ZLUDA 2025: The Multi-Vendor CUDA Solution Revolutionizing AI Workloads on Non-NVIDIA GPUs

ZLUDA 2025 transforms non-NVIDIA GPUs into CUDA-compatible powerhouses for AI workloads. Discover how this open-source solution achieves 90%+ CUDA performance on AMD/Intel GPUs, with new Q2 features like llm.c support and automated builds. Explore benchmarks, use cases, and installation guide.

AMD Engineer Implements GPU Memory Allocation via LLVM Libc – Boosting High-Performance Computing

 

AMD engineer Joseph Huber upstreams GPU malloc support in LLVM libc, enabling dynamic memory allocation for C++ & Fortran on ROCm. Boosts AI, HPC & gaming performance. Coming in LLVM 21.