Skip to content

⚡ Universal Hardware Support: Cross-Vendor Acceleration

Turing Engine provides universal, cross-vendor hardware execution, automatically discovering and dispatching computations to the highest performance accelerator available on your machine.


🌐 Supported Silicon & Hardware Matrix

Hardware Vendor Supported GPUs & Silicon Primary Compute Backend Acceleration Strategy
NVIDIA RTX 3090 / 4090 / 5090, L4, A100, H100 CUDA + Triton 3.x Custom Tensor Core SwiGLU & Flash-Tree Triton kernels
AMD Radeon RX 7900 XTX / 7900 GRE / 8000, Instinct MI250X / MI300X ROCm (HIP) + Triton Wave32 (RDNA) / Wave64 (CDNA) Matrix Core heuristics
Intel Intel Arc A770 / A750 / B580 Battlemage, Data Center GPU Max Intel XPU (SYCL / OneAPI) Intel XMX Matrix Engines + IPEX bindings
Apple M1 / M2 / M3 / M4 (Pro, Max, Ultra) Metal (MPS) Metal Performance Shaders + Vectorized Subspace Slicing
Cross-Vendor Intel, AMD APUs, Qualcomm Snapdragon Adreno, ARM Mali Vulkan Compute SPIR-V Compute Shaders + Host-Visible Coherent Memory
x86_64 / ARM CPU Intel Xeon, AMD EPYC, Ampere Altra, Apple Silicon CPU C++20 AVX2 / NEON SIMD 64-byte aligned SIMD fused FMA + zero-copy mmap

🔍 Dynamic Hardware Auto-Discovery

Turing Engine automatically detects your system's hardware configuration at startup:

from turing.kernels.dispatch import get_hardware_backend_info

info = get_hardware_backend_info()
print(f"Backend: {info['backend']} ({info['vendor']}) on {info['device_name']}")

You can also run the hardware probe directly from your terminal:

python scripts/test_universal_backends.py


🛠️ Ecosystem Setup Guides

1. AMD ROCm / HIP Setup (Linux)

On systems with AMD Radeon or Instinct GPUs:

# Install PyTorch with ROCm support:
pip install torch --index-url https://download.pytorch.org/whl/rocm6.2

# Install Turing Engine:
pip install turing-engine

2. Intel Arc / XPU Setup (Linux & Windows)

On systems with Intel Arc GPUs:

# Install PyTorch with Intel XPU / SYCL support:
pip install torch intel-extension-for-pytorch

# Launch with XPU acceleration:
turing chat --model deepseek-r1-1.5b --device xpu

3. Apple Silicon Metal (macOS)

On any Mac with M1/M2/M3/M4 chip:

# Direct pip install (pre-compiled native ARM64 wheel):
pip install turing-engine

# Launch instant terminal chat:
turing chat --model smollm2

4. Cross-Vendor Vulkan & CPU SIMD Fallback

On systems without proprietary vendor drivers or headless servers:

# Automatic fallback to C++20 AVX2 SIMD:
turing chat --model deepseek-r1-1.5b --device cpu