C++20 GPU compute runtime with CUDA-like ergonomics.
-
Updated
Jul 24, 2026 - C++
#
cuda-alternative
Here are 11 public repositories matching this topic...
CUDA-equivalent tensor-core acceleration for Apple Silicon. C-ABI kernel library wrapping simdgroup_matrix (M1+) and mpp::tensor_ops (M5+): GEMM, FlashAttention, Conv2D, Q4_0/Q8_0 quantized inference, GGUF reader, full transformer training kernels. One binary, M1 → M5.
metal objective-c-plus-plus gemm mlx mixed-precision apple-silicon llm-inference flash-attention gguf cuda-alternative simdgroup-matrix tensor-ops flashattention2
-
Updated
Aug 30, 2026 - Python
Environment diagnostics, compatibility resolver, and a live-tracked map of the Python AI ecosystem on AMD ROCm.
-
Updated
Sep 5, 2026 - Python
Native Apple Silicon Metal port of mumax3, extending CUDA based micromagnetic simulation to macOS with MPSGraph FFT and validated physics compatibility.
macos golang metal simulation scientific-computing computational-physics gpu-acceleration gpu-computing magnetism micromagnetics metal-performance-shaders spintronics skyrmion mumax3 apple-silicon mpsgraph nanomagnetism cuda-alternative micromagnetic-simulation domain-wall
-
Updated
Jul 31, 2026 - Go
Native C++/HIP Qwen3.8-27B inference engine for 4x AMD MI50 (gfx906), with chained MTP speculative decoding.
cpp amd gpu transformers cpp17 hip quantization rocm int8 inference-engine amd-gpu tensor-parallelism llm llm-inference qwen speculative-decoding qwen3 gfx906 mi50 cuda-alternative
-
Updated
Oct 7, 2026 - HIP
Système d'exploitation Exokernel "Bare-Metal" et langage dédié (Neuro-Lang) pour l'IA. Élimination de la "Taxe d'Abstraction" : exécution Ring-0, mémoire unifiée SASOS et accès GPU direct sans latenc
-
Updated
Jan 6, 2026
Run a local LLM on a Mac with an NVIDIA eGPU: custom DriverKit driver (no CUDA runtime), RTX 3090 over Thunderbolt — Qwen3.8-27B at 75.81 tok/s decode @100k ctx, bit-exact, OpenAI-compatible API.
macos inference nvidia thunderbolt egpu driverkit apple-silicon openai-api tinygrad llm local-llm qwen speculative-decoding rtx-3090 cuda-alternative
-
Updated
Oct 5, 2026 - Python
Full local AI stack on AMD ROCm — no NVIDIA, no CUDA. Self-hosted LLM chat, image generation, web search & agent tools on one GPU.
amd mcp self-hosted homelab rocm radeon amd-gpu podman llm stable-diffusion llamacpp llama-cpp comfyui local-llm private-gpt local-ai open-webui ai-stack rdna4 cuda-alternative
-
Updated
Oct 15, 2026 - Python
GPU-accelerated Sobel edge detection using OpenCL. Features a high-performance "Universal Tiling" implementation with local memory (SRAM) caching and strided loading, achieving a 138x speedup on NVIDIA K20 hardware.
computer-vision hpc opencl university-project cpp17 gpu-computing performance-optimization heterogeneous-computing parrallel-computing sobel-filter cuda-alternative
-
Updated
Apr 17, 2026 - C++
From-scratch LLM inference engine for Apple Silicon: hand-written Metal kernels (simdgroup_matrix GEMM, FlashAttention prefill, paged-attention decode), paged KV cache, continuous batching, OpenAI-compatible server. C++17, no PyTorch/llama.cpp/MLX.
-
Updated
Aug 5, 2026 - C++
An open, vendor-neutral compute language — CUDA's grid-of-threads model without the lock-in. One language for AI, hashing and graphics; targets Apple Metal today, FPGA/custom silicon next.
programming-language open-source fpga metal compiler gpu rasterizer gpgpu sha256 compute apple-silicon cuda-alternative
-
Updated
Sep 13, 2026 - Python
Add this topic to your repo
To associate your repository with the cuda-alternative topic, visit your repo's landing page and select "manage topics."