NVIDIA H100 NVL
H100 NVL is a 94 GB PCIe GPU often discussed as a two-card, 188 GB pair. Always count how many cards the listing includes.
HBM3
Published peak, not measured application throughput
188 GB is the pair total, not the memory on one GPU. Azure NCads H100 v5 uses NVL; Azure ND H100 v5 is a different system configuration.
When this is a sensible choice
Start here if…
The bridged pair is a useful candidate for a 70B BF16 deployment: roughly 140 GB of raw weights can be split across two 94 GB cards. Runtime buffers and attention state consume the remaining capacity.
Choose another configuration if…
Use a switched SXM system for larger tensor-parallel groups when its topology fits the job. A pair’s total memory tells you nothing about a larger server’s cross-pair communication.
Specifications with their boundaries attached
- Architecture
- Hopper
- Memory
- 94 GB HBM3
- Memory bandwidth
- 3.9 TB/s per accelerator
- Peak compute
- 835.5 TFLOPS BF16/FP16; 1,670.5 TFLOPS FP8, dense per GPU
- Scale-up interconnect
- NVLink: 600 GB/s bidirectional per GPU; bridged configuration
- Host attachment
- PCIe 5.0 x16
- Power
- 350–400 W
- Partitioning
- Up to seven MIG instances
- Catalogue status
- Documented product
Compute figures are theoretical peaks at the stated precision. Dense and structured-sparse rates must not be mixed. Bandwidth labelled bidirectional combines both directions. See the source documents.
Three bandwidths, three different jobs
NVLink: 600 GB/s bidirectional per GPU; bridged configuration
InfiniBand, RoCE, EFA or provider-specific transport
Where it appears in provider documentation
| Provider / machine | Network scope | What changes the decision |
|---|---|---|
Azure NCads H100 v5 / ND H200 v5 ↗ | Different NC and ND system configurations | NCads H100 uses NVL; ND H200 uses H200. Confirm full size specifications before comparing them. |
Model fit and software support
These publisher or serving-engine documents mention this hardware family. They have not been reproduced on our machines.
No model-specific recipe in our reviewed set certifies this exact hardware. The memory calculator can narrow candidates, but it cannot establish software support. Read the model register.
What is the memory floor?
Start with total parameters, then add the memory the workload needs. This arithmetic does not certify a serving configuration. All output sizes below are decimal GB.
Override device capacity with the memory exposed by your allocation, particularly for cloud B300 and partitioned devices. The starting 32 GB budget and 15% reserve are editable teaching assumptions. They are not measurements for the selected model. Mixed-precision tensors, quantization scales, vision encoders and draft models can increase the weight payload.
Raw weights only
320B × 8 bits ÷ 8
Budget per device
94 GB × (1 − 15%)
Arithmetic lower bound
round up ((weights + 32) ÷ budget)
This assumes perfectly balanced sharding. A result of three does not prove that a three-device parallel layout is supported. Check layer/expert divisibility, actual allocatable memory, precision kernels and the fabric before renting.
320B is the model card’s total; 18B active is not its storage size. The vLLM recipe reports about 306 GiB for the native FP8 checkpoint. Hopper requires BF16 KV for this model; the documented ROCm path targets gfx950, not every Instinct GPU. Read the model source ↗
For full training, also budget gradients, optimizer states, activations and communication buffers. The inference weight estimate above is insufficient.
Tokens per second, TTFT and TPOT are not certified for this hardware in our reference. Use the benchmark checklist to compare an exact model, software revision and workload.
Nearby memory capacities, different tradeoffs
These are comparison candidates selected by memory capacity, not performance rankings or drop-in replacements.
RTX PRO 6000 Blackwell Server Edition
96 GB of GDDR7 makes this an interesting inference and mixed AI/graphics card. Its memory capacity sits above H100, but its bandwidth does not.
RTX PRO 6000 Blackwell Workstation Edition
A desktop card with 96 GB of memory for local model work and graphics. Its 1.792 TB/s specification differs from the Server Edition’s 1.597 TB/s.
NVIDIA A100 80 GB SXM
80 GB of HBM on the SXM4 form factor. A100 remains useful for BF16 training and supported inference, but cannot execute Hopper’s native FP8 Tensor Core path.
Sources and review scope
Manufacturer and cloud documentation checked 2026-09-12. We reviewed specifications and the stated product boundaries; we did not run training or serving benchmarks on this device. Cloud memory and availability can differ by configuration.
- NVIDIA H100 SXM and NVL specifications ↗ · checked 2026-09-12
Cite this reference
AI Infra Interviews. NVIDIA H100 NVL: specifications and workload fit. Reviewed . https://aiinfrainterviews.com/hardware/h100-nvl
Include your access date when citing a changing specification. Link to the specification section for hardware figures or the explanation for a sizing or workload decision.
Research method and limits
We compile manufacturer specifications, cloud documentation, model cards and serving recipes. The reference preserves source units and distinguishes individual devices from nodes and racks. Conflicting figures and unknown fields remain labelled.
Our contribution is the comparison, unit reconciliation, worked arithmetic and workload explanation. Published peaks are vendor specifications. Calculator results are estimates under the displayed assumptions. Neither is a measurement from our own accelerator lab.
For a manufacturer’s specification, consult the original source documents. Cite our page when using its analysis, and retain primary-source attribution for underlying figures. This curated reference does not establish market share, live availability or a universal performance ranking.
Found a discrepancy? Send a correction with the page URL, exact variant, disputed figure and supporting primary document.