PantheonSim / GPUs

Simulated GPUs: 37 NVIDIA and 10 AMD profiles

PantheonSim, the GPU simulator for CUDA and ROCm, carries 47 GPU profiles: 37 NVIDIA and 10 AMD, from the Turing T4 to Blackwell and Rubin, and from Radeon RX 6000 cards to Instinct accelerators. 16 of them were measured on a physical card. Each card has its own page with the numbers a program can ask it, the command to simulate it, and what its verified status means.

How to read the lists

A profile holds what a program can ask a GPU: its compute capability or gfx target, its multiprocessor count, its per-multiprocessor limits and memory, and the values nvidia-smi or rocm-smi show. The profile decides which kernels load. A GPU is named <vendor>/<model>, such as nvidia/h100 or amd/mi300x, and vgpu list-gpus prints every id.

The Measured column says whether the profile was read from a physical card by the simulator's characterization tools. A profile that was not is built from public documents only: some figures are derived or copied from a neighbouring part, and memory sizes are upper bounds. Memory is shown in GiB, as the profile reports it, which can differ from the marketed figure. None of this is about speed, because the simulator has no timing model.

NVIDIA profiles

AMD profiles

Using a profile

Pick a card by id and run a program you built with the real compiler:

vgpu list-gpus
vgpu info --gpu nvidia/h100
vgpu run --gpu nvidia/h100 ./my_cuda_app
vgpu run --gpu amd/mi300x ./my_hip_app

The browser playground offers a selection of these cards; the table on the home page lists them. To run one profile after another and compare, see vgpu test --matrix in the docs. For how to build your program for a profile, read running CUDA without a GPU or running ROCm and HIP without an AMD GPU.

Where this data comes from

The tables and pages here are generated from the profile files in the simulator's repository, as of commit 57bef051dc6c (2026-10-16). Newer commits may add cards or change a value, and the repository is the authority.

Questions