diff --git a/.github/actions/setup/action.yml b/.github/actions/setup/action.yml
index d12a1d16aa..d928b91df6 100644
--- a/.github/actions/setup/action.yml
+++ b/.github/actions/setup/action.yml
@@ -1,3 +1,5 @@
+# Note: this is only compatible with Linux runners.
+
 name: Test setup
 inputs:
   pull_token:
diff --git a/audits/rkm0959.md b/audits/rkm0959.md
index 3fb90318af..2c09f508d5 100644
--- a/audits/rkm0959.md
+++ b/audits/rkm0959.md
@@ -393,7 +393,7 @@ this passes each core verification, and since the RecursionPublicValue of proof
 - shard 1, 2's committed_value_digest = `0`
 - shard 3, 4's committed_value_digest = `x`
 
-this passes each core verification, as proof #2 thinks shard 3 is its "first" shard - so it actually thinks that the `committed_value_digest` didn't change. This means that the whole "no cpu chip means `commited_value_digest` equal" thing actually just passes. Then, in the compress verification, we'll just see the committed_value_digest go from `0` to `x`, which is also completely fine. However, the committed_value_digest will go `0, 0, x, x`, where the change occurs on a shard without cpu chip - which isn't supposed to happen.
+this passes each core verification, as proof #2 thinks shard 3 is its "first" shard - so it actually thinks that the `committed_value_digest` didn't change. This means that the whole "no cpu chip means `committed_value_digest` equal" thing actually just passes. Then, in the compress verification, we'll just see the committed_value_digest go from `0` to `x`, which is also completely fine. However, the committed_value_digest will go `0, 0, x, x`, where the change occurs on a shard without cpu chip - which isn't supposed to happen.
 
 While this is a slight incompatibility, the main invariant (if nonzero, public digest can only be one non-zero value) is preserved. Therefore, we did not fix this observation. 
 
diff --git a/book/generating-proofs/basics.md b/book/generating-proofs/basics.md
index 85273d96ba..7bde99bbc1 100644
--- a/book/generating-proofs/basics.md
+++ b/book/generating-proofs/basics.md
@@ -13,7 +13,8 @@ To make this more concrete, let's walk through a simple example of generating a
 You can run the above script in the `script` directory with `RUST_LOG=info cargo run --release`. Note that running the above script will generate a proof locally.
 
 <div class="warning">
-WARNING: Local proving often is much slower than the prover network and for certain proof types (e.g. Groth16, PLONK) require a significant amount of RAM and will likely not work on a laptop.
+WARNING: Local proving often is much slower than the prover network and for certain proof types (e.g. Groth16, PLONK) require a
+significant amount of RAM. You might only be able to generate proofs for small inputs locally.
 </div>
 
 We recommend using the [prover network](./prover-network.md) to generate proofs. Read more about the [recommended workflow](./recommended-workflow.md) for developing with SP1.
diff --git a/book/getting-started/hardware-requirements.md b/book/getting-started/hardware-requirements.md
index 828ea96080..3877d22e84 100644
--- a/book/getting-started/hardware-requirements.md
+++ b/book/getting-started/hardware-requirements.md
@@ -31,7 +31,9 @@ which can be parallelized with multiple cores.
 Our prover requires keeping large matrices (i.e., traces) in memory to generate the proofs. Certain steps of the prover
 have a minimum memory requirement, meaning that if you have less than this amount of memory, the process will OOM.
 
-This effect is most noticeable when using the Groth16 or PLONK provers.
+This effect is most noticeable when using the Groth16 or PLONK provers. If you're running the Groth16 or Plonk provers locally
+on Mac or Windows using docker, you might need to increase the memory limit for
+[docker desktop](https://docs.docker.com/desktop/settings-and-maintenance/settings/#resources).
 
 ### Disk
 
diff --git a/book/verification/onchain/getting-started.md b/book/verification/onchain/getting-started.md
index 8519d620a8..715b100d58 100644
--- a/book/verification/onchain/getting-started.md
+++ b/book/verification/onchain/getting-started.md
@@ -14,7 +14,8 @@ By default, the proofs generated by SP1 are not verifiable onchain, as they are
 
 > WARNING: The Groth16 and PLONK provers are only guaranteed to work on official releases of SP1. To
 > use Groth16 or PLONK proving & verification locally, ensure that you have Docker installed and have
-> at least 128GB of RAM.
+> at least 32GB of RAM. Note that you might need to increase the memory limit for
+> [docker desktop](https://docs.docker.com/desktop/settings-and-maintenance/settings/#resources) if you're running on Mac.
 
 ### Example
 
diff --git a/crates/recursion/gnark-ffi/src/ffi/docker.rs b/crates/recursion/gnark-ffi/src/ffi/docker.rs
index 119253d492..bf0f2865bd 100644
--- a/crates/recursion/gnark-ffi/src/ffi/docker.rs
+++ b/crates/recursion/gnark-ffi/src/ffi/docker.rs
@@ -47,8 +47,10 @@ fn call_docker(args: &[&str], mounts: &[(&str, &str)]) -> Result<()> {
     }
     cmd.arg(get_docker_image());
     cmd.args(args);
-    if !cmd.status()?.success() {
+    let result = cmd.status()?;
+    if !result.success() {
         log::error!("Failed to run `docker run`: {:?}", cmd);
+        log::error!("Execution result: {:?}", result);
         return Err(anyhow!("docker command failed"));
     }
     Ok(())