diff --git a/Project.toml b/Project.toml
index b0cd9950..14d29d97 100644
--- a/Project.toml
+++ b/Project.toml
@@ -39,7 +39,7 @@ Adapt = "4"
 BandedMatrices = "1"
 BlockBandedMatrices = "0.13"
 CUDA = "5"
-CUDSS = "0.5, 0.6"
+CUDSS = "0.6.1"
 ChainRules = "1"
 ChainRulesCore = "1"
 ChainRulesTestUtils = "1"
diff --git a/test/gpu/Project.toml b/test/gpu/Project.toml
index 353250f1..1ef4f102 100644
--- a/test/gpu/Project.toml
+++ b/test/gpu/Project.toml
@@ -1,3 +1,4 @@
 [deps]
 CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
-Metal = "dde4c033-4e86-420c-a63e-0dd931031962"
\ No newline at end of file
+CUDSS = "45b445bb-4962-46a0-9369-b4df9d0f772e"
+Metal = "dde4c033-4e86-420c-a63e-0dd931031962"
diff --git a/test/gpu/cuda.jl b/test/gpu/cuda.jl
index f57bdafa..2651a638 100644
--- a/test/gpu/cuda.jl
+++ b/test/gpu/cuda.jl
@@ -1,7 +1,20 @@
-using CUDA
+using CUDSS, CUDA, SparseArrays, LinearAlgebra
+using CUDA.CUSPARSE
 using ArrayInterface
 
 using Test
 
+A_cpu = Float32[1 0; 0 1]
+A_dense = CuMatrix(A_cpu)
+A_sparse = CuSparseMatrixCSR(sparse(A_cpu))
+
 # Test whether lu_instance throws an error when invoked with an gpu array
-@test !isa(try ArrayInterface.lu_instance(CUDA.CuArray([1.f0 1.f0; 1.f0 1.f0])) catch ex ex end, Exception) 
+lu_inst_dense = ArrayInterface.lu_instance(A_dense)
+lu_inst_sparse = ArrayInterface.lu_instance(A_sparse)
+
+# test that lu! is valid when using the inst as scratch
+lu_sparse = lu!(lu_inst_sparse, A_sparse)
+
+#test that the resulting lu works
+b = CuVector([1f0, 1f0])
+@test CUDA.@allowscalar lu_sparse \ b == [1, 1]