diff --git a/ext/CUDAExt.jl b/ext/CUDAExt.jl
index 0dc143b..2e7e99d 100644
--- a/ext/CUDAExt.jl
+++ b/ext/CUDAExt.jl
@@ -36,7 +36,7 @@ function _gcp(
         for n in 1:N
             V = reduce(.*, U[i]'U[i] for i in setdiff(1:N, n))
             U[n] = GCPDecompositions.mttkrp(X, U, n) / V
-            λ = CuArray(CUDA.norm.(eachcol(U[n])))
+            λ = vec(sqrt.(sum(abs2, U[n]; dims=1)))
             U[n] = U[n] ./ permutedims(λ)
         end
     end
diff --git a/src/gcp-opt.jl b/src/gcp-opt.jl
index 1e9ad20..5ac5933 100644
--- a/src/gcp-opt.jl
+++ b/src/gcp-opt.jl
@@ -166,7 +166,7 @@ function _gcp(
         for n in 1:N
             V = reduce(.*, U[i]'U[i] for i in setdiff(1:N, n))
             U[n] = mttkrp(X, U, n) / V
-            λ = norm.(eachcol(U[n]))
+            λ = vec(sqrt.(sum(abs2, U[n]; dims=1)))
             U[n] = U[n] ./ permutedims(λ)
         end
     end