The new gpu partition has just been open for production on Sherlock 2.0.

It currently features two kinds of accelerators:

• NVIDIA Tesla P100-PCIe GPUs, for applications requiring double-precision (64-bit) and deep-learning training workloads,
• NVIDIA Tesla P40 GPUs, for single- or half-precision (32, 16-bit) workloads and deep-learning inference jobs.

You'll find more details in the documentation.