Better integration with cudaMallocAsync , allowing memory to be scoped more tightly to specific graph nodes to prevent fragmentation. 🛠️ NVCC Compiler and Toolchain Updates
"Here goes nothing," he said.
"It feels like they cleaned the garage," Elias observed. "It's not just new tools; they organized the old ones." He noted the deprecation warnings for some legacy texture reference APIs—finally forcing him to clean up his old code—but the compiler warnings were clear and actionable, not the cryptic nonsense he was used to. cuda 12.6 release notes