GPUs for your projects · crypto payment without KYC How to rent
English
Open the console
KERNODECK / VOTRE CONFIGURATION

H100 SXM: 80 GB to expand your PyTorch project.

Your project needs more memory on a single card. Rent a NVIDIA H100 SXM 80GB from Kernodeck, with 80 GB per GPU for a Hopper-compatible application. The 3-day plan costs 203.57 USD for 1 lot of 1 GPUs: an option to consider when 48 GB limits your scenario.

  • 80 GB on one card
  • For your Hopper-compatible projects
  • Full price for 3, 7, or 30 days
At Kernodeck, no ID and no KYC process on this GPU rental path. Account required: first name, last name, email and password; that does not mean anonymity. Account data ↗
YOUR PERIOD BUDGET

Rent your GPU for 3, 7 or 30 days.

Each price covers 1 full lot for the entire duration. Choose your period, then the number of lots in the configurator.

1 GPUs includedHBM3
80GB per card

63 available cards.

Memory indicated per GPU. Choose the duration, then the number of lots in the configurator.

3 jours

1 GPU per lot

$203.57Rent 3 jours

7 jours

1 GPU per lot

$475.00Rent 7 jours

30 jours

1 GPU per lot

$1,690.00Rent 30 jours

A candidate when memory becomes your purchase criterion.

A PyTorch scenario that overflows a 48 GB card is worth comparing against the H100 SXM. For fine-tuning, count the activations and training states in addition to the weights; for inference, go by the allocations of your actual run. The choice becomes relevant if that need fits within 80 GB and your dependencies are Hopper-compatible.

For a series of configurations with evaluation and checkpoints, budget 475.00 USD for 1 lot of 1 H100 SXM over 7 days. If your priority is deciding whether the scenario suits this card, the 3-day plan is 203.57 USD. The period funds your work on the application, without guaranteeing a number of experiments or their outcome.

Same capacity or more room: compare the right criterion.

The H100 PCIe also has 80 GB per GPU and a lower plan in the Kernodeck catalog. For mostly single-card work, that's a useful comparison. The SXM name alone isn't enough to conclude there's an advantage for your application: the topology and the exchanges actually used remain decisive if you work with multiple cards.

The H200 SXM offers 141 GB per GPU. Consider it when the identified problem is the card's capacity, rather than ordering a second H100 on the assumption that the two memories will pool together. You're then comparing a higher memory need against its rental budget.

TO BREAK THE TIE

Two alternatives to compare.

7-day budgets for one lot. The choice depends on your application's memory and dependencies.

YOUR PAYMENT

Your plan, payable directly in crypto.

Pay for your Kernodeck rental directly in crypto, with no mandatory top-up: BTC on Bitcoin and USDT on Tron (TRC-20), among others. The process requires neither an ID document nor a KYC procedure; an account with first name, last name, email and password is required, with no promise of anonymity.

Compare the eight accepted assets and networks ↗

The terms that matter for your project

Availability. 63 declared batches. Declared quantities per model in the offers. They do not reserve future availability or a delivery date. Hosting country and served region to be confirmed before purchase if your project requires them.

Setup. Ubuntu, PyTorch, Blender or a custom request. Versions, CPU, RAM, storage, network and access mode to be confirmed per project; their inclusion is not implied by the GPU model.

Pricing and terms. USD amounts for the lot and the entire period. Tax status and any fees to be confirmed in the applicable terms.

Read the rental terms ↗
AVANT DE CHOISIR

Does this configuration fit your project?

How much does a lot of H100 SXM cost, and what does that unit include?

One lot corresponds to 1 H100 SXM of 80 GB. The full plans are 203.57 USD for 3 days, 475.00 USD for 7 days, and 1,690.00 USD for 30 days. The quantity chosen in the configurator corresponds to the number of lots; each additional lot adds its card and its price.

When do the H100 SXM's 80 GB justify this choice?

When your scenario exceeds a lower capacity and can fit within 80 GB, with a Hopper-compatible stack. For training, include activations, gradients, and optimizer states. For inference, take into account the application's allocations beyond the weights alone.

Why compare the H100 PCIe before choosing SXM?

The H100 PCIe keeps 80 GB per GPU with a lower plan in the Kernodeck catalog. So it's worth considering for a single-card project. A need for inter-card exchanges must be evaluated with the actual topology; the SXM label alone doesn't prove an application-level gain.

Should you take two H100s or move up to the H200 if 80 GB isn't enough?

If your application expects more memory on a single card, first compare the H200 SXM's 141 GB. Two H100s remain two 80 GB memories: using them together requires suitable software and a distribution plan. The number of cards doesn't replace that application architecture choice.

TO PREPARE YOUR WORK

From choosing a plan to your application.

Check the computation before comparing

Mixed precision and stabilityDecide when to use AMP and check what changes numerically.Profile a PyTorch stepScope the measurement and read a CPU/GPU trace without jumping to conclusions.

Start with a complete training step

Your first test should run through data loading, the forward pass, gradient computation, and the optimizer update. Measure this full cycle before increasing the batch. Weights are only part of the memory used: activations and training states must also fit within the 80 GB.

Keep a representative sequence of inputs, including the large examples. You'll avoid sizing the entire campaign based on a particularly easy first batch.

Mixed precision and controlled comparison

PyTorch lets you choose a precision suited to certain operations with autocast. Test this setting on your loss function and validation criteria. Enabling an option is not proof of numerical stability: keep a comparison run, metrics, and the exact parameters.

For multiple cards, distinguish per-GPU batch from global batch. The number of processes and gradient accumulation are part of the protocol, just like the learning rate.

SXM, PCIe, or more memory

The H100 PCIe keeps an 80 GB capacity and is worth comparing for primarily single-card work. For frequent GPU-to-GPU exchanges, measure the topology actually used by your application; the SXM name does not replace this check. Prefer the H200 if the identified problem is primarily a lack of memory per card.

Reserve a period that includes recovery

Choose 3 days to validate a loop and a checkpoint, 7 days for an ablation series, or 30 days for extended experiments. Include evaluation and export in that schedule, not just training epochs.

In the configurator, select the PyTorch or custom setup, your batches, and your duration. Create your account or log in before saving the order and choosing crypto payment. The "I've paid" button is used to report the transfer; payment tracking remains visible in your account.

Read this page in Markdown ↗