World Chip Center
Chip family / GPU

Parallel acceleration

Process many similar operations together for graphics, AI and scientific workloads.

Where to start

AI training and inference, rendering and parallel scientific applications.

What to compare

Memory capacity and bandwidth, supported precision, interconnect, drivers and libraries.

A question worth asking

Test with the model, batch size, precision and software version you expect to use. Peak FLOPS is only one input.

Bring the system into view

Define the task and acceptable response time. Identify the working data size, supported software and operating constraints. Ask for measurements that reflect those conditions.

From understanding to action

Start with a better chip brief.

Turn a workload, a constraint and a question into a useful next step.

Build your brief