From 12fb1c58ec679f45c8fb44548cdcf16639ff8525 Mon Sep 17 00:00:00 2001
From: Jared Van Bortel <jared@nomic.ai>
Date: Sun, 26 Nov 2023 22:20:18 -0500
Subject: [PATCH] cuda : tweak mm stride to double perf on P40 + GTX 970

---
 ggml-cuda.cu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/ggml-cuda.cu b/ggml-cuda.cu
index 5b80e4ae3..fc82f80d9 100644
--- a/ggml-cuda.cu
+++ b/ggml-cuda.cu
@@ -467,7 +467,7 @@ static_assert(K_QUANTS_PER_ITERATION == 1 || K_QUANTS_PER_ITERATION == 2, "K_QUA
 #define GGML_CUDA_PEER_MAX_BATCH_SIZE 128
 #endif // GGML_CUDA_PEER_MAX_BATCH_SIZE
 
-#define MUL_MAT_SRC1_COL_STRIDE 128
+#define MUL_MAT_SRC1_COL_STRIDE 4096
 
 #define MAX_STREAMS 8
 static cudaStream_t g_cudaStreams[GGML_CUDA_MAX_DEVICES][MAX_STREAMS] = { { nullptr } };