NVIDIA GPU tuning on Linux gets voltage boost and new sensors with LACT
LACT, the Linux GPU configuration and monitoring tool, has received a major update for NVIDIA GPU users. Version 0.10.0 adds new overclocking controls while expanding sensor support for GeForce RTX cards, including Blackwell GPUs and GDDR7 memory.
More NVIDIA controls on Linux
The update introduces NVIDIA PowerMizer control, allowing users to force the GPU into its highest performance state. LACT compares this option with the performance-level control already available for AMD GPUs.
© LACT A new voltage boost slider has also been added. This does not provide a full voltage offset. Instead, it allows the GPU to use slightly higher voltages and provides extra headroom for some overclocking configurations. The developers compare it directly with the voltage slider available in MSI Afterburner on Windows.
Blackwell hotspot and per-chip VRAM temperatures
LACT 0.10.0 also changes how Blackwell hotspot temperatures are read. The software now uses a Blackwell-specific sensor path rather than the previous method that could report incorrect values.
The software can now read temperatures from individual NVIDIA memory chips on GPUs using GDDR6X and GDDR7. This includes separate readings from both sides of the PCB on cards using a clamshell memory configuration. The project credits Paulo Gomes for finding these sensors and sharing the information with the developers.
AMD users also receive a new GTT memory usage indicator. This tracks system memory accessible to the GPU and can be useful with integrated graphics when applications exceed the available dedicated or reserved graphics memory. Other changes include a new service setup interface, improved suspend/resume handling, GPU detach and reattach commands for VM setups and polkit authentication for administrative actions.
LACT supports AMD, NVIDIA and Intel GPUs and provides monitoring, overclocking, fan control and power configuration through a graphical interface. NVIDIA support requires the proprietary NVIDIA driver with CUDA libraries installed.
Source: LACT (Github) , TechPowerUP
