Analysis and design for the naval recovery of a piece of Lake Lario's history.
The steamship Plinio, which entered service on Lake Lario in the early twentieth century, has lain since 2010 on the bed of Lake Mezzola. Its period construction, the depth, the location and its historic value make it impossible to use traditional recovery systems without risking serious damage to the structure.
For this reason we designed a series of dedicated apparatus that make recovery possible while minimizing the chance of the operations failing. They are modular: they can be used to recover any vessel in special conditions.
The Plinio project also lives on through Ingegneri per il Lario APS, a social promotion association that puts technical skills at the service of protecting and enhancing the historic and naval heritage of Lake Lario.
Registered office: Via San Giorgio 8, 22036 Erba (CO)
We develop a lot of software, from local AI tools to automation, and we share part of it openly: because technical knowledge, like the heritage of the lake, grows when it is shared.
Computing for artificial intelligence on AMD Radeon VII GPUs (“Vega” architecture, gfx906), cards that manufacturers no longer support: we put them back to work on Windows, documenting every step.
Marconi, a ROCm kit for Vega GPUs on Windows: compute libraries (rocBLAS, MIOpen, hipBLASLt) and optimized kernels for cards that have lost official support.
Open on GitHub → PyTorch · ROCmPyTorch compiled natively with ROCm on Windows for the Radeon VII: about 5 times faster than going through ZLUDA, GPU linear algebra, flash attention and LoRA training.
Open on GitHub → ZLUDA · RustZLUDA to run PyTorch on AMD GPUs: all 86 cuDNN symbols used by PyTorch translated to MIOpen and verified, from convolutions to recurrent networks.
Open on GitHub → MIOpen · HIPMIOpen on the Radeon VII: kernel compatibility with current drivers and fixes for silent defects that produced wrong gradients without any warning.
Open on GitHub →