diff --git a/.github/scripts/defaults.sh b/.github/scripts/defaults.sh
index 872484b6..738f939f 100755
--- a/.github/scripts/defaults.sh
+++ b/.github/scripts/defaults.sh
@@ -30,7 +30,7 @@ REMOTE_COURSIER_CACHE=$REMOTE_WORK_DIR/.coursier-cache
 declare -A grouping
 grouping["group-cores"]="chipyard-cva6 chipyard-ibex chipyard-rocket chipyard-hetero chipyard-boom chipyard-sodor chipyard-digitaltop chipyard-multiclock-rocket chipyard-nomem-scratchpad chipyard-spike chipyard-clone"
 grouping["group-peripherals"]="chipyard-dmirocket chipyard-dmiboom chipyard-spiflashwrite chipyard-mmios chipyard-nocores chipyard-manyperipherals chipyard-chiplike chipyard-tethered"
-grouping["group-accels"]="chipyard-mempress chipyard-sha3 chipyard-hwacha chipyard-gemmini chipyard-manymmioaccels"
+grouping["group-accels"]="chipyard-mempress chipyard-sha3 chipyard-hwacha chipyard-gemmini chipyard-manymmioaccels chipyard-nvdla"
 grouping["group-constellation"]="chipyard-constellation"
 grouping["group-tracegen"]="tracegen tracegen-boom"
 grouping["group-other"]="icenet testchipip constellation rocketchip-amba rocketchip-tlsimple rocketchip-tlwidth rocketchip-tlxbar"
@@ -44,6 +44,7 @@ mapping["chipyard-sha3"]=" CONFIG=Sha3RocketConfig"
 mapping["chipyard-mempress"]=" CONFIG=MempressRocketConfig"
 mapping["chipyard-digitaltop"]=" TOP=DigitalTop"
 mapping["chipyard-manymmioaccels"]=" CONFIG=ManyMMIOAcceleratorRocketConfig"
+mapping["chipyard-nvdla"]=" CONFIG=SmallNVDLARocketConfig"
 mapping["chipyard-hetero"]=" CONFIG=LargeBoomAndRocketConfig"
 mapping["chipyard-boom"]=" CONFIG=MediumBoomCosimConfig"
 mapping["chipyard-dmiboom"]=" CONFIG=dmiMediumBoomCosimConfig"
diff --git a/.github/scripts/run-tests.sh b/.github/scripts/run-tests.sh
index 89859eb7..20e9ce03 100755
--- a/.github/scripts/run-tests.sh
+++ b/.github/scripts/run-tests.sh
@@ -91,12 +91,13 @@ case $1 in
 	# test streaming-fir
         run_binary BINARY=$LOCAL_CHIPYARD_DIR/tests/streaming-fir.riscv
 
-	# test nvdla
-        run_binary BINARY=$LOCAL_CHIPYARD_DIR/tests/nvdla.riscv
-
 	# test fft
         run_binary BINARY=$LOCAL_CHIPYARD_DIR/tests/fft.riscv
 	;;
+    chipyard-nvdla)
+	make -C $LOCAL_CHIPYARD_DIR/tests
+        run_binary BINARY=$LOCAL_CHIPYARD_DIR/tests/nvdla.riscv
+	;;
     chipyard-manyperipherals)
 	# SPI Flash read tests
         make -C $LOCAL_CHIPYARD_DIR/tests
diff --git a/.github/workflows/chipyard-run-tests.yml b/.github/workflows/chipyard-run-tests.yml
index 72d6f654..11bb2cd2 100644
--- a/.github/workflows/chipyard-run-tests.yml
+++ b/.github/workflows/chipyard-run-tests.yml
@@ -764,6 +764,29 @@ jobs:
           group-key: "group-accels"
           project-key: "chipyard-manymmioaccels"
 
+  chipyard-nvdla-run-tests:
+    name: chipyard-nvdla-run-tests
+    needs: prepare-chipyard-accels
+    runs-on: self-hosted
+    steps:
+      - name: Delete old checkout
+        run: |
+            ls -alh .
+            rm -rf ${{ github.workspace }}/* || true
+            rm -rf ${{ github.workspace }}/.* || true
+            ls -alh .
+      - name: Checkout
+        uses: actions/checkout@v3
+      - name: Git workaround
+        uses: ./.github/actions/git-workaround
+      - name: Create conda env
+        uses: ./.github/actions/create-conda-env
+      - name: Run tests
+        uses: ./.github/actions/run-tests
+        with:
+          group-key: "group-accels"
+          project-key: "chipyard-nvdla"
+
   chipyard-mempress-run-tests:
     name: chipyard-mempress-run-tests
     needs: prepare-chipyard-accels
@@ -1037,6 +1060,7 @@ jobs:
             chipyard-sha3-run-tests,
             chipyard-gemmini-run-tests,
             chipyard-manymmioaccels-run-tests,
+            chipyard-nvdla-run-tests,
             chipyard-mempress-run-tests,
             chipyard-constellation-run-tests,
             tracegen-boom-run-tests,
diff --git a/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala b/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala
index 6a7710d9..16e6b313 100644
--- a/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala
+++ b/generators/chipyard/src/main/scala/config/MMIOAcceleratorConfigs.scala
@@ -61,7 +61,6 @@ class LargeNVDLARocketConfig extends Config(
 class ManyMMIOAcceleratorRocketConfig extends Config(
   new chipyard.iobinders.WithDontTouchIOBinders(false) ++   // TODO: hack around dontTouch not working in SFC
   new fftgenerator.WithFFTGenerator(numPoints=8, width=16, decPt=8) ++ // add 8-point mmio fft at the default addr (0x2400) with 16bit fixed-point numbers.
-  new nvidia.blocks.dla.WithNVDLA("small") ++               // add a small NVDLA
   new chipyard.example.WithStreamingPassthrough ++          // use top with tilelink-controlled streaming passthrough
   new chipyard.example.WithStreamingFIR ++                  // use top with tilelink-controlled streaming FIR
   new freechips.rocketchip.subsystem.WithNBigCores(1) ++
diff --git a/generators/mempress b/generators/mempress
index 295ae085..c3d4901a 160000
--- a/generators/mempress
+++ b/generators/mempress
@@ -1 +1 @@
-Subproject commit 295ae0854a429182e4a38b120c8771a4e898834e
+Subproject commit c3d4901ab0f3c28bfd4101ffd62f707a86fc377e
diff --git a/generators/sha3 b/generators/sha3
index b19b7c76..eb3822a2 160000
--- a/generators/sha3
+++ b/generators/sha3
@@ -1 +1 @@
-Subproject commit b19b7c76fc865598b6a9806c8ef799064fc496ab
+Subproject commit eb3822a2bcc33d5b36456b537681da626ab4b2af
diff --git a/software/firemarshal b/software/firemarshal
index fb93e311..f553a22b 160000
--- a/software/firemarshal
+++ b/software/firemarshal
@@ -1 +1 @@
-Subproject commit fb93e3116fed07191e669291b941e9eabb565ee3
+Subproject commit f553a22b2831a37868e921afa98cf30a6c633a61