NVIDIA Hopper

NVIDIA H200 GPU cloud

Run NVIDIA H200 SXM GPUs with 141GB HBM3e, the same Hopper compute as the H100 with far more memory bandwidth, built for long-context LLMs and larger models without sharding.

Reserve capacity See pricing & availability

NVIDIA H200 SXM

GPU memory: 141GB HBM3e
Memory bandwidth: 4.8 TB/s

Technical specifications

Architecture: Hopper
GPU memory: 141GB HBM3e
Memory bandwidth: 4.8 TB/s
NVLink: 900 GB/s
FP16/BF16 (Tensor): 1,979 TFLOPS
FP8 (Tensor): 3,958 TFLOPS
Max TDP: 700W
GPUs per node: 8 (HGX H200)

*Peak performance with sparsity, per NVIDIA official specs. Final specs may vary by node configuration.

Pricing & availability

NVIDIA H200 SXMReserved only

Talk to sales

What's the H200 best for?

Long-context & large-model inference

141GB HBM3e fits 70B-class models in FP8 (and big KV caches) on a single GPU, so you serve longer context windows without tensor-parallel sharding.

Memory-bound training & fine-tuning

Larger batches and longer sequences fit in memory; 4.8 TB/s bandwidth keeps Hopper FP8 tensor cores fed for memory-bound workloads.

Drop-in Hopper upgrade

Same CUDA, PyTorch, and NeMo stack as the H100: move memory-constrained workloads over with no code changes and more headroom.

Compare NVIDIA data-center GPUs

	A100 Ampere	H100 Hopper	H200 You're viewing	B200 Blackwell	B300 Blackwell	GB300 Blackwell
Architecture	Ampere	Hopper	Hopper	Blackwell	Blackwell	Blackwell
GPU memory	80GB HBM2e	80GB HBM3	141GB HBM3e	192GB HBM3e	up to 288GB HBM3e	up to 288GB HBM3e
Memory bandwidth	2.0 TB/s	3.35 TB/s	4.8 TB/s	8 TB/s	8 TB/s	8 TB/s
FP8 (Tensor)	Not supported	3,958 TFLOPS	3,958 TFLOPS	9 PFLOPS	9 PFLOPS	10 PFLOPS
Access	from $1.48/hr	from $2.98/hr	Reserved only	Reserved only	Reserved only	Reserved only
Best for	Cost-efficient fine-tuning & inference	Cost-efficient training & inference	Long-context & large-model inference	Frontier-scale training (FP4)	Largest models & reasoning inference	Cluster-scale training & inference

Why AI teams run GPUs on VESSL Cloud

From A100 to Blackwell Ultra

Run A100 and H100 on demand. Reserve H200 and Blackwell GPUs for larger runs, so each workload gets the GPU it needs.

Issues caught early

Every VM Cluster node is health-checked continuously, with its status refreshed about once a minute. Workspaces and Jobs alert you when a run crosses a resource threshold or hits a runtime issue.

From one node to a cluster

Train and serve on up to 8 GPUs on one node, and move to a multi-node VM Cluster when one node is not enough.

Enterprise-ready

SOC 2 Type II compliance, with dedicated support for production AI.

Frequently asked questions

How do I get access to NVIDIA H200 GPUs?

H200 runs as reserved VM Cluster capacity. Talk to our team for current availability and pricing.

What's the difference between the H100 and H200?

Both share the same Hopper compute (1,979 TFLOPS FP16 / 3,958 TFLOPS FP8, with sparsity). The H200 carries 141GB HBM3e at 4.8 TB/s vs the H100's 80GB HBM3 at 3.35 TB/s, so it fits larger models, bigger batches, and longer context windows.

How much memory does the H200 have?

The H200 SXM has 141GB of HBM3e memory at 4.8 TB/s bandwidth, about 76% more capacity and 43% more bandwidth than the H100.

Can I run multi-node H200 training?

Yes. H200 runs as a VM Cluster: HGX H200 nodes (8 GPUs each) joined by high-speed InfiniBand for distributed training.

Should I pick the H200 or a Blackwell B200?

The H200 is the memory-rich Hopper option for large-model inference. If you need FP4 acceleration and 192GB+ for frontier-scale work, look at the Blackwell B200/B300. Talk to us and we'll help you choose.