diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 6b6c7091e7..f3b27b0e7a 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -27,7 +27,7 @@ env:
   BUILDTOOLS_VERSION: '3.0.0'
   BUILDIFIER_SHA256SUM: 'e92a6793c7134c5431c58fbc34700664f101e5c9b1c1fcd93b97978e8b7f88db'
   BUILDOZER_SHA256SUM: '3d58a0b6972e4535718cdd6c12778170ea7382de7c75bc3728f5719437ffb84d'
-  TENSORFLOW_VERSION: 'tf-nightly'
+  TENSORFLOW_VERSION: 'tf-nightly==2.12.0.dev20230201'
 
 jobs:
   build:
@@ -115,19 +115,6 @@ jobs:
           bazel test //tensorboard/summary/writer:event_file_writer_s3_test &&
           bazel test //tensorboard/compat/tensorflow_stub:gfile_fsspec_test &&
           bazel test //tensorboard/summary/writer:event_file_writer_fsspec_test
-      - name: 'Bazel: build Pip package (with TensorFlow only)'
-        if: matrix.tf_version_id == 'tf'
-        run: |
-          ./tensorboard/tools/update_version.sh
-          rm -rf /tmp/tb_nightly_pip_package && mkdir /tmp/tb_nightly_pip_package
-          bazel run //tensorboard/pip_package:build_pip_package -- /tmp/tb_nightly_pip_package
-      - name: 'Upload Pip package as an artifact (with TensorFlow only)'
-        # Prevent uploads when running on forks or non-master branch.
-        if: matrix.tf_version_id == 'tf' && github.repository == 'tensorflow/tensorboard' && github.ref == 'refs/heads/master'
-        uses: actions/upload-artifact@83fd05a356d7e2593de66fc9913b3002723633cb # v3.1.1
-        with:
-          name: tb-nightly
-          path: /tmp/tb_nightly_pip_package/*
 
   build-data-server-pip:
     runs-on: ${{ matrix.platform }}
diff --git a/RELEASE.md b/RELEASE.md
index d3de1b5d21..bdbe46f314 100644
--- a/RELEASE.md
+++ b/RELEASE.md
@@ -1,3 +1,29 @@
+# Release 2.12.0
+
+The 2.12 minor series tracks TensorFlow 2.12.
+
+## Features
+
+- Time Series dashboard visualization improvements: (#6137)
+  - Allows selection of a step or range of steps on a scalar chart, and shows a table with data from those steps under it.
+  - Enables linking the selected steps across all charts in the Time Series dashboard.
+- Time Series dashboard now sorts runs in tooltip by pixel distance (matching the Scalars dashboard) (#6116).
+- Fast data loading mode (`--load_fast`, aka “RustBoard”)  improvements:
+  - Supports more ways to authenticate to GCS, including GKE service accounts, via  `gcp_auth` (#5939, thanks @Corwinpro).
+  - Now available on `manylinux2014` platforms (#6101, thanks @adamjstewart).
+
+## Bug Fixes
+
+- Fixes long standing breakage in standalone version of the Projector visualization (#6069).
+- Fixes broken help dialog button in projector plugin (#6024, thanks @mromanelli9).
+- Fixes a bug in which a deadlock could cause the event writer to hang (#6168, thanks @crassirostris).
+
+
+## Breaking Changes
+
+- Drops support for Python 3.7 and marks 3.11 as supported (#6144).
+- Drops support for protobuf < 3.19.6 and adds support for 4.x (#6147).
+
 # Release 2.11.0
 
 The 2.11 minor series tracks TensorFlow 2.11.
diff --git a/tensorboard/compat/proto/config.proto b/tensorboard/compat/proto/config.proto
index 400e02f28c..6b97e1af73 100644
--- a/tensorboard/compat/proto/config.proto
+++ b/tensorboard/compat/proto/config.proto
@@ -18,9 +18,9 @@ option java_package = "org.tensorflow.framework";
 option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/protobuf/for_core_protos_go_proto";
 
 message GPUOptions {
-  // Fraction of the available GPU memory to allocate for each process.
+  // Fraction of the total GPU memory to allocate for each process.
   // 1 means to allocate all of the GPU memory, 0.5 means the process
-  // allocates up to ~50% of the available GPU memory.
+  // allocates up to ~50% of the total GPU memory.
   //
   // GPU memory is pre-allocated unless the allow_growth option is enabled.
   //
@@ -239,6 +239,19 @@ message GPUOptions {
     // hopes that another thread will free up memory in the meantime.  Setting
     // this to true disables the sleep; instead we'll OOM immediately.
     bool disallow_retry_on_allocation_failure = 12;
+
+    // Memory limit for "GPU host allocator", aka pinned memory allocator.  This
+    // can also be set via the envvar TF_GPU_HOST_MEM_LIMIT_IN_MB.
+    float gpu_host_mem_limit_in_mb = 13;
+
+    // If true, then the host allocator allocates its max memory all upfront and
+    // never grows.  This can be useful for latency-sensitive systems, because
+    // growing the GPU host memory pool can be expensive.
+    //
+    // You probably only want to use this in combination with
+    // gpu_host_mem_limit_in_mb, because the default GPU host memory limit is
+    // quite high.
+    bool gpu_host_mem_disallow_growth = 14;
   }
 
   // Everything inside experimental is subject to change and is not subject
@@ -582,7 +595,8 @@ message ConfigProto {
     // If set, this can be used by the runtime and the Ops for debugging,
     // monitoring, etc.
     //
-    // NOTE: This is currently used and propagated only by the direct session.
+    // NOTE: This is currently used and propagated only by the direct session
+    // and EagerContext.
     SessionMetadata session_metadata = 11;
 
     // If true, the session may treat the graph as being static for optimization
@@ -615,18 +629,9 @@ message ConfigProto {
       MLIR_BRIDGE_ROLLOUT_ENABLED = 1;
       // Disabling the MLIR bridge disables it for all graphs in this session.
       MLIR_BRIDGE_ROLLOUT_DISABLED = 2;
-      // Enable the MLIR bridge on a per graph basis based on an analysis of
-      // the features used in the graph. If the features used by the graph are
-      // supported by the MLIR bridge, the MLIR bridge will be used to run the
-      // graph.
-      MLIR_BRIDGE_ROLLOUT_SAFE_MODE_ENABLED = 3;
-      // Enable the MLIR bridge in a fallback mode on a per graph basis based
-      // on an analysis of the features used in the graph.
-      // Running the MLIR bridge in the fallback mode means that it is
-      // executed and it commits all the changes to the TF graph in case
-      // of success. And it does not in case of failures and let the old bridge
-      // to process the TF graph.
-      MLIR_BRIDGE_ROLLOUT_SAFE_MODE_FALLBACK_ENABLED = 4;
+      reserved 3, 4;
+      reserved "MLIR_BRIDGE_ROLLOUT_SAFE_MODE_ENABLED",
+          "MLIR_BRIDGE_ROLLOUT_SAFE_MODE_FALLBACK_ENABLED";
     }
     // Whether to enable the MLIR-based TF->XLA bridge.
     MlirBridgeRollout mlir_bridge_rollout = 17;
@@ -675,7 +680,18 @@ message ConfigProto {
     // Distributed coordination service configurations.
     CoordinationServiceConfig coordination_config = 23;
 
-    // Next: 24
+    // If true, the session will treat the graph as being non-static for
+    // optimization purposes.
+    //
+    // If this option is set to true when a session is created, the full
+    // GraphDef will be retained to enable calls to Session::Extend().
+    // Calling Extend() without setting this flag will result in errors.
+    //
+    // This option is meant to replace `optimize_for_static_graph` and it
+    // aims to negate its value.
+    bool disable_optimize_for_static_graph = 24;
+
+    // Next: 25
   }
 
   Experimental experimental = 16;
diff --git a/tensorboard/compat/proto/full_type.proto b/tensorboard/compat/proto/full_type.proto
index 1f595f045d..56856c6852 100644
--- a/tensorboard/compat/proto/full_type.proto
+++ b/tensorboard/compat/proto/full_type.proto
@@ -186,6 +186,17 @@ enum FullTypeId {
   //   TFT_ENCODING[TFT_INT32, TFT_STRING] is an integer encoded as string.
   TFT_ENCODED = 1004;
 
+  // The type of "shape tensors" where the runtime value is the shape of
+  // some tensor(s), i.e. the output of tf.shape.
+  // Shape tensors have special, host-only placement, in contrast to
+  // TFT_TENSOR[TFT_INT32] which is the type of a normal numeric tensor
+  // with no special placement.
+  //
+  // Examples:
+  //   TFT_SHAPE_TENSOR[TFT_INT32] is the most common
+  //   TFT_SHAPE_TENSOR[TFT_INT64] is also allowed
+  TFT_SHAPE_TENSOR = 1005;
+
   // Type attributes. These always appear in the parametrization of a type,
   // never alone. For example, there is no such thing as a "bool" TensorFlow
   // object (for now).
diff --git a/tensorboard/compat/proto/proto_test.py b/tensorboard/compat/proto/proto_test.py
index 0874e4a977..f76604c6bb 100644
--- a/tensorboard/compat/proto/proto_test.py
+++ b/tensorboard/compat/proto/proto_test.py
@@ -148,8 +148,9 @@
     ("tensorflow/core/framework/", "tensorboard/compat/proto/"),
     ("tensorflow/core/protobuf/", "tensorboard/compat/proto/"),
     ("tensorflow/core/profiler/", "tensorboard/compat/proto/"),
-    ("tensorflow/python/framework/", "tensorboard/compat/proto/"),
     ("tensorflow/core/util/", "tensorboard/compat/proto/"),
+    ("tensorflow/python/framework/", "tensorboard/compat/proto/"),
+    ("tensorflow/tsl/protobuf/", "tensorboard/compat/proto/"),
     ('package: "tensorflow.tfprof"', 'package: "tensorboard"'),
     ('package: "tensorflow"', 'package: "tensorboard"'),
     ('type_name: ".tensorflow.tfprof', 'type_name: ".tensorboard'),
diff --git a/tensorboard/compat/proto/tensor.proto b/tensorboard/compat/proto/tensor.proto
index 64300b4dea..801bbef3b5 100644
--- a/tensorboard/compat/proto/tensor.proto
+++ b/tensorboard/compat/proto/tensor.proto
@@ -83,6 +83,10 @@ message TensorProto {
 
   // DT_UINT64
   repeated uint64 uint64_val = 17 [packed = true];
+
+  // DT_FLOAT8_*, use variable-sized set of bytes
+  // (i.e. the equivalent of repeated uint8, if such a thing existed).
+  bytes float8_val = 18;
 }
 
 // Protocol buffer representing the serialization format of DT_VARIANT tensors.
diff --git a/tensorboard/compat/proto/tfprof_log.proto b/tensorboard/compat/proto/tfprof_log.proto
index 72d51d2da3..68a2b2fe8f 100644
--- a/tensorboard/compat/proto/tfprof_log.proto
+++ b/tensorboard/compat/proto/tfprof_log.proto
@@ -1,11 +1,12 @@
 syntax = "proto3";
 
 package tensorboard;
-option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/profiler/tfprof_log_go_proto";
 
 import "tensorboard/compat/proto/attr_value.proto";
 import "tensorboard/compat/proto/step_stats.proto";
 
+option go_package = "github.com/tensorflow/tensorflow/tensorflow/go/core/profiler/protos_all_go_proto";
+
 // It specifies the Python callstack that creates an op.
 message CodeDef {
   repeated Trace traces = 1;
diff --git a/tensorboard/compat/proto/types.proto b/tensorboard/compat/proto/types.proto
index 4191bd60e3..ddc8751b32 100644
--- a/tensorboard/compat/proto/types.proto
+++ b/tensorboard/compat/proto/types.proto
@@ -30,7 +30,7 @@ enum DataType {
   DT_QINT8 = 11;     // Quantized int8
   DT_QUINT8 = 12;    // Quantized uint8
   DT_QINT32 = 13;    // Quantized int32
-  DT_BFLOAT16 = 14;  // Float32 truncated to 16 bits.  Only for cast ops.
+  DT_BFLOAT16 = 14;  // Float32 truncated to 16 bits.
   DT_QINT16 = 15;    // Quantized int16
   DT_QUINT16 = 16;   // Quantized uint16
   DT_UINT16 = 17;
@@ -40,6 +40,9 @@ enum DataType {
   DT_VARIANT = 21;  // Arbitrary C++ data types
   DT_UINT32 = 22;
   DT_UINT64 = 23;
+  DT_FLOAT8_E5M2 = 24;    // 5 exponent bits, 2 mantissa bits.
+  DT_FLOAT8_E4M3FN = 25;  // 4 exponent bits, 3 mantissa bits, finite-only, with
+                          // 2 NaNs (0bS1111111).
 
   // Do not use!  These are only for parameters.  Every enum above
   // should have a corresponding value below (verified by types_test).
@@ -66,6 +69,8 @@ enum DataType {
   DT_VARIANT_REF = 121;
   DT_UINT32_REF = 122;
   DT_UINT64_REF = 123;
+  DT_FLOAT8_E5M2_REF = 124;
+  DT_FLOAT8_E4M3FN_REF = 125;
 }
 // DISABLED.ThenChange(
 //    https://www.tensorflow.org/code/tensorflow/c/tf_datatype.h,
diff --git a/tensorboard/data/server/descriptor.bin b/tensorboard/data/server/descriptor.bin
index 0ac4e114bf..39dc1c8a09 100644
Binary files a/tensorboard/data/server/descriptor.bin and b/tensorboard/data/server/descriptor.bin differ
diff --git a/tensorboard/data/server/tensorboard.pb.rs b/tensorboard/data/server/tensorboard.pb.rs
index 8bce5d1ccd..ca3e7fc4d2 100644
--- a/tensorboard/data/server/tensorboard.pb.rs
+++ b/tensorboard/data/server/tensorboard.pb.rs
@@ -111,7 +111,7 @@ pub enum DataType {
     DtQuint8 = 12,
     /// Quantized int32
     DtQint32 = 13,
-    /// Float32 truncated to 16 bits.  Only for cast ops.
+    /// Float32 truncated to 16 bits.
     DtBfloat16 = 14,
     /// Quantized int16
     DtQint16 = 15,
@@ -126,6 +126,12 @@ pub enum DataType {
     DtVariant = 21,
     DtUint32 = 22,
     DtUint64 = 23,
+    /// 5 exponent bits, 2 mantissa bits.
+    DtFloat8E5m2 = 24,
+    /// 4 exponent bits, 3 mantissa bits, finite-only, with
+    DtFloat8E4m3fn = 25,
+    // 2 NaNs (0bS1111111).
+
     /// Do not use!  These are only for parameters.  Every enum above
     /// should have a corresponding value below (verified by types_test).
     DtFloatRef = 101,
@@ -151,6 +157,8 @@ pub enum DataType {
     DtVariantRef = 121,
     DtUint32Ref = 122,
     DtUint64Ref = 123,
+    DtFloat8E5m2Ref = 124,
+    DtFloat8E4m3fnRef = 125,
 }
 /// Protocol buffer representing a handle to a tensorflow resource. Handles are
 /// not valid across executions, but can be serialized back and forth from within
@@ -262,6 +270,10 @@ pub struct TensorProto {
     /// DT_UINT64
     #[prost(uint64, repeated, tag="17")]
     pub uint64_val: ::prost::alloc::vec::Vec<u64>,
+    /// DT_FLOAT8_*, use variable-sized set of bytes
+    /// (i.e. the equivalent of repeated uint8, if such a thing existed).
+    #[prost(bytes="bytes", tag="18")]
+    pub float8_val: ::prost::bytes::Bytes,
 }
 /// Protocol buffer representing the serialization format of DT_VARIANT tensors.
 #[derive(Clone, PartialEq, ::prost::Message)]
diff --git a/tensorboard/version.py b/tensorboard/version.py
index d3e7667d80..3f48529683 100644
--- a/tensorboard/version.py
+++ b/tensorboard/version.py
@@ -15,4 +15,4 @@
 
 """Contains the version string."""
 
-VERSION = "2.12.0a0"
+VERSION = "2.12.0"