Hopper architecture
NVIDIA H200
The NVIDIA H200 pairs the Hopper architecture with 141 GB of HBM3e memory at 4.8 TB/s — the first GPU to ship with HBM3e — for demanding AI and long-context workloads.
- GPU memory
- 141 GB HBM3e
- Memory bandwidth
- 4.8 TB/s
- FP8 tensor
- 3,958 TFLOPS
Specifications as published by NVIDIA; Tensor Core ratings shown are sparsity-accelerated figures.
Overview
Why H200 fits
Hopper architecture, HBM3e memory
H200 is built for work that has outgrown smaller memory pools: large language models with long context windows, recommendation and search models over big embedding tables, and scientific workloads where bandwidth is the bottleneck. NVIDIA's current generation memory keeps data closer to compute, so fewer multi-GPU shards are needed for the same model.
On Scalenta's side, H200-class capacity is sourced through verified data center and hardware partners. We scope the whole instance — GPU form factor, count, interconnect, CPU, memory and storage — around the workload, and confirm what is genuinely deliverable before anything is committed in writing.
Specifications
H200 technical specifications
Published figures for the accelerator, followed by the variant and configuration caveats that apply.
- Architecture
- NVIDIA Hopper
- GPU memory
- 141 GB HBM3e
- Memory bandwidth
- 4.8 TB/s
- FP8 Tensor Core
- 3,958 TFLOPS (with sparsity)
- FP16 / BF16 Tensor Core
- 1,979 TFLOPS (with sparsity)
- TF32 Tensor Core
- 989 TFLOPS (with sparsity)
- FP64
- 34 TFLOPS
- Interconnect
- NVLink 900 GB/s · PCIe Gen5 128 GB/s
- Form factor
- SXM
- Max TDP
- Up to 700 W (configurable)
- Multi-Instance GPU
- Up to 7 partitions
- Media engines
- 7 NVDEC / 7 JPEG
- Confidential computing
- Supported
Variant & configuration note
Rows describe the H200 SXM form factor as published by NVIDIA. The PCIe-based H200 NVL variant differs (up to 600 W, 2- or 4-way NVLink bridge, up to 8 GPUs per server). Host configuration — GPU count, CPU, memory, storage and interconnect — is scoped per engagement; no fixed server build is implied.
Specifications describe the GPU itself as published by NVIDIA — not a fixed server configuration. Scalenta scopes complete instances (GPU count, interconnect, CPU, memory, storage) per engagement and states them in writing.
Use cases
Where H200 fits
Workload shapes this platform suits — the scoping conversation maps them to a concrete configuration.
Large-model training
Pretraining and fine-tuning runs where 141 GB of memory per GPU reduces the number of shards a model has to be split across.
Long-context inference
Serving LLMs with large context windows and retrieval pipelines where the KV cache and embeddings want a bigger memory pool.
HPC and science
Memory-bandwidth-bound simulation and analytics work — the H200 keeps the same power envelope as H100 with more memory and bandwidth.
Delivery model
Partner-sourced, scoped in writing
How GPU capacity is delivered at Scalenta — the same three commitments on every model page.
Partner-sourced capacity
Accelerator capacity is sourced through verified data center and hardware partners — no facility-ownership claims, no stock promises.
Scoped per engagement
Model, variant, GPU count, interconnect and host configuration are confirmed in writing before anything is committed.
Operated with 24/7 support
Deployments are covered by 24/7 support and SLA-backed managed services scoped to your environment.
Related GPU models
Nearby options in the catalogue
NVIDIA H100
The proven Hopper accelerator for LLM training and inference, with FP8 Transformer Engine support and 900 GB/s NVLink scale-up.
- GPU memory
- 80 GB HBM3
- Memory bandwidth
- 3.35 TB/s
- FP8 tensor
- 3,958 TFLOPS
NVIDIA A100
A proven training-and-HPC accelerator with an 80 GB memory pool per GPU and mature tooling across the AI ecosystem.
- GPU memory
- 80 GB HBM2e
- Memory bandwidth
- 2,039 GB/s
- FP16 tensor
- 624 TFLOPS
NVIDIA L40S
One card for generative AI, 3D graphics, rendering and video — the pragmatic choice where workloads refuse to stay in one lane.
- GPU memory
- 48 GB GDDR6 ECC
- Memory bandwidth
- 864 GB/s
- FP8 tensor
- 1,466 TFLOPS
FAQ
H200 — common questions
Do I need H200, or is H100 enough?
Is H200 capacity available today?
Can I start with one GPU and scale later?
Request a GPU consultation
Talk to us about NVIDIA H200 — workload-first scoping, written findings and a proposal with clear scope. No obligation.