| |
Nvidia is extending CUDA support to RISC-V processors, opening the architecture for GPU compute applications including machine learning. To make this work, Nvidia requires RISC-V CPUs to meet server-grade specifications including RVA23 compliance, vector extensions with predication support, ACPI support (now available through the UEFI forum's 2025 addition), PCIe coherency guarantees, and peer-to-peer PCIe communication capabilities. These requirements ensure CUDA can run efficiently without compromising performance through lowest-common-denominator software implementations.
Read Full Article →
← More Tech news