33.2b vGPU profiles: Q-series (Quadro/Workstation), C-series (Compute/AI), A-series (apps)¶

🔍 Context Introduction¶

When you share a physical NVIDIA GPU across multiple virtual machines (VMs) using vGPU technology, you need to choose the right profile for each VM. Think of a vGPU profile as a predefined slice of the GPU's capabilities — it determines how much memory, compute power, and graphics performance each VM gets. NVIDIA offers three main profile families: Q-series, C-series, and A-series. Each is designed for a specific type of workload, and picking the wrong one can lead to poor performance or unexpected behavior.

⚙️ What Are vGPU Profiles?¶

A vGPU profile defines the resources allocated to a single VM from the physical GPU. These resources include:

• Frame buffer (video memory) — how much GPU memory the VM can use
• Compute units — how many CUDA cores or Tensor Cores are available
• Display capabilities — whether the VM can output to a monitor

Each profile is named with a letter (Q, C, or A) followed by a number (e.g., Q‑4, C‑8, A‑2). The letter indicates the use case, and the number typically represents the amount of frame buffer in gigabytes.

📊 Profile Families at a Glance¶

🛠️ Deep Dive into Each Series¶

🖥️ Q-series (Quadro / Workstation)¶

The Q-series profiles are designed for professional graphics workloads. If a VM needs to run applications like AutoCAD, SolidWorks, or Blender with full GPU acceleration, you want a Q-series profile.

Key characteristics: - Supports display outputs — the VM can drive physical monitors - Full support for OpenGL and DirectX APIs - Certified for ISV (Independent Software Vendor) applications - Includes NVIDIA Quadro drivers inside the VM - Typically used in design, engineering, and media environments

When to use Q-series: - You need to render 3D models with high polygon counts - The VM must output video to a physical display - The application requires Quadro-specific driver optimizations

🧮 C-series (Compute / AI)¶

The C-series profiles are built for compute-intensive workloads — specifically AI training, inference, and scientific simulations. These profiles disable display output to dedicate all GPU resources to computation.

Key characteristics: - No display output — the VM cannot drive monitors - Full access to CUDA cores and Tensor Cores - Optimized for batch processing and parallel workloads - Uses NVIDIA Compute drivers inside the VM - Supports NVIDIA AI Enterprise software stack

When to use C-series: - You are training deep learning models (PyTorch, TensorFlow) - Running inference pipelines for AI applications - Performing HPC simulations (molecular dynamics, weather modeling) - The VM does not need a graphical desktop

🖱️ A-series (Apps)¶

The A-series profiles are designed for virtual desktop infrastructure (VDI) and application streaming. They provide a balance between graphics performance and resource efficiency, making them ideal for office productivity and light creative work.

Key characteristics: - Supports display outputs for virtual desktop sessions - Optimized for application streaming (e.g., Microsoft Remote Desktop, VMware Horizon) - Lower memory footprint compared to Q-series - Good for office applications, web browsing, and video playback - Uses NVIDIA GRID drivers inside the VM

When to use A-series: - You are deploying virtual desktops for knowledge workers - Users need to run Microsoft Office, web browsers, or video conferencing - You want to maximize the number of VMs per physical GPU - Lightweight graphics acceleration is sufficient

📊 Visual Representation: vGPU Profile sizing and naming structure¶

This diagram breaks down vGPU profiles: assigning fixed VRAM slices and matching specific compute targets.

🕵️ How to Choose the Right Profile¶

Choosing between Q, C, and A series depends on three questions:

1. Does the VM need a display output?
2. Yes → Q-series or A-series

3. No → C-series

4. What is the primary workload?

5. Professional 3D graphics → Q-series
6. AI/ML training or HPC → C-series

7. Office productivity or app streaming → A-series

8. How many VMs do you need per GPU?

9. Few VMs with high performance → Q-series or C-series (larger profiles)
10. Many VMs with moderate performance → A-series (smaller profiles)

🧪 Practical Example: Mapping Workloads to Profiles¶

Imagine you have one physical NVIDIA A100 GPU with 40 GB of memory. You want to support three different VMs:

• VM1: A designer running SolidWorks (needs display output, high graphics performance)

• Profile: Q‑16 (16 GB frame buffer, Quadro drivers)

• VM2: A data scientist training a neural network (no display needed, needs CUDA)

• Profile: C‑16 (16 GB frame buffer, Compute drivers)

• VM3: A remote worker using Microsoft Office (needs display, light graphics)

• Profile: A‑8 (8 GB frame buffer, GRID drivers)

Total allocated: 16 + 16 + 8 = 40 GB — exactly matching the physical GPU memory.

✅ Key Takeaways for New Engineers¶

• Q-series = Professional graphics + display output (think: CAD, 3D rendering)
• C-series = Compute + no display (think: AI training, HPC)
• A-series = Virtual desktops + light graphics (think: office apps, VDI)
• The letter tells you the workload type; the number tells you the memory allocation
• Always match the driver type inside the VM to the profile series (Quadro, Compute, or GRID)
• You can mix different profile series on the same physical GPU, as long as total memory does not exceed the GPU's capacity

📚 Further Learning¶

• NVIDIA vGPU documentation: NVIDIA Virtual GPU Software Documentation
• Profile compatibility matrix: NVIDIA vGPU Supported GPUs and Profiles
• Driver selection guide: NVIDIA vGPU Driver Types and Installation