From “what is a GPU” to “how to calculate cost per TFLOP” — everything you need to understand GPU cloud infrastructure. Every concept connects back to the GIS specification.
The building blocks. Processors, memory, storage, and why the cloud exists.
Everything about GPUs — architecture, VRAM, TFLOPS, interconnects, form factors.
vCPUs, RAM, storage, networking — the non-GPU parts of a cloud instance.
On-demand, spot, reserved. Cost per GPU hour. The money side of cloud compute.
Regions, availability zones, SLAs, uptime — the operational side.
The GPU cloud landscape. Hyperscalers, specialists, marketplaces.
ML training, inference, fine-tuning, LLMs — what hardware each workload needs.
Understanding the GPU Instance Specification — field by field.
New to GPU cloud? Follow these 10 articles in order. By the end, you'll understand every field in a GIS document.
A Graphics Processing Unit is a chip designed to do thousands of simple calculations at the same time. Originally built for rendering pixels, now the engine behind AI and scientific computing.
CPUs excel at complex sequential tasks. GPUs excel at simple parallel tasks. Understanding the difference is key to choosing the right cloud instance.
Video RAM is the GPU's own dedicated memory. It determines how large a model you can load, how big your batch sizes can be, and ultimately what workloads a GPU can handle.
TFLOPS (Tera Floating-Point Operations Per Second) measures how many trillion math operations a GPU can perform each second. It's the raw compute power metric.
A virtual CPU is a portion of a physical CPU core allocated to your cloud instance. It's how cloud providers measure and sell CPU compute.
A millicore is 1/1000th of a CPU core. It's how Kubernetes measures CPU allocation — 1000m = 1 full vCPU. Essential for container-based GPU workloads.
How GPU cloud pricing works: on-demand, spot, and reserved. Billing granularity, hidden costs, and how to compare prices across providers.
The universal metric for comparing GPU cloud pricing. Normalizes multi-GPU instances, different billing units, and provider-specific pricing into one comparable number.
The GPU Instance Specification is an open, vendor-neutral JSON format for describing any GPU cloud offering. One document per instance. Machine-readable. Comparable.
A hands-on walkthrough of a real GIS document — field by field, section by section. By the end, you'll be able to read any GIS document and understand exactly what it describes.