diff --git a/artiq/firmware/runtime/session.rs b/artiq/firmware/runtime/session.rs
index 1cb26509e..8b42329ee 100644
--- a/artiq/firmware/runtime/session.rs
+++ b/artiq/firmware/runtime/session.rs
@@ -522,7 +522,7 @@ fn process_kern_queued_rpc(stream: &mut TcpStream,
         let length = NetworkEndian::read_u32(slice) as usize;
         host_write(stream, host::Reply::RpcRequest { async: true })?;
         debug!("{:?}", &slice[4..][..length]);
-        stream.write(&slice[4..][..length])?;
+        stream.write_all(&slice[4..][..length])?;
         Ok(())
     })
 }
diff --git a/artiq/test/coredevice/test_stress.py b/artiq/test/coredevice/test_stress.py
new file mode 100644
index 000000000..a7037f156
--- /dev/null
+++ b/artiq/test/coredevice/test_stress.py
@@ -0,0 +1,29 @@
+import os
+import time
+import unittest
+
+from artiq.experiment import *
+from artiq.test.hardware_testbench import ExperimentCase
+
+
+artiq_low_latency = os.getenv("ARTIQ_LOW_LATENCY")
+
+
+class _Stress(EnvExperiment):
+    def build(self):
+        self.setattr_device("core")
+
+    @rpc(flags={"async"})
+    def sink(self, data):
+        pass
+
+    @kernel
+    def async_rpc(self, n):
+        for _ in range(n):
+            self.sink(b"")
+
+
+class StressTest(ExperimentCase):
+    def test_async_rpc(self):
+        exp = self.create(_Stress)
+        exp.async_rpc(16000)