Operator: v1.NodeSelectorOpExists,
											},
										},
									},
								},
							},
						},
					},
				},
			},
			labels: map[string]string{
				"GPU": "NVIDIA-GRID-K1",
			},
			runPreFilter: true,
		},
		{
			name: "Pod with affinity that don't match node's labels won't schedule onto the node",
			pod: &v1.Pod{
				Spec: v1.PodSpec{
					Affinity: &v1.Affinity{

  Args:
    environ_cp: copy of the os.environ.
    var_name: string for name of environment variable, e.g. "TF_NEED_CUDA".
    query_item: string for feature related to the variable, e.g. "CUDA for
      Nvidia GPUs".
    enabled_by_default: boolean for default behavior.
    question: optional string for how to ask for user input.
    yes_reply: optional string for reply when feature is enabled.

  // it to NCHW before padding, and does all computations in NCHW (this is the
  // default setup for ResNet model trained in fp32 on GPU).
  //
  // To be able to use Tensor Cores on latest NVIDIA GPUs this model has to be
  // converted to NHWC data format.

  // Padding in spatial dimension (NCHW)
  %0 = "tf.Const"() {value = dense<[[0, 0], [0, 0], [3, 3], [3, 3]]> : tensor<4x2xi32>} : () -> tensor<4x2xi32>

Neon link: https://neon.tech/blog/deploy-a-serverless-fastapi-app-with-neon-postgres-and-aws-app-runner-at-any-scale title: Deploy a Serverless FastAPI App with Neon Postgres and AWS App Runner at any scale - author: Kurtis Pykes - NVIDIA link: https://developer.nvidia.com/blog/building-a-machine-learning-microservice-with-fastapi/ title: Building a Machine Learning Microservice with FastAPI - author: Ravgeet Dhillon - Twilio link: https://www.twilio.com/en-us/blog/booking-appointments-twilio-notion-fastapi...

tf_cuda_cc_test(
    name = "pjrt_compile_util_test",
    srcs = ["pjrt_compile_util_test.cc"],
    tags = [
        "config-cuda-only",
        "no_oss",  # This test only runs with GPU.
        "requires-gpu-nvidia",
        "xla",
    ],
    deps = [
        ":pjrt_compile_util",
        ":test_util",
        ":xla_gpu_jit",
        "//tensorflow/cc:function_ops",
        "//tensorflow/cc:math_ops",

			TotalUsageBytes: &totalUsageBytes,
			BaseUsageBytes:  &baseUsageBytes,
			InodeUsage:      &inodeUsage,
		},
		Accelerators: []cadvisorapiv1.AcceleratorStats{
			{
				Make:        "nvidia",
				Model:       "Tesla K80",
				ID:          "foobar",
				MemoryTotal: uint64(seed + offsetMemUsageBytes),
				MemoryUsed:  uint64(seed + offsetMemUsageBytes),

The `tensorflow` pip package has a new, optional installation method for Linux that installs necessary Nvidia CUDA libraries through pip. As long as the Nvidia driver is already installed on the system, you may now run `pip install tensorflow[and-cuda]` to install TensorFlow's Nvidia CUDA library dependencies in the Python environment. Aside from the Nvidia driver, no other pre-existing Nvidia CUDA packages are necessary.

*   Enable JIT-compiled i64-indexed kernels on GPU for large tensors...

build:cuda_clang --config=cuda
# Enable TensorRT optimizations https://developer.nvidia.com/tensorrt
build:cuda_clang --config=tensorrt
build:cuda_clang --action_env=TF_CUDA_CLANG="1"
build:cuda_clang --@local_config_cuda//:cuda_compiler=clang
# Select supported compute capabilities (supported graphics cards).
# This is the same as the official TensorFlow builds.
# See https://developer.nvidia.com/cuda-gpus#compute

	EM_TILEPRO       Machine = 188 /* Tilera TILEPro multicore architecture family */
	EM_MICROBLAZE    Machine = 189 /* Xilinx MicroBlaze 32-bit RISC soft processor core */
	EM_CUDA          Machine = 190 /* NVIDIA CUDA architecture */
	EM_TILEGX        Machine = 191 /* Tilera TILE-Gx multicore architecture family */
	EM_CLOUDSHIELD   Machine = 192 /* CloudShield architecture family */

  if (one_by_one && trivial_strides && trivial_dilations) {
    return "NHWC";
  }

  // If filter spatial dimensions are unknown or not 1x1 we prefer NCHW, because
  // it's the fastest option on NVIDIA GPUs with cuDNN library support.
  return "NCHW";
}

//===----------------------------------------------------------------------===//
// Conv2dBackpropFilterOp