Run AI models on H100, H200 and L40S GPU resources in a cloud located in Türkiye
Regions: tr-ist-01 (İstanbul) · tr-ank-01 (Ankara).
Training data sits in object storage and is read into the GPU node in the chosen Türkiye region.
The multi-GPU node trains the model; regular checkpoints preserve progress.
An approved checkpoint is written to the model registry and released to the inference endpoint as a version.
Application requests reach the inference endpoint, which scales with demand.
Answer
GPU / AI Compute provides GPU capacity for training, fine-tuning and inference. You choose capacity without buying hardware.
Include GPU, storage and networking needs in your cost estimate.Calculate Cost
Live capacity and quota depend on the selected resource; service level terms are explained on the SLA page.
Features
Choose the compute and data resources your model needs together.
How it works
Define your workload, prepare your resources and start with a small test.
Define your model's memory need, dataset size and target runtime. Training, fine-tuning and inference may need different resource profiles. Compare H100, H200 and L40S options with those requirements, and run a small test on your own data.
Plan CPU and RAM, working disks, dataset storage and network traffic alongside the GPU. Object Storage can hold datasets and model outputs. Decide who needs access and what must be backed up before deployment.
Total cost depends on GPU model and count, runtime, CPU/RAM, storage and network usage. Attached disks, public IPs and backup are separate line items. You can see current prices in the cost calculator.
Choose a resource profile and create your server in the console. Install a GPU driver compatible with the operating system and the dependencies your application needs. Follow the Ubuntu NVIDIA driver guide on Discuss. Run a small sample of your dataset, measure memory use and runtime, and adjust capacity using those results.
PortvMind Cloud provides GPU resource access and the control plane. Models, data, the application environment and access permissions belong to your team.
Scenarios
Consider GPU resources for the workloads below. Resource needs vary with your model and dataset.
Decision criteria
Before you start, review operational responsibilities, setup steps and cost components.
FAQ
Yes. PortvMind Cloud offers GPU servers for AI/ML training, inference and compute-intensive workloads. The options are H100, H200 and L40S.
Start with model memory needs, data volume and target runtime. Training and inference may need different resource profiles. Compare the options, test with your own workload and review the results with the team to choose capacity.
GPU server cost varies by GPU model and count, CPU/RAM, storage, network usage and runtime. Configure current options in the PortvMind Cloud cost calculator or consult the expert team for a custom workload.
After creating your server in the console, install a GPU driver compatible with your operating system and the dependencies your application needs. For Ubuntu, follow the NVIDIA driver installation guide on Discuss. Start with a small dataset to check the setup, memory use and runtime.
Next step
Create the resource in the console. For a migration or scaling plan, continue together in an architecture consultation.
Model your scenario in the cost calculator.Calculate Cost
