GPU
Definition
Graphics Processing Unit - specialized hardware designed for parallel processing that accelerates AI training and graphics rendering.
Use Cases
- Netflix: Video encoding and transcoding acceleration for streaming — Netflix has used GPU-accelerated encoding to speed up compute-intensive video processing workflows, running encoding jobs on GPU-capable infrastructure where it improves throughput and efficiency. (Faster encoding pipelines and improved efficiency for compute-heavy video processing, helping deliver high-quality streams at scale.)
- BMW Group: AI and simulation workloads for engineering and autonomous driving research — BMW has used GPU-accelerated computing in high-performance environments to run large-scale simulations and train machine learning models that benefit from parallel processing. (Reduced time-to-insight for simulation and model training, enabling faster iteration in R&D workflows.)
- Adobe: Accelerating creative workloads such as image processing and AI-powered features — Adobe applications leverage GPU acceleration (locally and in cloud-backed services) to speed up parallelizable operations like rendering, filters, and AI-enhanced features. (More responsive creative tools and faster processing for graphics and AI-assisted workflows.)
Provider Equivalents
- AWS: Amazon EC2 Accelerated Computing (GPU instances, e.g., P5, P4d, G5, G6)
- Azure: Azure Virtual Machines (GPU VMs, e.g., ND, NC, NV series)
- GCP: Compute Engine GPU (NVIDIA GPUs attached to VMs; also A2 accelerator-optimized VMs)
- OCI: OCI Compute GPU Instances (e.g., NVIDIA A100/H100-based shapes, depending on region)
Frequently Asked Questions
- What's the difference between a GPU and a CPU?
- A CPU is designed for general-purpose computing and excels at a few complex tasks at a time (low parallelism). A GPU is designed to run many similar operations in parallel (high parallelism), which makes it much faster for workloads like deep learning training, matrix math, image/video processing, and 3D rendering.
- When should I use a GPU in the cloud?
- Use a GPU when your workload is highly parallel and spends most of its time on math-heavy operations—common examples are training deep learning models, running high-throughput AI inference, 3D rendering, video transcoding, and scientific simulations. If your workload is mostly web serving, typical databases, or single-threaded logic, a CPU instance is usually more cost-effective.
- How much does a cloud GPU cost?
- GPU cost depends on the GPU model (e.g., NVIDIA T4 vs A100 vs H100), the number of GPUs, VM size (CPU/RAM), region, and pricing model (on-demand vs reserved/committed use vs spot/preemptible). GPUs are typically priced per hour (or per second/minute depending on provider) and can range from relatively low-cost inference GPUs to very expensive top-end training GPUs. Also factor in storage, data transfer, and managed service fees if you use a platform like a managed ML service.
Category: hardware
Difficulty: intermediate
Related Terms
See Also