From 645b60eeb138ba6616ae8f080f0a7dc2ad768ce8 Mon Sep 17 00:00:00 2001
From: Oliver Simons <osimons@nvidia.com>
Date: Fri, 16 Oct 2026 19:01:46 +0200
Subject: [PATCH] Guard __nv_cvt_fp8x2_to_bf162raw on CTK 13.4

Only cvt.rn.bf16x2.e4m3x2 is exposed from 13.2 onwards, so
93e98b9203dfacb85f9ac665f5f0482a8c6b4192 was incorrect fix
---
 ggml/src/ggml-cuda/mmvf.cu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ggml/src/ggml-cuda/mmvf.cu b/ggml/src/ggml-cuda/mmvf.cu
index 21e2dec43f..c708f9f79a 100644
--- a/ggml/src/ggml-cuda/mmvf.cu
+++ b/ggml/src/ggml-cuda/mmvf.cu
@@ -10,7 +10,7 @@ using mmvf_y_t = std::conditional_t<std::is_same_v<T, ggml_fp8_e4m3_t>, nv_bfloa
 
 #if defined(FP8_AVAILABLE) && !defined(GGML_USE_HIP) && !defined(GGML_USE_MUSA)
 static __device__ __forceinline__ nv_bfloat162 mmvf_f8x2_e4m3_to_bf162(__nv_fp8x2_storage_t x) {
-#if CUDART_VERSION >= 13020
+#if CUDART_VERSION >= 13040
     return static_cast<nv_bfloat162>(__nv_cvt_fp8x2_to_bf162raw(x, __NV_E4M3));
 #else
     return make_bfloat162(