GPU Backends

Morph selects GPU computing backends in priority order with CPU fallback.

---

Backend Priority

Priority	Backend	Requirements
1	CUDA	NVIDIA GPU + CUDA toolkit
2	Vulkan	Any GPU with Vulkan driver
3	CPU	Always available (fallback)

---

Environment Variables

Names match the tensor / GPU runtime probes (e.g. tests/runtime/test_tensor_runtime.cpp), not the branding macros in Branding.generated.h:

Variable	Description
MORPHLANG_GPU_FORCE_BACKEND	Force a specific backend (cuda, vulkan, cpu)
MORPHLANG_GPU_DUMP_METRICS	Print GPU performance timing after each block
MORPHLANG_GPU_READBACK_OPT	Skip unnecessary GPU-to-CPU data transfers

---

Checking GPU Availability

The runtime automatically detects available GPUs. Use MORPHLANG_GPU_DUMP_METRICS=1 to see which backend was selected.

---

Next Steps

• Scope Batching — Batched GPU dispatch