# ThunderGBM (ROCm-Finance)

Use ThunderGBM when parallelism and raw throughput dominate the story:
massively parallel trees, simulation-scale batches, and scenario grids that want the accelerator
to do the heavy lifting today—not in some speculative later hardware generation. ThunderGBM collapses the
signal-to-decision distance, optimized for highly parallel,
GPU-intensive training runs on [AMD Instinct](https://www.amd.com/en/products/accelerators/instinct.html) silicon.

::::{grid} 2 2 2 2

:::{grid-item-card}
:padding: 1
:img-top: ../images/finance-6.png
:class-img-top: small-sd-card-img-top
:class-body: small-sd-card
:class: small-sd-card
+++
<a href="https://rocm.docs.amd.com/projects/thundergbm/en/latest/" class="card-header-link">
  <h2 class="card-header">Documentation</h2>
</a>
<p class="paragraph"> ThunderGBM on ROCm—component documentation is available on the ROCm Documentation site.
</p>
:::

:::{grid-item-card}
:padding: 1
:img-top: ../images/finance-3.png
:class-img-top: small-sd-card-img-top
:class-body: small-sd-card
:class: small-sd-card
+++
<a href="https://github.com/ROCm/ThunderGBM/" class="card-header-link">
  <h2 class="card-header">Github</h2>
</a>
<p class="paragraph"> View the ThunderGBM source code on Github.
</p>
:::

::::