GPU Types and Configurations

A GPU is a physical piece of hardware, usually a rectangular circuit board with a large cooling system attached. But the term “GPU” also refers to the chip on that board — and that chip can exist in many different forms, including multi‑chip modules and full multi‑GPU clusters.

Below is a clear, structured breakdown with visuals.

🖼️ What a GPU looks like

A typical discrete GPU (the kind you plug into a desktop PC) looks like:

• A long rectangular card
• One to three cooling fans
• A metal heatsink under the fans
• A PCIe connector at the bottom
• Video output ports (HDMI, DisplayPort) on the back
  Under the cooler is the GPU die — the actual processor — plus VRAM chips around it.

Integrated GPUs look different: they’re just part of the CPU package and have no separate card.

🧩 The different kinds of GPUs

Below are the major categories, each starting with a Guided Link so you can dive deeper.

• Consumer GPUs — Gaming cards like RTX 4090, 4080, 3080. Powerful but limited VRAM. gpucorehost.com. Different GPU Types Explained – Choosing the Right GPU for AI Workloads
• Professional workstation GPUs — NVIDIA RTX A6000, Quadro. More VRAM, more stable drivers. gpucorehost.com. Different GPU Types Explained – Choosing the Right GPU for AI Workloads
• Data‑center training GPUs — A100, H100. Massive VRAM (40–80GB+), designed for multi‑GPU clusters. gpucorehost.com. Different GPU Types Explained – Choosing the Right GPU for AI Workloads
• Data‑center inference GPUs — L4, T4. Optimized for serving models, not training. gpucorehost.com. Different GPU Types Explained – Choosing the Right GPU for AI Workloads
• Integrated GPUs — Built into CPUs, share system RAM. Low power, low performance. altinix.io. Understanding GPU Types and Their Real-World Use Cases | Altinix
• Embedded/automotive GPUs — NVIDIA Drive PX, used in autonomous vehicles. Semiconductor for You. Graphics Processing Unit (GPU): Types, Functions, and Applications Explained - Semiconductor for You
• External GPUs (eGPUs) — GPU in a Thunderbolt enclosure for laptops. Semiconductor for You. Graphics Processing Unit (GPU): Types, Functions, and Applications Explained - Semiconductor for You

🧱 Are there “stacks” of GPUs?

Yes — multiple meanings of “stack” exist in GPU computing.

1. Hardware GPU stacks (multi‑GPU systems)

This is the literal “stacking” of GPUs:

• Servers with 4, 8, or 16 GPUs connected together
• High‑speed links like NVLink or NVSwitch for fast GPU‑to‑GPU communication
• Used for training large AI models
  NVIDIA A100 and H100 specifically support NVLink for clustering. Servermall. Server GPU vs Consumer GPU: ECC, VRAM, MIG/vGPU, NVLink & TCO (2026) 🚀

2. Software GPU stacks

This refers to the layers of software required to use GPUs:

• Drivers
• CUDA / ROCm
• Frameworks (PyTorch, TensorFlow)
• Distributed training libraries
  NVIDIA calls this the AI software stack.

3. Multi‑chip GPU “stacks” (chiplets)

Some modern GPUs use multiple silicon dies inside one package — a vertical or horizontal “stack” of compute tiles.

Intel and NVIDIA have explored multi‑chip GPU architectures. Intel. Multi-Stack GPU Architecture

🧠 Why your downloaded models run fine on your PC (even with few GPUs)

Since you asked earlier:
Models like Gemma 4 or Qwen 3.6 run on your desktop because:

• They’re quantized (smaller versions)
• They fit into your GPU’s VRAM
• Ollama automatically chooses CPU fallback when needed
• Modern consumer GPUs (even mid‑range ones) are extremely capable for inference
  Training huge models is impossible on a small GPU, but running them is often easy.

🧭 Want to go deeper?

Pick one:

• GPU architecture basics
• How multi‑GPU clusters work
• Why NVLink matters
• Difference between GPU and CPU parallelism
  Jud, if you want, I can also explain exactly how your PC handles Gemma/Qwen with limited VRAM — that’s a fun one.

[gpucorehost.com

Different GPU Types Explained – Choosing the Right GPU for AI Workloads](https://gpucorehost.com/resources/guides/different-gpu-types-explained/?utm_source=copilot.com)\[altinix.io

Understanding GPU Types and Their Real-World Use Cases | Altinix](https://altinix.io/blog/understanding-gpu-types-and-their-real-world-use-cases/?utm_source=copilot.com)Message Copilot