NVIDIA updated its R520 Linux kernel module to 520.61.05 early this morning which add support for its Hopper PCIe GPUs. Readers may remember that the NVIDIA H100 GPU will be used for artificial intelligence data centers.

NVIDIA's R520 Linux driver begins support for NVIDIA H100 AI data center graphics cards before the release of RTX 40 series GPUs

CUDA allows for support of the two new GPU architectures to see use in the NVIDIA Jetson systems. The new 520.61.05 update to the R520 Linux kernel module comes a week after the enablement of NVIDIA Ada Lovelace and Hopper in the CUDA 11.8 release. The difference with the most recent 520.61.05 update is that the driver will now be a standalone driver. It is not currently found on NVIDIA's support page but will probably arrive after Ada Lovelace.

Ada Lovelace is the consumer-level GPU to release tomorrow and will fall in the RTX 40 series graphics cards. Hopper is the next-gen AI GPU to aid in developing advanced technology, sciences, and breakthroughs that the company has planned out for this decade. In its recent online presentation, the company laid out all of the advancements it focuses on and new technology introduced in the following year. There will be two board configurations for NVIDIA Hopper — one consisting of an SXM5 board form factor and one that will be a PCIe 5.0 board form factor.

The Hopper chipset inside the new GPU will be the first high-speed 4nm data center chip to produce up to 4000 TFLOPs of computing performance and offer HBM3 3 TB/s memory onboard. The TDP between the two boards will be between 600W to 700W power consumption.

Additional fixes were completed for the updated NVIDIA version 520.61.05 Linux driver, even if they are minor. The Makefile was fixed, the nvLogBase2 was enhanced for better efficiency, and now the nv-pci: fixed will show the true expression by default. There is anticipation that the driver will also offer support for updated Vulkan drivers, but at this time, they are only adding the Hopper support.

News Sources: Phoronix, NVIDIA GitHub page