Provision GPU Resources

Provision GPU Resources
Lesson 3 · Provisioning

Provision GPU Resources

Not every workload wants to run on a CPU. When graphics, rendering, or parallel math become the bottleneck, offloading to a GPU co-processor is usually the fix — the question is whether that GPU is shared or dedicated.

3.4

Provision GPU Resources

Workloads can sometimes be offloaded from the CPU and passed to other co-processors or hardware accelerators. A common solution for this offloading is the Graphics Processing Unit (GPU). AWS, Azure, and GCP all use NVIDIA GPUs for this purpose.

CPU
general-purpose processing
offloads →
GPU co-processor
parallel / graphics workload

These processors benefit VMs running specific kinds of work. Tap a use case:

Game streaming
Cloud gaming VMs
Remote graphics workstations
Design & CAD work
3D model development
Modeling & rendering
3.4a

Shared vGPUs vs. Pass-Through GPUs

Like CPUs, GPUs are exposed to VMs as available hardware — a Virtual GPU, or vGPU. But there are two ways to hand that hardware out: sliced across many VMs, or dedicated entirely to one.

3.4b

Preconfigured GPU Instances

Just like memory- or storage-optimized instances, cloud administrators select GPU-capable configurations from what the provider already offers — rather than assembling one from scratch.

NVIDIA on the AWS Marketplace

NVIDIA offers preconfigured graphics-intensive VM workstation instances directly on the AWS Marketplace. For example, an administrator could select an Ubuntu Linux platform with an NVIDIA GPU configured specifically for cloud-based gaming — no manual GPU driver or graphics stack setup required.

Ubuntu + NVIDIA
Cloud gaming
Workstation-ready
3.4c

Manage Clock Speed

Clock speed is a common performance measurement for CPUs. Tuning it for the workload — a computationally intensive science rendering program, for instance — can buy real time efficiency. As high-performance computing workloads scale up, the impact on CPU performance grows with them.

Relative processing time
§ 3.4 · Provision GPU Resources