Give accelerated workloads their own physical environment in Switzerland. Tensor combines NVIDIA Tesla options with dedicated server resources for compatible inference, rendering and GPU processing workflows.
Dedicated GPU hardware combining dual Xeon E5-2680 v4 processors, a listed Tesla P4 / T4 GPU configuration, 64 GB of DDR4 memory and a 1 TB SSD. The network specification is a 1 Gbps port with unmetered traffic.
A GPU configuration with a 10 Gbps port for accelerated projects that also move larger datasets. The listed hardware includes dual Xeon E5-2680 v4 processors, Tesla P4 / T4 GPUs, 64 GB of DDR4 memory and a 1 TB SSD.
Dual Xeon Gold 6134 processors accompany the listed Tesla P4 / T4 GPU configuration for dedicated accelerated compute. This plan includes 64 GB of DDR4 memory, a 1 TB SSD and a 1 Gbps port with unmetered traffic.
Compare the listed specifications and base prices. Final availability, delivery and applicable tax are confirmed on your quote. Illustrations are conceptual.
Understand what makes this service family useful.
The Tensor range is listed as bare metal. Plan your operating system, drivers and application stack around the selected hardware rather than a generic virtual instance.
The configurator distinguishes NVIDIA Tesla P4 and T4 options. Compare the listed GPU memory with the model or task you want to run.
A GPU workflow also needs host memory, data preparation and storage. Review the Xeon platform and select RAM or disk additions where useful.
Compare the port and traffic characteristics across Tensor configurations. Dataset imports and output distribution can be as important as the accelerator itself.
Run inference workloads whose framework, precision and memory requirements match the selected NVIDIA accelerator.
Prepare a rendering or visual-processing workflow after checking the engine's GPU and driver compatibility.
Use an accelerator-aware application for repeatable image, media or numerical tasks on data you are entitled to process.
Before selecting a server, check whether the application and its working data fit the chosen accelerator's memory. The model architecture, precision and batch size all affect this requirement. A larger host-RAM option does not automatically increase the GPU's own memory.
Choose an operating system supported by your software, then verify the NVIDIA driver and framework requirements. CPU preprocessing, storage reads and data transfer can limit a workflow even when the GPU is suitable. Benchmark a representative task before scheduling a large production job.
Tensor gives you a Swiss location for an application stack you configure. Decide who may access datasets, models and results, and keep sensitive material out of public directories. Use licensed software and authorised datasets; no external AI service or data-processing claim is implied by the presence of a GPU.
Treat the selected accelerator, driver and application as one compatibility decision. CUDA-based work, TensorFlow or PyTorch, scientific processing and GPU rendering need a supported software combination. Video workflows may also use FFmpeg and hardware encoding, but supported codecs and practical throughput must be checked for the chosen P4 or T4. Confirm installation work and operator access before delivery.
Available choices depend on the selected configuration. See included selections and option prices in the configurator.
Specific answers for this service family.
Compare the P4 and T4 options with your software's support list and memory needs. The selected option, rather than the family headline, defines the accelerator configuration.
No. System RAM and the GPU's own memory are different resources. Review the accelerator option and host-memory option separately.
The catalogue lists operating-system and software choices. Confirm any required framework, driver and installation work as part of your deployment plan.
Check the application's supported accelerator, driver and operating-system requirements. A GPU listed in a server specification does not establish compatibility with every engine.
Compare the CPU platform and network port as well as the GPU selection. A workload with large data transfers may need a different balance from a compute-heavy local batch.
The published Tensor choices list Tesla P4 and T4. Other accelerators or several GPUs are not included catalogue selections and would require a separately validated configuration.
Check the chosen GPU, driver, toolkit and framework versions together. Installation and a working test job need confirmation; listing an NVIDIA accelerator does not make a framework preconfigured.
Evaluate FFmpeg, hardware-encoding support and the codecs required by your workflow. HLS, DASH, WebRTC or RTMP support depends on the software you deploy, and no stream count or latency is promised.
The plan lists its port, unmetered traffic and five IPs. Confirm root or administrator access, recovery controls and baseline protection with the operator; additional IP and software options are separate.
Account for weights, activations, caches and temporary buffers, then test the peak workload rather than the model file alone.
Read the guideSeparate model inference from orchestration, size the actual worker and limit what its tools can change before putting an AI workflow on a server.
Read the guideChoose your resources. Keep control of your identity.