From 537a4bcedfdb8df63459880cc09e84c7c8be70d1 Mon Sep 17 00:00:00 2001
From: Duane Merrill <duane.merrill@gmail.com>
Date: Tue, 5 Dec 2017 22:54:49 -0500
Subject: [PATCH] Update README.md

---
 README.md | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/README.md b/README.md
index 4e5f38cc..11fe5a1c 100644
--- a/README.md
+++ b/README.md
@@ -27,7 +27,8 @@ in CUDA C++"](https://devblogs.nvidia.com/parallelforall/cutlass-linear-algebra-
 
 ![ALT](/media/cutlass-performance-plot.png "Relative performance of CUTLASS and cuBLAS for large matrices")
 
-CUTLASS is very efficient, with performance comparable to cuBLAS for scalar GEMM 
+CUTLASS primitives are very efficient.  When used to construct device-wide GEMM kernels, 
+they exhibit performance comparable to cuBLAS for scalar GEMM 
 computations. The above figure shows CUTLASS performance relative to cuBLAS 
 for large matrix dimensions (M=10240, N=K=4096) running on an NVIDIA Tesla V100 GPU 
 when compiled with CUDA 9.0.