Remove obsolete HIP workaround

Author: sARY77

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -119,13 +119,6 @@ static cudaError_t ggml_cuda_device_malloc(void ** ptr, size_t size, int device)
 }
 
 static ggml_cuda_device_info ggml_cuda_init() {
-#ifdef __HIP_PLATFORM_AMD__
-    // Workaround for a rocBLAS bug when using multiple graphics cards:
-    // https://github.com/ROCmSoftwarePlatform/rocBLAS/issues/1346
-    rocblas_initialize();
-    CUDA_CHECK(cudaDeviceSynchronize());
-#endif
-
     ggml_cuda_device_info info = {};
 
     cudaError_t err = cudaGetDeviceCount(&info.device_count);