cuda : fix UMA detection on discrete GPUs. (llama/17537)

diff --git a/src/ggml-cuda/ggml-cuda.cu b/src/ggml-cuda/ggml-cuda.cu
index 0b29074f33d75fbc3b5962dee6d25380c89e7093..aa6570765ad1b002fc5d74f44b8ef6b87ca1d3fa 100644 (file)
--- a/src/ggml-cuda/ggml-cuda.cu
+++ b/src/ggml-cuda/ggml-cuda.cu
@@ -3837,7 +3837,7 @@ static void ggml_backend_cuda_device_get_memory(ggml_backend_dev_t dev, size_t *
 
     // Check if UMA is explicitly enabled via environment variable
     bool uma_env = getenv("GGML_CUDA_ENABLE_UNIFIED_MEMORY") != nullptr;
-    bool is_uma = prop.unifiedAddressing > 0 || uma_env;
+    bool is_uma = prop.integrated > 0 || uma_env;
 
     if (is_uma) {
         // For UMA systems (like DGX Spark), use system memory info