Space for Fakhrul Islam
Space for Fakhrul Islam
Breadcrumbs

Cloud Provider Support for Pixel Streaming

Our platform supports deployment across all major cloud providers worldwide. Customers are free to choose the cloud provider and region that best fits their latency, compliance, and business requirements, as long as the infrastructure meets the required GPU specifications.

This flexibility allows you to deploy Pixel Streaming workloads close to your end users while maintaining high performance and reliability.



Supported Cloud Providers

We support any cloud provider offering NVIDIA GPU–enabled virtual machines, including but not limited to:

Global Cloud Providers

  • AWS (Amazon Web Services) – Multiple regions worldwide

  • Microsoft Azure – Multiple global regions

  • Google Cloud Platform (GCP) – Regions with NVIDIA GPU availability

Regional / Local Providers

  • Oracle Cloud Infrastructure (OCI) – Regional availability varies

  • Other sovereign or local cloud providers offering NVIDIA GPU instances

Any cloud provider operating in your target region with supported NVIDIA GPUs can be used with our platform.





GPU Requirement

⚠️ NVIDIA GPU is mandatory

To run Pixel Streaming workloads, your cloud infrastructure must meet the following requirement:

  • NVIDIA GPU–based virtual machines

  • Supported GPUs include (depending on provider availability):

    • NVIDIA T4

    • NVIDIA A10

    • NVIDIA L4

    • NVIDIA A16

    • Other NVIDIA GPUs compatible with Unreal Engine Pixel Streaming

CPU-only or non-NVIDIA GPU instances are not supported.




On-Demand VM Availability Considerations

Some regions may experience high demand for GPU-based virtual machines. As a result:

  • On-demand GPU VMs may not always be immediately available

  • VM creation may fail during peak usage periods

This is a cloud-provider limitation, not a platform issue.



To avoid availability issues, we strongly recommend:

  • Pre-booking or reserving GPU VMs where possible

  • Using Reserved Instances, Capacity Reservations, or Committed-Use Discounts, depending on the cloud provider

  • Keeping VMs provisioned ahead of critical launches, demos, or events



Benefits of pre-booking:

  • Guaranteed GPU availability

  • Stable and predictable perform

  • No last-minute deployment failures



If local GPU capacity is exhausted:

  • Deploy in other nearby regions with available NVIDIA GPU VMs

  • This ensures uninterrupted service and avoids failures due to regional demand spikes





Real-World Example

One of our customers serves end users based in the Middle East and initially deployed their application in the AWS UAE (Middle East) region.

Due to high demand for GPU-based on-demand VMs in that region, they frequently encountered VM provisioning failures. To resolve this:

  • The customer migrated their deployment to the AWS Mumbai region

  • GPU VM availability was immediately restored

  • The application has been running reliably since the region change

This example highlights how regional capacity constraints can impact on-demand GPU availability and why region flexibility or VM pre-booking is often the best solution.






Auto-Scaling Support

Auto-scaling allows cloud environments to dynamically adjust the number of active virtual machines based on the application’s real-time usage and demand.

  • When user demand increases, additional VMs are automatically provisioned to maintain performance.

  • When demand decreases, VMs are de-provisioned to optimize cost.

Supported Providers for Auto-Scaling:

  • AWS

  • CoreWeave

  • Google Cloud Platform (GCP)

  • Microsoft Azure

Note: Auto-scaling is not currently available on other cloud providers. Customers using these providers must manually scale VM resources or contact their cloud provider for assistance.








Deployment Recommendations

  1. Choose regions close to end users to minimize latency.

  2. Pre-book or reserve GPU instances in high-demand periods to ensure availability.

  3. Consider auto-scaling if supported by your provider to handle variable traffic efficiently.

  4. Select GPU type based on workload:

    • T4, A10, L4: General Pixel Streaming workloads

    • A16 or higher: Graphics-intensive or VR applications

  5. Validate on-demand VM availability before critical launches.