Update app.py

app.py

@@ -161,10 +161,17 @@ acts = round(2*bs*l*(d/h)*2*n/1e9, 2)
 c2.write(str(acts))
 
 st.subheader("Approximations")
-st.markdown("We estimate the time for each matrix multiplication")
+st.markdown("[We use the [following crude approximation](https://docs.nvidia.com/deeplearning/performance/dl-performance-gpu-background/index.html#understand-perf) to estimate the execution time for each matrix multiplication.")
 
-st.latex("C = A \dot B, A \in R^{NxM}")
+st.latex("C = A \cdot B")
+st.latex("A \in \mathbb{R}^{MxK}, B \in R^{KxN}, C \in \mathbb{R}^{MxN}")
 
+st.markdown('''
+To execute this operation on the GPU, we need to 
+- Read A, B from memory
+- Perform math operations
+- Write C to memory
+''')
 
 
 breakdown = st.checkbox("Show breakdown per operation")