How to choose the right GPU instance at Hostinger

Hostinger GPU instance selection depends on matching Video Random Access Memory (VRAM) capacity and hardware allocation to your specific machine learning workload. Selecting the optimal model involves evaluating VRAM requirements, choosing between shared and dedicated infrastructure, and managing hourly compute billing.

Available Hostinger GPU instance models

hPanel displays four Hostinger GPU instance models to support Artificial Intelligence (AI) and Machine Learning (ML) workloads:

VRAM capacity and workload requirements

Video Random Access Memory (VRAM) is the memory dedicated to the Graphics Processing Unit (GPU). VRAM determines the maximum model size, dataset size, and batch volume your Hostinger GPU instance can load. Compute execution speed depends on GPU architecture rather than total VRAM capacity.

Match your VRAM capacity to your specific AI task requirements:

• Lower VRAM capacities suit lighter workloads, including running smaller language models or basic inference tasks.
• Higher VRAM capacities support intensive workloads, including training large language models or generating high-resolution media.

Tasks fail to load or execute when a workload requires more VRAM than the Hostinger GPU instance provides. Selecting a model with extra VRAM does not increase task processing speed, but increases your hourly compute cost.

VRAM vs instance RAM

Video Random Access Memory (VRAM) and instance RAM are two different types of memory, and they are not related. VRAM is memory built into the Graphics Processing Unit (GPU) itself. Each GPU model has a fixed amount of VRAM, and it does not change no matter which instance size you choose. VRAM determines whether your AI model, dataset, or batch size can run on the GPU at all.

Instance RAM is separate memory used by the rest of the instance, not the GPU. You choose the amount of instance RAM when configuring your instance size, along with CPU cores and storage. Instance RAM handles tasks like loading files and running your operating system, not GPU computation.

For example, an instance with an L40S GPU has 48 GB of VRAM regardless of size. If you select a larger instance size for that same GPU, only the CPU cores, instance RAM, and storage increase. The 48 GB of VRAM stays the same.

NOTE

• You select your instance size, including CPU cores, instance RAM, and storage, only at deployment. You cannot upgrade the instance size of an active Hostinger GPU instance afterward.

Differences between B200 and B200 (Dedicated)

Both the B200 and B200 (Dedicated) models provide 192 GB of VRAM. Hardware allocation creates the operational distinction between these two models:

• B200 runs on shared infrastructure alongside workloads from other Hostinger customers.
• B200 (Dedicated) provides non-shared, dedicated access to underlying server hardware.

Choose B200 (Dedicated) if your production workload requires consistent performance without resource variance from shared hardware. Choose the standard B200 model if your workload tolerates slight performance variation at a lower cost.

Match a Hostinger GPU instance to your workload

Select a Hostinger GPU instance model based on your specific task complexity and memory requirements:

Balance cost and performance

Hostinger bills GPU instances on an hourly basis, with higher-specification models costing more per hour. Review current rates on the Hostinger GPU pricing page.

Deploy a smaller GPU instance first to optimize compute expenses. Move to a larger model only after encountering VRAM or compute limitations.

Frequently asked questions

Can I switch GPU models or upgrade my instance size after deploying?

No. You cannot change the GPU model or upgrade the instance size, including CPU cores, instance RAM, or storage, of an active Hostinger GPU instance after deployment. Deploy a new Hostinger GPU instance to use a different GPU model or a different instance size.

Is a GPU instance guaranteed to be available?

GPU instances do not currently have specific availability constraints. Availability may fluctuate during the beta release period.

Which GPU instance should I choose if I am not sure?

Deploy an L40S instance for lightweight workloads or an A100 80GB PCIe instance for general machine learning tasks if your requirements are uncertain. You can deploy a new Hostinger GPU instance with a larger model later if your workload requires additional VRAM.

Selecting the right Hostinger GPU instance requires aligning VRAM capacity with model size and choosing between shared or dedicated hardware. Starting with a smaller instance allows you to verify performance requirements while maintaining cost efficiency.

NOTES

• Check template documentation for specific VRAM requirements when deploying an application template, including Unsloth or Ollama. Unsloth QLoRA fine-tuning fits on smaller GPU instances, but full fine-tuning requires 48 GB of VRAM or more.
• See how to choose between a Hostinger VPS and GPU instance to decide whether to use GPU or VPS.