diff --git a/benchmark/README.md b/benchmark/README.md
new file mode 100644
index 000000000..989a1865f
--- /dev/null
+++ b/benchmark/README.md
@@ -0,0 +1,53 @@
+# Benchmarking Suite
+
+The following benchmarking suite uses `PkgBenchmark.jl` and `BenchmarkTools.jl` to compare the performance of different branches of Gridap. 
+
+## Running the benchmarks
+
+### Running as CI job
+
+The benchmarks are setup as a manual Github Actions workflow in `.github/workflows/benchmark.yml`. To run the workflow, you will need administrator access to the Gridap repository, then follow instructions [here](https://docs.github.com/en/actions/managing-workflow-runs-and-deployments/managing-workflow-runs/manually-running-a-workflow).
+
+The workflow will has two inputs: `target` and `base`, which are the branches/tags/commits you want to compare. The workflow will run the benchmarks on the `target` branch and compare them with the `base` branch (`master` by default).
+
+### Running Locally
+
+To run the benchmarks locally, you can have a look at the [documentation for `PkgBenchmark.jl`](https://juliaci.github.io/PkgBenchmark.jl/stable/run_benchmarks/).
+
+Alternatively, you can run the CI script locally from a local copy of the repository. From the Gridap root directory, run the following commands:
+
+```bash
+# Instantiate Gridap and Gridap/benchmark
+julia -e 'using Pkg; Pkg.develop(PackageSpec(path=pwd())); Pkg.instantiate()'
+julia --project=benchmark/ -e 'using Pkg; Pkg.develop(PackageSpec(path=pwd())); Pkg.instantiate()'
+# Run the benchmarks
+export BM_TARGET = "your target branch"
+export BM_BASE = "your reference branch"
+julia --project=benchmark/ --color=yes benchmark/run_benchmarks.jl
+```
+
+where `BM_TARGET` and `BM_BASE` are the branches/tags/commits you want to compare.
+
+## Adding a new benchmark
+
+To add a new benchmark suite `xyx`, create a new file `bm/bm_xyx.jl` that with the following structure:
+
+```julia
+module bm_xyx
+
+using PkgBenchmark, BenchmarkTools
+
+const SUITE = BenchmarkGroup()
+
+[... Add your benchmarks here ...]
+
+end # module
+```
+
+Then, add the following line to the `benchmarks.jl` file:
+
+```julia
+@include_bm "bm_xyz"
+```
+
+This should automatically include the new benchmarks into the global benchmarking suite.
diff --git a/benchmark/benchmarks.jl b/benchmark/benchmarks.jl
index e14106e0a..ad8aa0aa6 100644
--- a/benchmark/benchmarks.jl
+++ b/benchmark/benchmarks.jl
@@ -1,29 +1,14 @@
 using BenchmarkTools
+using PkgBenchmark
 using Gridap
 
-include("drivers.jl")
+macro include_bm(SUITE,name)
+  quote
+    include("bm/$($name).jl")
+    SUITE["$($name)"] = $(Symbol(name)).SUITE
+  end
+end
 
 const SUITE = BenchmarkGroup()
 
-for (D,ncells) in [(2,20),(3,8)]
-  for order in [1,2,3]
-    basis_cases = [
-      ("lagrangian",lagrangian,Float64,order),
-      ("vector_lagragian",lagrangian,VectorValue{D,Float64},order),
-      ("raviart_thomas",raviart_thomas,Float64,order-1),
-    ]
-    for (basis_name,basis,T,degree) in basis_cases
-      biform_cases = [
-        ("mass",mass,2*order),
-        ("laplacian",laplacian,2*(order-1)),
-      ]
-      for (biform_name,biform,qdegree) in biform_cases
-        reffe = ReferenceFE(basis, T, degree)
-        name = "assembly_$(D)D_$(basis_name)_$(biform_name)_$(order)"
-        SUITE[name] = @benchmarkable bm_matrix_assembly(
-          $(D),$(ncells),$(reffe),$(qdegree),$(biform)
-        )
-      end
-    end
-  end
-end
+@include_bm SUITE "bm_assembly"
diff --git a/benchmark/bm/bm_assembly.jl b/benchmark/bm/bm_assembly.jl
new file mode 100644
index 000000000..80b378163
--- /dev/null
+++ b/benchmark/bm/bm_assembly.jl
@@ -0,0 +1,59 @@
+module bm_assembly
+
+using PkgBenchmark, BenchmarkTools
+using Gridap
+using Gridap.Geometry
+
+mass(u,v,dΩ) = ∫(u⋅v)dΩ
+laplacian(u,v,dΩ) = ∫(∇(u)⊙∇(v))dΩ
+graddiv(u,v,dΩ) = ∫((∇⋅u)⋅(∇⋅v))dΩ
+
+function driver(
+  Ω       :: Triangulation,
+  reffe   :: Tuple,
+  qdegree :: Integer,
+  biform  :: Function
+)
+  model = get_background_model(Ω)
+  dΩ = Measure(Ω,qdegree)
+  V = TestFESpace(model, reffe)
+  a(u,v) = biform(u,v,dΩ)
+  A = assemble_matrix(a, V, V)
+end
+
+const SUITE = BenchmarkGroup()
+
+for (D,n) in [(2,10),(3,6)]
+  domain = Tuple([repeat([0,1], D)...])
+  partition = Tuple(fill(n, D))
+  model = UnstructuredDiscreteModel(CartesianDiscreteModel(domain, partition))
+  trian_cases = [
+    ("bulk",Triangulation(model)),
+    ("view",Triangulation(model,collect(1:div(n^D,2)))), # About half of the cells
+  ]
+  for (trian_name, trian) in trian_cases
+    for order in [1,2,3]
+      basis_cases = [
+        ("lagrangian",lagrangian,Float64,order),
+        ("vector_lagragian",lagrangian,VectorValue{D,Float64},order),
+        ("raviart_thomas",raviart_thomas,Float64,order-1),
+      ]
+      for (basis_name,basis,T,degree) in basis_cases
+        biform_cases = [
+          ("mass",mass,2*order),
+          ("laplacian",laplacian,2*(order-1)),
+          ("graddiv",graddiv,2*(order-1)),
+        ]
+        for (biform_name,biform,qdegree) in biform_cases
+          reffe = ReferenceFE(basis, T, degree)
+          name = "assembly_$(D)D_$(trian_name)_$(basis_name)_$(biform_name)_$(order)"
+          SUITE[name] = @benchmarkable driver(
+            $(trian),$(reffe),$(qdegree),$(biform)
+          )
+        end
+      end
+    end
+  end
+end
+
+end # module
\ No newline at end of file
diff --git a/benchmark/drivers.jl b/benchmark/drivers.jl
deleted file mode 100644
index 6ac5bf6a5..000000000
--- a/benchmark/drivers.jl
+++ /dev/null
@@ -1,22 +0,0 @@
-
-using Gridap.Geometry
-
-mass(u,v,dΩ) = ∫(u⋅v)dΩ
-laplacian(u,v,dΩ) = ∫(∇(u)⊙∇(v))dΩ
-
-function bm_matrix_assembly(
-  D :: Integer,
-  n :: Integer,
-  reffe :: Tuple,
-  qdegree :: Integer,
-  biform :: Function
-)
-  domain = Tuple([repeat([0,1], D)...])
-  partition = Tuple(fill(n, D))
-  model = UnstructuredDiscreteModel(CartesianDiscreteModel(domain, partition))
-  Ω = Triangulation(model)
-  dΩ = Measure(Ω,qdegree)
-  V = TestFESpace(model, reffe)
-  a(u,v) = biform(u,v,dΩ)
-  A = assemble_matrix(a, V, V)
-end