The facts you need to understand the offering
Kernodeck is the trade name presented on this site. The catalog and the order file provide the information needed to choose and track a rental. A card's memory and a batch's composition are read separately from the preparation terms.
Scroll the table to read all columns.| Topic | Value or rule | Condition or limit |
|---|---|---|
| Audience | Developers and self-reliant users | The choice of software, models and processing belongs to the user. |
| Catalog | 15 NVIDIA and AMD models | A GPU's capabilities do not by themselves describe its server. |
| Availability | Stock shown per model in the catalog | Stated stock; the delivery date and actual configuration are verified separately. |
| Periods | 3, 7, or 30 days | Rental durations; no compute time is guaranteed by this choice. |
| Quantities | 1 to 10 lots | One GPU per batch, except B200: two GPUs. |
| Pricing | USD per batch and per duration | The total multiplies this plan by the number of batches; the cards in the batch are already included. |
| Software preparation | Ubuntu, PyTorch, Blender or custom | A preference requested during configuration, not proof of an already-tested installation. |
| Account and payment | Email/password account; KYC-free crypto payment | First name, last name and email are still required. No identity document is requested. |
| Tracking | Order, payment and preparation | Reporting a transfer, its confirmation and its availability are distinct. |
Project autonomy, with order tracking
You choose your code, libraries, data and the computations to run. Kernodeck does not inspect the content of your files, prompts or computations. If you request help, select the relevant technical elements and remove any access secrets before sharing them.
This policy does not mean that no data is retained. The account, contact details, orders, balance and payment operations have their own tracking information. Your application may also generate its own logs. The absence of a KYC procedure is therefore not a promise of absolute anonymity.
Runnable resources, with their scope of proof
Kernodeck guides start from a concrete problem: understanding why PyTorch does not see a GPU, then verifying that training resumes from the same point. The downloadable files include code, instructions and reports. You can examine the method before adapting it to your project.
The diagnosis was tested on CPU and on an RTX 5070 with CUDA; the resumption exercise was tested on CPU. These tests validate the resources in the documented environments, without measuring the GPUs in the catalog or describing the software installed on a rental. The files specify the versions and results.
Scroll the table to read all columns.| Resource | Input and protocol | Result provided | Limitation |
|---|---|---|---|
| PyTorch diagnostic v1 | 2 × 2 float32 matrices; product, gradient and requested device check. | Product and gradient verified; loss 196 on CPU and CUDA in the described environment. | Neither a benchmark, nor a business-model test, nor a ROCm hardware run. |
| Resumption v1 | 24 synthetic rows; 10 steps versus 5 + 5; four fresh processes, CPU float64. | Full resumption: zero deviations in the proof. Omitting the RNGs: divergence detected. | No GPU proof, no AMP, no distributed training; absolute tolerance 1e-12 and relative tolerance 0. |
Link the command to a project you know how to resume
Before setting up a rental, keep a version of the code, the dependencies, a representative input and a result criterion. Specify the software preparation you want, then check the pipeline actually provided. This method helps change one variable at a time when you compare trials.
The record keeps the GPU, the duration, the batches and the total. After the crypto transfer, "I've paid" signals your action; settlement confirmation follows verification. Preparation is tracked separately. For your own work, plan a usable checkpoint and verify the export of results during the chosen period.