version links in pairs at best, and only if the provider has fitted a bridge. Software still has to split the work, so the memory doesn't act as one big pool.
Specifications
Headline values are SXM. NVL (PCIe): FP32 60, FP64 30 TFLOPS. NVIDIA H200 documents opened give no CUDA core count or NVLink generation. maxGpusInFabric is the largest NVLink domain the H200 page documents (HGX boards with 4 or 8 GPUs; H200 NVL bridges up to 4). Only H200 NVL is specified as air-cooled, so cooling stays null.
Memory
- VRAM
- 141 GB
- Memory type
- HBM3e
- Memory bandwidthHow fast the GPU reaches its memory. Often limits LLM speed
- 4,800 GB/s
Compute
- FP64Speed at 64-bit precision, used for scientific simulation
- 34 TFLOPS
- FP32Speed at 32-bit precision
- 67 TFLOPS
- FP16Speed at 16-bit precision
- 1,979 TFLOPS
- BF1616-bit AI format. Standard for training and full-quality inference
- 1,979 TFLOPS
- FP88-bit AI format. About half the memory of BF16, slightly lower quality
- 3,958 TFLOPS
- INT88-bit integer speed, used by quantized models
- 3,958 TOPS
- FP8 support
- Yes
- FP4 support
- No
- Transformer EngineHardware that speeds up FP8 and FP4
- Yes
Multi-GPU and connectivity
- Interconnect
- NVLink
- Interconnect bandwidthHow fast linked GPUs exchange data
- 900 GB/s
- Max GPUs in fabric
- 8
- PCIe generation
- 5
- PCIe bandwidth
- 128 GB/s
Sharing and security
- MIG partitioningSplits one GPU into several isolated smaller GPUs
- Yes
- Max MIG instances
- 7
- Confidential computing
- Yes
Hardware
- Architecture
- Hopper
- Launch year
- 2023
- Process node
- TSMC 4N
- Transistors
- 80 billion
- Power (TDP)Maximum power draw
- 700 W
Video
- NVDEC engines
- 7
Sources (31)
- https://www.nvidia.com/en-us/data-center/h200/
- https://dam-cdn.nvd.orangelogic.com/AssetLink/5o2qgy5d2835ve2pm11i62kv8mphqta8.pdf
- https://docs.nvidia.com/datacenter/tesla/mig-user-guide/supported-gpus.html
- https://docs.nvidia.com/590trd1-trusted-computing-solutions-release-notes.pdf
- https://nvidianews.nvidia.com/news/nvidia-supercharges-hopper-the-worlds-leading-ai-computing-platform
- https://developer.nvidia.com/blog/inside-nvidia-blackwell-ultra-the-chip-powering-the-ai-factory-era/