diff --git a/drivers/gpu/nvgpu/common/fifo/submit.c b/drivers/gpu/nvgpu/common/fifo/submit.c
index 09885128a..cc50ed741 100644
--- a/drivers/gpu/nvgpu/common/fifo/submit.c
+++ b/drivers/gpu/nvgpu/common/fifo/submit.c
@@ -460,6 +460,8 @@ static int nvgpu_submit_channel_gpfifo(struct nvgpu_channel *c,
 		 *   allocated, not pooled)
 		 * - dependency on sync framework for post fences
 		 * - buffer refcounting, which is O(n)
+		 * - aggressive sync destroy is enabled (sync objects are not
+		 *   held during channel lifetime but refcounted for submits)
 		 * - channel wdt, which needs periodic async cleanup
 		 *
 		 * If none of the conditions are met, then deferred clean-up
@@ -469,7 +471,8 @@ static int nvgpu_submit_channel_gpfifo(struct nvgpu_channel *c,
 		need_deferred_cleanup = !nvgpu_channel_is_deterministic(c) ||
 					!nvgpu_has_syncpoints(g) ||
 					(flag_sync_fence && flag_fence_get) ||
-					!skip_buffer_refcounting;
+					!skip_buffer_refcounting ||
+					g->aggressive_sync_destroy_thresh != 0U;
 
 #ifdef CONFIG_NVGPU_CHANNEL_WDT
 		need_deferred_cleanup = need_deferred_cleanup || c->wdt.enabled;
diff --git a/drivers/gpu/nvgpu/common/init/nvgpu_init.c b/drivers/gpu/nvgpu/common/init/nvgpu_init.c
index 62c4c773c..1c6dd731a 100644
--- a/drivers/gpu/nvgpu/common/init/nvgpu_init.c
+++ b/drivers/gpu/nvgpu/common/init/nvgpu_init.c
@@ -767,18 +767,25 @@ int nvgpu_init_gpu_characteristics(struct gk20a *g)
 			true);
 
 	/*
+	 * Full deterministic submit means that synchronization (pre and post
+	 * fences; implies job tracking) can be used. If such submits can be
+	 * guaranteed as long as the channel is set up correctly by userspace
+	 * (e.g., watchdog disabled), this bit is set.
+	 *
 	 * Sync framework is needed when we don't have syncpoint support
 	 * because we don't have a means to expose raw gpu semas in a way
-	 * similar to raw syncpts. Use of the framework requires heavy stuff
-	 * like deferred job cleanup and wrapping syncs in FDs which prevents
-	 * deterministic submits. This is supported otherwise, provided that
-	 * the user doesn't request anything that depends on deferred cleanup.
+	 * similar to raw syncpts. Use of the framework requires unpredictable
+	 * actions including deferred job cleanup and wrapping syncs in FDs.
+	 *
+	 * Aggressive sync destroy causes the channel syncpoint to be abruptly
+	 * allocated and deleted during submit path and deferred cleanup.
 	 *
 	 * Note that userspace expects this to be set for usermode submits
 	 * (even if kernel-mode submits aren't enabled where full deterministic
 	 * features matter).
 	 */
-	if (nvgpu_has_syncpoints(g)) {
+	if (nvgpu_has_syncpoints(g) &&
+			g->aggressive_sync_destroy_thresh == 0U) {
 		nvgpu_set_enabled(g,
 				NVGPU_SUPPORT_DETERMINISTIC_SUBMIT_FULL,
 				true);
diff --git a/drivers/gpu/nvgpu/os/linux/module.c b/drivers/gpu/nvgpu/os/linux/module.c
index 4b70f0460..3a70684a7 100644
--- a/drivers/gpu/nvgpu/os/linux/module.c
+++ b/drivers/gpu/nvgpu/os/linux/module.c
@@ -723,7 +723,7 @@ static struct of_device_id tegra_gk20a_of_match[] = {
 		.data = &gv11b_vgpu_tegra_platform},
 #endif
 #ifdef CONFIG_NVGPU_GR_VIRTUALIZATION
-	{ .compatible = "nvidia,tegra124-gk20a-vgpu",
+	{ .compatible = "nvidia,tegra186-gp10b-vgpu",
 		.data = &vgpu_tegra_platform },
 #endif
 #if defined(CONFIG_NVGPU_NEXT) && defined(CONFIG_NVGPU_NON_FUSA)
diff --git a/drivers/gpu/nvgpu/os/linux/platform_gk20a.h b/drivers/gpu/nvgpu/os/linux/platform_gk20a.h
index 62d61e784..cc1532483 100644
--- a/drivers/gpu/nvgpu/os/linux/platform_gk20a.h
+++ b/drivers/gpu/nvgpu/os/linux/platform_gk20a.h
@@ -61,7 +61,7 @@ enum tegra_chip_id {
 	TEGRA_186,
 	TEGRA_194,
 	TEGRA_194_VGPU,
-	TEGRA_124_VGPU,
+	TEGRA_186_VGPU,
 	TEGRA_234,
 };
 
diff --git a/drivers/gpu/nvgpu/os/linux/vgpu/platform_vgpu_tegra.c b/drivers/gpu/nvgpu/os/linux/vgpu/platform_vgpu_tegra.c
index 765f6fefa..6066fef6b 100644
--- a/drivers/gpu/nvgpu/os/linux/vgpu/platform_vgpu_tegra.c
+++ b/drivers/gpu/nvgpu/os/linux/vgpu/platform_vgpu_tegra.c
@@ -67,8 +67,6 @@ struct gk20a_platform vgpu_tegra_platform = {
 #ifdef CONFIG_TEGRA_GK20A_NVHOST
 	.has_syncpoints = true,
 #endif
-	.aggressive_sync_destroy_thresh = 64,
-
 	/* power management configuration */
 	.can_railgate_init	= false,
 	.can_elpg_init          = false,
@@ -88,7 +86,7 @@ struct gk20a_platform vgpu_tegra_platform = {
 	.clk_round_rate = vgpu_plat_clk_round_rate,
 	.get_clk_freqs = vgpu_plat_clk_get_freqs,
 
-	.platform_chip_id = TEGRA_124_VGPU,
+	.platform_chip_id = TEGRA_186_VGPU,
 
 	/* frequency scaling configuration */
 	.devfreq_governor = "userspace",