diff --git a/ci/ci_common/bench-common.libsonnet b/ci/ci_common/bench-common.libsonnet
index d5af90b929ad..27014ae5d71e 100644
--- a/ci/ci_common/bench-common.libsonnet
+++ b/ci/ci_common/bench-common.libsonnet
@@ -34,7 +34,7 @@
   # max number of threads to use for benchmarking in general
   # the goal being to limit parallelism on very large servers which may not be respresentative of real-world scenarios
   bench_max_threads:: {
-   restrict_threads:: 36
+   restrict_threads:: if self.arch == 'amd64' then 36 else 32,
   },
 
   bench_no_thread_cap:: {
@@ -99,7 +99,27 @@
       default_numa_node:: 0,
       num_threads:: 160,
       hyperthreading:: false
-    }
+    },
+    hr350a:: common.linux_aarch64 + self._bench_machine + {
+      machine_name:: "hr350a",
+      capabilities+: ["tmpfs25g"],
+      numa_nodes:: [0],
+      default_numa_node:: 0,
+      num_threads:: 32,
+      hyperthreading:: false
+    },
+    osprey:: common.linux_aarch64 + self._bench_machine + {
+      machine_name:: "osprey",
+      capabilities+: ["tmpfs25g"],
+      numa_nodes:: [0],
+      default_numa_node:: 0,
+      num_threads:: 32,
+      hyperthreading:: false
+    },
+    local contains(a_str, substr) = std.length(std.findSubstr(substr, a_str)) > 0,
+    # Useful to distribute the benchmark load between two machine types, but always stay consistent for the same benchmark suite
+    # This guarantees comparability of results, whatever the platform/context, for each suite
+    hr350a_or_osprey(suite=null):: if suite != null && std.count([contains(s, suite) for s in ["barista", "renaissance"]], true) > 0 then self.osprey else self.hr350a,
   },
 
   hwloc_cmd(cmd, num_threads, node, hyperthreading, max_threads_per_node)::
diff --git a/compiler/ci/ci_common/benchmark-builders.jsonnet b/compiler/ci/ci_common/benchmark-builders.jsonnet
index d0bd3761d8bf..da0a1190d85b 100644
--- a/compiler/ci/ci_common/benchmark-builders.jsonnet
+++ b/compiler/ci/ci_common/benchmark-builders.jsonnet
@@ -45,17 +45,17 @@
   ])),
 
   local weekly_aarch64_forks_builds = std.flattenArrays([
-    bc.generate_fork_builds(c.weekly + hw.a12c + jdk + cc.libgraal + suite, subdir='compiler')
+    bc.generate_fork_builds(c.weekly + hw.hr350a_or_osprey(suite.suite) + jdk + cc.libgraal + suite, subdir='compiler')
   for jdk in cc.product_jdks
   for suite in bench.groups.weekly_forks_suites
   ]),
 
   local aarch64_builds = [
-    c.daily + hw.a12c + jdk + cc.libgraal + suite,
+    c.daily + hw.hr350a_or_osprey(suite.suite) + jdk + cc.libgraal + suite,
   for jdk in cc.product_jdks
   for suite in bench.groups.main_suites
   ] + [
-    c.monthly + hw.a12c + jdk + cc.libgraal + bench.specjbb2015,
+    c.monthly + hw.hr350a_or_osprey(bench.specjbb2015.suite) + jdk + cc.libgraal + bench.specjbb2015,
   for jdk in cc.product_jdks
   ],
 
diff --git a/compiler/ci/ci_common/benchmark-suites.libsonnet b/compiler/ci/ci_common/benchmark-suites.libsonnet
index 043984244c31..df5b10a17753 100644
--- a/compiler/ci/ci_common/benchmark-suites.libsonnet
+++ b/compiler/ci/ci_common/benchmark-suites.libsonnet
@@ -29,7 +29,7 @@
     run+: [
       self.benchmark_cmd + [self.suite + ":*", "--"] + self.extra_vm_args
     ],
-    timelimit: "30:00",
+    timelimit: "1:00:00",
     forks_batches:: null,
     forks_timelimit:: null,
     min_jdk_version:: 8,
diff --git a/compiler/ci/ci_includes/baseline-benchmarks.jsonnet b/compiler/ci/ci_includes/baseline-benchmarks.jsonnet
index 81782dc7cfc1..679b161eae1b 100644
--- a/compiler/ci/ci_includes/baseline-benchmarks.jsonnet
+++ b/compiler/ci/ci_includes/baseline-benchmarks.jsonnet
@@ -9,7 +9,7 @@
   local hotspot_builds = std.flattenArrays([
     [
       c.weekly + hw.x52 + jdk + cc.c2 + suite,
-      c.weekly + hw.a12c + jdk + cc.c2 + suite
+      c.weekly + hw.hr350a_or_osprey(suite.suite) + jdk + cc.c2 + suite
     ]
   for jdk in cc.product_jdks
   for suite in bench.groups.all_suites
@@ -18,9 +18,9 @@
   local hotspot_profiling_builds = std.flattenArrays([
     [
     c.on_demand + hw.x52  + jdk + cc.c2 + suite + cc.enable_profiling    + { job_prefix:: "bench-compiler-profiling" },
-    c.on_demand + hw.a12c + jdk + cc.c2 + suite + cc.enable_profiling    + { job_prefix:: "bench-compiler-profiling" },
+    c.on_demand + hw.hr350a_or_osprey(suite.suite) + jdk + cc.c2 + suite + cc.enable_profiling    + { job_prefix:: "bench-compiler-profiling" },
     c.on_demand + hw.x52  + jdk + cc.c2 + suite + cc.footprint_tracking  + { job_prefix:: "bench-compiler-footprint" },
-    c.on_demand + hw.a12c + jdk + cc.c2 + suite + cc.footprint_tracking  + { job_prefix:: "bench-compiler-footprint" },
+    c.on_demand + hw.hr350a_or_osprey(suite.suite) + jdk + cc.c2 + suite + cc.footprint_tracking  + { job_prefix:: "bench-compiler-footprint" },
     c.monthly + hw.x52_root + jdk + cc.c2 + suite + cc.energy_tracking   + { job_prefix:: "bench-compiler-energy" }
     ]
   for jdk in cc.product_jdks
@@ -37,7 +37,7 @@
   ])),
 
   local weekly_forks_aarch64_builds = std.flattenArrays([
-    bc.generate_fork_builds(c.weekly + hw.a12c + jdk + cc.c2 + suite)
+    bc.generate_fork_builds(c.weekly + hw.hr350a_or_osprey(suite.suite) + jdk + cc.c2 + suite)
   for jdk in cc.product_jdks
   for suite in bench.groups.weekly_forks_suites
   ]),